Roughly 80% of global sulfur production comes from mandatory desulfurization of sour crude and natural gas. Environmental regulations force refineries to strip hydrogen sulfide during processing. Out comes elemental sulfur, whether anyone wants it or not. For decades, sulfur was so cheap and so unwanted that Canadian oil sands operators stacked 10 million tonnes of it in solid blocks across Alberta because nobody would pay to ship it.

It is now the most underappreciated commodity chokepoint in the global economy. And the Strait of Hormuz shut it off.

Operation Epic Fury, the coordinated US-Israeli strikes on Iranian military and nuclear facilities that began on February 28, closed the Strait of Hormuz. Over 400 vessels now idle in the Gulf of Oman. Oil spiked above $119 Brent before settling into the current $106-110 range.

That is what every cable news anchor and every macro strategist wants to talk about.

In our oil analysis , we covered the crude repricing. In our tanker and LNG work “Operation Epic Fury” and “8 Days to Kuwait”. “Oil Isn’t The Trade”, we covered the shipping dislocation. In “Party’s Over” and “The Reversion Trade”, we flagged fertilizer as “the second-order effect with the longest tail and the least market attention” and promised a dedicated piece.

This is that piece. The tail is longer than we expected.

Here is the short version for anyone scanning this at 6am: half the world’s seaborne sulfur trade passes through a strait that is currently closed, and there are no strategic sulfur reserves anywhere on earth. The USGS explicitly states “Government Stockpile: None” for the United States. No other country maintains one either. US producer yearend stocks in 2024 totaled 110,000 tonnes against 8.2 million tonnes of production, about five days of output. The industry runs on just-in-time logistics because sulfur was always too cheap to justify building storage infrastructure. Molten sulfur requires steam-jacketed pipelines and heated tanks maintained at 120 to 155°C.

That matters because sulfuric acid is the irreplaceable input for phosphate fertilizer production, Indonesian battery-grade nickel processing, DRC cobalt mining, and the amino acid that sets global chicken prices. The downstream effects of sulfur starvation run through all four simultaneously. Goldman, HSBC, and SFA Oxford have each published on separate legs. Nobody has connected them.

I think the connection is where the alpha is.

The numbers: Gulf states produce about 24% of global sulfur, but the figure that matters is trade. China produces 19,000 kt but imported 9.95 million tonnes in 2024. The United States produces 8,100 kt and consumes 9,100 kt. India, Japan, South Korea: all net importers. The Gulf states are the only major producers that actually export in volume. According to S&P Global Trade Atlas data cited by NDSU, Gulf producers account for 44% of all seaborne sulfur trade. The Fertilizer Institute and Bloomberg put it closer to 50%.

The Ras Laffan strikes also removed approximately 180,000 tonnes of annual sulfur exports, roughly 6% of Qatar's total, compounding the seaborne sulfur deficit through infrastructure damage rather than mere transit disruption. Unlike the Hormuz shipping blockade, this sulfur supply does not return with a ceasefire.

In my oil analysis, I described the “air pocket,” the lag between the last laden tankers departing the Gulf and the arrival of physical scarcity at consuming refineries. The sulfur air pocket operates on the same mechanics but with shorter runways: 2-4 weeks at phosphate complexes, roughly 30-45 days at Indonesian HPAL plants, 4-6 weeks at DRC leaching operations. The last sulfur cargoes are arriving now.

A ceasefire by June still produces the most severe fertilizer supply disruption since 2008. If the closure holds into Q3, the word “disruption” stops being adequate.

Prediction markets provide the duration framework, and readers of the Epic Fury series will recognize the scenario architecture. Our Taco scenario (ceasefire within 3-5 weeks, 20% weight as of “The Reversion Trade”) maps to Polymarket’s 39% chance of normalization by end of April, which has declined from ~80% earlier in March as the crisis proved more intractable than initial consensus assumed. Our Grind scenario (2-4 months, 50% weight) aligns with Kalshi’s 67% cumulative probability by June 1. The Unthinkable (6+ months, 30% weight) is the tail where the word “disruption” stops being adequate. The severity of each downstream channel maps directly to these duration thresholds, and we structure the analysis accordingly.

If you are new to TSCS, this is Part V of our Epic Fury series. The series has tracked the investment implications of the Hormuz closure in real time since March 1, covering crude, tankers and LNG, second-order commodity plays, and the broader geopolitical framework. Paid subscribers receive the full series archive and our scenario updates as prediction market probabilities shift.

The analysis below is for TSCS subscribers only.

Let’s get into it.

Subscribe now

Read more

In October 2025, it quietly surpassed MRI machines as the world’s largest end use. Nobody outside the specialty gas industry noticed.

Five months later, Iranian drones shut down the facility that produces a third of global supply.

You have read the panic version of this story. Sherwood ran it. CNBC ran it. Tom’s Hardware ran it. Helium gone, chips disrupted, AI buildout at risk. We are not writing that piece. Not because the disruption is fake, it is real, but because the panic narrative is built on a number that is almost certainly wrong, and the number that replaces it cuts in a direction nobody is discussing.

Here is what we mean. The headline says semiconductors consume 21 to 24% of global helium. That figure is accurate. It is also gross throughput before recycling, and quoting it as net demand is like measuring an oil refinery’s crude intake and calling it consumption without accounting for the fact that most of it comes out the other end as product. Modern advanced fabs recapture 80 to 95% of the helium they use. The semiconductor industry’s actual fresh helium requirement may be closer to 2 to 5% of the global market.

The entire crisis narrative might be anchored to the wrong number.

But before the bulls exhale: the right number is still scary enough. South Korea sources 64.7% of its helium imports from Qatar and fabricates roughly two-thirds of the world’s memory chips. The memory market entered this crisis with DRAM prices up 240% year over year, finished inventory at three weeks against a four-week floor, and SK Hynix having presold its entire 2026 output. There is no slack. There is no quiet quarter where this lands softly. And the recycling rate that determines whether this is a manageable inconvenience or a binding production constraint is, to be direct, a number nobody can actually measure.

We built the full transmission model: fab-level helium consumption by process step, the nonlinear yield degradation curve that makes conservation self-defeating past a threshold nobody publishes, the distributor allocation mechanics that determine which fabs produce chips and which ones don’t, the China variable that turns a commodity shortage into a geopolitical allocation crisis, and the natural hedge between energy costs and helium scarcity that almost nobody is modelling.

Our base case: a three-to-five month disruption producing meaningful spot price increases and allocation tightening but not binding production constraints at major Korean fabs. Probability: roughly 50%. Tail risk, six months or longer, forcing allocation triage on top of the steepest memory supercycle in the modern era: roughly 25%. Benign resolution within weeks: roughly 25%.

We think the most probable outcome is that the semiconductor industry navigates this with manageable pain and the narrative overshoots reality.

We could be wrong. Below, we show you exactly how to know if we are.

Share

Why you should be reading TSCS

Everything above is free because you can find some version of it elsewhere.

What follows, the recycling arithmetic that collapses the headline narrative, the nonlinear yield curve nobody is modelling, and the five indicators that will tell you whether this is a footnote or a structural repricing, you cannot.

That is what a TSCS subscription gets you.

Subscribe now

Let’s get into it.

Read more

You probably knew that, vaguely, in the way you vaguely know that seatbelts save lives or that smoking causes cancer. You filed it under “technically true but not actionable” and moved on. Most people do.

Here is what you probably do not know.

You do not know that 44% of the world’s uranium enrichment capacity sits inside Russia, and that the West has no way to replace it before the early 2030s. You do not know that the entire Western conversion supply chain, the step between mining uranium and enriching it, runs through three buildings on two continents operating at 91% utilization with zero buffer for disruption. You do not know that a single landlocked Central Asian country produces 39% of the world’s uranium, that more than half its output now flows to Chinese buyers under long-term contracts, and that 90% of its exports historically travel by rail through Russia.

You do not know that the US natural gas price is the single largest determinant of whether Constellation Energy’s nuclear fleet is worth $50 billion or $150 billion. You do not know that interest during construction, not reactor physics, not materials science, not regulatory obstruction, is the mechanism that turned every Western nuclear project of the past two decades into a financial catastrophe. You do not know that the difference between China building a reactor in five years for $2.7 billion and France building one in seventeen years for $23.7 billion is not technology. It is institutional coherence. Same concrete. Same steel. Same physics. Different result.

You do not know that the spot uranium price, the number that shows up on financial websites, is a sideshow derived from a handful of weekly trades, sometimes just one, while 85-90% of the real market moves through private bilateral contracts that will not become public for months. You do not know that when enrichment prices quadruple (they did), the rational response for utilities is to use more uranium per kilogram of fuel, not less, creating a second-order demand tailwind that most supply models ignore entirely. You do not know that Form Energy is building 100-hour iron-air batteries that could narrow the addressable market for new nuclear construction by 2030, a risk the bulls have not priced and the bears have not articulated clearly.

This primer exists because the gap between “nuclear is coming back” and “here is exactly how the fuel cycle works, where the chokepoints are, who controls them, what the bear cases are, and how to actually position a portfolio” is approximately the size of the Grand Canyon. Most of what has been written to fill that gap is either too technical for generalists, too shallow for professionals, or too ideological to be useful for either.

We wrote 45,000 words to fix that.

What follows is a first-principles walkthrough of nuclear power, from the physics of fission to the structure of the uranium market to the geopolitics of enrichment to the investable universe, written for the kind of person who wants to actually understand what they own and why they own it. We assume you are intelligent but not yet an expert. By the time you finish, you will be dangerous.

Share

What is in this Primer?

Part 1 builds the demand case: why global electricity consumption is accelerating, why intermittent renewables cannot fill the gap alone, and why nuclear’s physics gives it a structural advantage no other clean energy source can match.

Parts 2 and 3 take you inside the machine: how fission works, why reactor designs matter, and the five-stage fuel cycle supply chain where every bottleneck has a name, a price, and an owner.

Parts 4 and 5 map the market and the geopolitics: why uranium does not trade like any commodity you have seen before, where the structural deficit is real (and where the bulls overstate it), and why 44% of global enrichment capacity sitting inside Russia is the most underpriced risk in energy.

Parts 6 and 7 confront the hard questions: why China builds reactors in five years and the West takes seventeen, whether NRC reform changes the equation, what the safety record actually says, and why SMRs are five to ten years behind the narrative.

Parts 8 and 9 deploy everything: the investable universe across five distinct markets, the TSCS portfolio framework, the five propositions every nuclear position maps to, and every bear case we could find, including the ones the bulls skip.

A note to existing subscribers, and to those considering it:

This primer gives you the map. Part 2 gives you the coordinates.

Over the coming weeks, we will release the TSCS Nuclear Equity Roadmap, a companion piece built exclusively for paid subscribers. It takes every name discussed in this primer and subjects it to the analysis we did not publish here: company-specific scenario modelling, risk-reward frameworks weighted by the five propositions from Part 8, entry logic, position sizing rationale, and the specific structural trades we believe offer the most asymmetric exposure to the nuclear fuel cycle over the next three to five years.

The primer tells you what enrichment scarcity means. The Roadmap tells you exactly how we would express it in a portfolio, which names carry the thesis, which ones are traps, and where the consensus is mispricing the bottleneck.

If you read 45,000 words and your first thought is “now tell me what to buy,” the Roadmap is where that conversation happens.

Subscribe now

Let’s get into it.

Part 1: Why Nuclear, Why Now

1.1 The Dirtiest Clean Energy Story Ever Told

To understand why nuclear power is about to have its moment, you need to understand why it lost its moment in the first place.

In 1996, nuclear power generated 17.6% of the world’s electricity. The atom was winning. France had gone all-in during the 1970s oil shocks and built 58 reactors in roughly two decades, a feat of industrial coordination that makes China’s Belt and Road look like a school fundraiser. The United States had 104 operating reactors. Japan was building them like Toyota builds Corollas. The future looked fissile.

Then the world collectively lost its nerve.

Three Mile Island in 1979 started the rot in America, not because the accident was catastrophic (it wasn’t; the containment held, nobody died, the maximum offsite dose was equivalent to a chest X-ray), but because it was televised. The China Syndrome had hit theaters twelve days earlier. Hollywood could not have scripted a better PR disaster for the nuclear industry if it had tried. And it did try, repeatedly, for the next four decades.

Chernobyl in 1986 was the real kill shot. An RBMK reactor, a Soviet design with no containment building and a positive void coefficient (meaning it gets more reactive as it loses coolant, which is roughly the engineering equivalent of designing a car whose brakes make it accelerate), blew its lid during a safety test conducted by operators who had disabled most of the safety systems. The explosion and subsequent graphite fire spread radioactive material across Europe. The political fallout was worse than the literal fallout. Italy shut down its entire nuclear program. Germany began its long, slow march toward Atomausstieg. Sweden held a referendum. The global nuclear pipeline didn’t just slow down; it froze.

Then Fukushima in 2011 finished the job. A magnitude 9.0 earthquake followed by a 14-meter tsunami overwhelmed the backup diesel generators at Fukushima Daiichi, and without cooling, three cores melted down. Japan shut down all 54 of its reactors within 14 months. Germany accelerated its nuclear exit. Global nuclear capacity flatlined. The share of world electricity from nuclear dropped from that 17.6% peak to roughly 9%, where it has sat, comatose, for over a decade.

Here is the punchline that almost nobody in the general public understands: across all three of these accidents, spanning three decades and three different reactor designs on three different continents, the total confirmed death toll from acute radiation exposure is fewer than 50 people. All from Chernobyl.

The WHO’s most aggressive long-term projections estimate around 4,000 additional cancer deaths over the lifetimes of the most exposed populations, though the linear no-threshold model underlying that estimate is itself debated among radiation biologists (we will examine this in Part 7).

The one clearly documented radiogenic outcome is a sharp increase in thyroid cancer among children exposed to iodine-131 in the immediate aftermath, most of whom survived with treatment, but which remains the strongest evidence that Chernobyl caused measurable population-level harm beyond the acute deaths.

That is a tragedy. It is also fewer people than coal kills every single day through air pollution. The WHO estimates that outdoor air pollution, of which coal combustion is a leading contributor, kills roughly 4.2 million people per year globally.

Nuclear power killed fewer people in 70 years than coal kills before lunch on a Tuesday.

But radiation is invisible and scary, and coal smoke is familiar and slow, so here we are. The human brain is not wired to assess risk rationally. It is wired to assess risk emotionally. And few things trigger primal fear like the word “nuclear.” The mushroom cloud. The Geiger counter. The hazmat suit. The entire aesthetic of nuclear catastrophe has been baked into our cultural firmware since Hiroshima, and no amount of statistical evidence has been able to overwrite it.

Until now.

Something changed. And what changed was not public sentiment, not some grassroots awakening to the comparative safety data. What changed was the electricity bill.

1.2 The Grid is Hungry and Nobody Has Enough Food

Here is the situation, stated plainly.

Global electricity demand grew by roughly 4.3% in 2024 (subsequently revised to 4.4% in the IEA’s Mid-Year Update), crossing 30,000 TWh for the first time in history. That is over 1,200 TWh of incremental demand in a single year.

To put that in perspective, 1,172 TWh is roughly the entire annual electricity consumption of Germany and France combined. The world added the equivalent of those two countries’ power demand in one year, and it barely made the news.

This was not a one-off. The IEA projects that global electricity demand will continue accelerating through the end of the decade, driven by a convergence of structural forces that are not cyclical and are not going away.

Force number one: electrification. Every internal combustion engine that becomes an EV is a new mouth at the electricity table. Every gas furnace replaced by a heat pump. Every industrial process that moves from fossil fuels to electric power. The world has decided, for better or worse, that the path to decarbonization runs through the electrical grid. This is probably correct. But nobody seems to have done the math on what happens when you try to electrify everything while simultaneously retiring the power plants that currently keep the lights on.

Force number two: air conditioning. This one is underappreciated. There are roughly 2.8 billion people living in tropical and subtropical regions who do not yet have air conditioning. As incomes rise in India, Southeast Asia, and Sub-Saharan Africa, the demand for cooling will be staggering. The IEA estimates that global cooling demand could triple by 2050. Air conditioning is not optional in a 45-degree heatwave. It is a survival technology. And it is extraordinarily power-hungry.

Force number three, and this is the one that broke the dam: artificial intelligence.

Data center electricity consumption is projected to more than double to approximately 945 TWh by 2030. That is roughly equivalent to the entire electricity consumption of Japan. AI-optimized servers alone are growing at roughly 30% per year in power consumption. In the United States, data centers are on track to account for nearly half of all electricity demand growth between now and 2030. Read that again. Half. The US economy will consume more electricity for processing data by the end of the decade than for manufacturing aluminum, steel, cement, and chemicals combined.

This is not a forecast from some breathless AI startup trying to raise a Series B. This is the IEA’s base case projection, published in April 2025.

And here is the problem: you cannot power a data center with hopes and vibes. You cannot power it intermittently. When Microsoft or Google or Amazon builds a 500MW hyperscale campus, it needs 500MW of power available 24 hours a day, 7 days a week, 365 days a year. Not “500MW when the wind blows” or “500MW between 10am and 4pm on a sunny day.” Five hundred megawatts, all the time, no exceptions, with contractual penalties if the power goes out.

This brings us to the core physics problem that the entire energy transition has been trying very hard to ignore.

1.3 The Capacity Factor Problem, or Why Solar Panels Don’t Work at Night

There is a concept in the power industry called “capacity factor.” It is the single most important number for understanding why the energy mix looks the way it does, and it is the number that renewable energy advocates most desperately wish did not exist.

Capacity factor is the ratio of actual electricity generated over a given period to the maximum possible generation if the plant ran at full output 100% of the time. It tells you how often a power source actually shows up for work.

Here are the 2024 numbers for the United States:

Nuclear: ~92%, Natural gas (combined cycle): ~57-61% (varying by year; 2024 EIA data showed ~60.5%), Coal: ~43%, Wind: ~34%, Solar: ~23%

Read those numbers carefully. A nuclear plant runs at full tilt more than 90% of the time. It only stops for scheduled refueling every 18 to 24 months, and even those outages are getting shorter. The US fleet hit 92% capacity factor in 2024 and has not dropped below 90% in any year since 2012. That is not a theoretical number. That is operational performance across 94 reactors, many of which are 40+ years old.

Solar, meanwhile, manages 23%. Not because the panels are broken. Not because the maintenance crew is lazy. Because the sun goes down. Every single night. And it is dim in the morning. And it is dim in the evening. And clouds exist. And winter exists. That 23% is a hard physical constraint dictated by the rotation of the Earth and the tilt of its axis. No amount of engineering innovation changes the fact that Ohio in January receives roughly one-third the solar irradiance of Arizona in July.

Wind does better at 34%, but with a different problem: unpredictability. The sun at least has the decency to follow a schedule. Wind does whatever it wants. A wind drought, which is exactly what it sounds like, can last days or even weeks. In January 2024, large parts of Europe experienced a multi-day period of low wind coinciding with a cold snap, which is of course exactly when you need the most electricity. This is called “Dunkelflaute” in German, literally “dark doldrums,” and it is not a theoretical risk. It is a recurring meteorological reality.

Now, the standard rebuttal here is “batteries.” And yes, batteries are improving rapidly and costs are falling. But let us do some basic arithmetic. A Tesla Megapack, one of the most commercially deployed utility-scale batteries, stores about 3.9 MWh of energy. A single 1GW nuclear reactor generates roughly 24,000 MWh per day. To store one day of that reactor’s output in Megapacks, you would need approximately 6,150 of them, covering something like 120 acres of land, at a cost of roughly $7-9 billion at current installed pricing (down from $10-12 billion just two years ago, as battery costs continue to fall, though this only reinforces how enormous the storage challenge remains). And that gives you one day of backup. Dunkelflaute can last a week.

The math simply does not work for grid-scale seasonal storage. Not today, not with any battery chemistry currently in production, not with any chemistry on the near-term horizon. Batteries are excellent for short-duration storage (4-8 hours, maybe 12). They are a rounding error on the scale of multi-day or seasonal storage.

So if you need a power source that runs all the time, produces no carbon emissions, fits on a relatively small footprint of land, and can be located close to demand centers, your list of options is extremely short.

You have hydropower, which is excellent but geographically constrained. You cannot build a dam where there is no river, and most of the best sites are already taken.

You have geothermal, which is promising but even more geographically constrained, and is decades away from being available at scale outside of Iceland, parts of East Africa, and the western US.

And you have nuclear.

That is it. That is the list.

1.3.1 How Nuclear Fits Into Grids

A necessary footnote on how nuclear fits into grids with growing renewable penetration: nuclear’s traditional role as “baseload” (running flat-out, 24/7, never adjusting output) is increasingly awkward in electricity systems where solar and wind provide cheap but variable generation.

In markets with high renewable penetration (California, parts of Germany, South Australia), midday electricity prices can go negative when solar output peaks. A nuclear plant running at full power into a negative-price market is losing money on every MWh it generates. France has already begun operating some of its reactors in “load-following” mode, ramping output up and down to complement variable renewables, but this reduces capacity factor (and therefore revenue) and introduces thermal cycling that accelerates wear on reactor components.

The IAEA and the US National Renewable Energy Laboratory (NREL) are studying “nuclear-renewable hybrid systems” that pair reactors with industrial heat applications, hydrogen production, or desalination during periods of low electricity prices, allowing the reactor to run at full thermal power but redirect the heat to non-electric uses. These are promising concepts but largely unproven at commercial scale.

For investors, the grid integration question matters because the 92% capacity factor that makes nuclear’s economics so compelling was achieved in a world where nuclear ran as baseload against coal and gas. In a future grid where solar provides the cheapest daytime MWh and wind provides cheap-but-variable nighttime MWh, nuclear’s capacity factor could decline not because the reactors stop working but because the economics of running them at full power in every hour deteriorate. This is a slow-moving risk, most relevant on a 15-30 year horizon, but it should be part of any long-duration nuclear investment thesis.

The counter-argument: data centers and industrial electrification are creating a new category of 24/7 demand that is perfectly matched to nuclear’s flat output profile. A data center does not care about the duck curve. It wants the same 500 MW at noon as it does at midnight. This demand profile effectively resurrects the baseload model that high-renewable grids were eroding, and it is the primary reason tech companies are signing nuclear PPAs rather than renewable PPAs. The grid integration risk is real for merchant nuclear plants selling into wholesale markets. It is far less relevant for contracted nuclear plants selling behind-the-meter or under long-term PPAs to industrial offtakers. Structure your portfolio accordingly.

1.4 The Energy Density Cheat Code

Before we go further, it is worth pausing on the physics, because the physics is the reason nuclear exists at all, and it is the reason nuclear will always have a structural advantage over every chemical fuel source on Earth.

Chemical energy, the kind you get from burning coal or gas or oil, comes from rearranging electrons. When carbon combusts with oxygen, electrons shift between atomic orbitals, and the energy difference is released as heat. This is the same basic process whether you are burning wood, gasoline, or natural gas. The energies involved are on the order of a few electron volts per reaction.

Nuclear energy comes from rearranging the nucleus itself. When a uranium-235 atom absorbs a neutron and fissions into two lighter atoms, the energy released is roughly 200 million electron volts per reaction. Not a few. Two hundred million.

This is not a small difference. It is a difference of roughly eight orders of magnitude. It means that a single uranium fuel pellet, a little ceramic cylinder about the size of the tip of your pinky finger, weighing around 10 grams, contains as much energy as one ton of coal, or 149 gallons of oil, or 17,000 cubic feet of natural gas.

Let me state that differently, because people tend to glaze over large numbers. Ten grams of uranium. That is lighter than two nickels stacked together. It replaces a ton of coal. A literal ton. Two thousand pounds. Enough coal to fill a large bathtub. You could hold the uranium pellet between your thumb and forefinger. You would need a pickup truck for the coal.

This energy density gap is not something that can be engineered away. It is baked into the fundamental physics of the strong nuclear force versus electromagnetic force. No improvement in solar panel efficiency, no breakthrough in battery chemistry, no advance in wind turbine design will ever close this gap, because they are all operating in the electromagnetic regime. Nuclear operates in a different regime entirely, and the energy densities are separated by a factor of millions.

This is why a single nuclear power plant can sit on a few hundred acres and produce as much electricity as a solar farm covering 50+ square miles, or a wind farm stretching across hundreds of square miles. This is why France, a country the size of Texas, generates about 65-70% of its electricity from nuclear with 56 operating reactors occupying a negligible fraction of its land area, while Germany, which chose renewables, has covered its landscape in turbines and panels and still burns more coal than any other country in Western Europe.

Land is not free. Land has opportunity costs. Land near cities, where the power is actually needed, is especially expensive. The energy density of nuclear is not just a fun physics fact. It is a hard economic advantage that compounds every year as land and permitting costs rise.

1.5 The Vibe Shift

So we have a structural demand problem that is large, accelerating, and concentrated in the exact kind of loads (data centers, industrial electrification) that require 24/7 baseload power. We have a supply problem because the two dominant clean energy sources are intermittent and cannot economically provide storage at the scale needed. And we have a fuel source with energy density advantages that are literally millions of times better than the alternatives.

You might think the conclusion here is obvious. You would be right. And yet it took the world 15 years of dithering after Fukushima to arrive at it.

The turning point came not from environmentalists or engineers or policy wonks. It came from the tech industry. When you have Microsoft signing a 20-year power purchase agreement to restart Three Mile Island Unit 1 (yes, that Three Mile Island), when Amazon is buying a nuclear-powered data center campus from Talen Energy, when Google is signing deals with Kairos Power for small modular reactors, you know the vibe has shifted. These are not speculative bets by junior analysts. These are multi-billion dollar commitments by companies whose core business depends on having reliable, scalable, carbon-free power.

At the policy level, the shift has been equally dramatic. Twenty-two countries signed a pledge at COP28 in December 2023 to triple nuclear energy capacity by 2050, a number that has since grown to over 30 (by COP29 in November 2024) and 38 by the March 2025 Nuclear Energy Summit.

The IAEA has raised its nuclear capacity projections for five consecutive years. In the US, President Trump signed executive orders in May 2025 to accelerate nuclear adoption, targeting an expansion from approximately 100 GW today to 400 GW by 2050. In the high-case IAEA projection, global nuclear capacity could reach 992 GW by 2050, roughly 2.6 times the 2024 level.

Goldman Sachs projects that global nuclear generating capacity will rise from 378 GW to 575 GW by 2040, increasing nuclear’s share of the global electricity mix from roughly 9% to 12%. The total fleet of around 440 operable reactors today (approximately 416 actively generating) is set to expand to roughly 500 by 2030, with over 400 additional reactors in the planning or proposal stage.

The nuclear fleet set a generation record in 2024, producing 2,667 TWh of electricity, surpassing the previous record of 2,660 TWh set in 2006. But intellectual honesty demands a caveat: outside China, global nuclear generation in 2024 remained roughly 360 TWh below its 2006 level, a nearly 14% decline. The record is real, but it is substantially a Chinese story. The comeback is underway in the data, but it is geographically uneven, and the narrative is still catching up.

And this is where the story gets interesting for investors. Because while the demand case for nuclear is becoming consensus, the supply chain required to actually deliver on this demand is riddled with bottlenecks, geopolitical dependencies, and structural underinvestment. The uranium market. The enrichment bottleneck. The conversion duopoly. The construction workforce that evaporated after Fukushima. The regulatory apparatus that takes a decade to approve what China builds in five years.

This primer exists because the gap between “nuclear is coming back” (which is the easy part) and “here is exactly how the nuclear fuel cycle works, where the bottlenecks are, what the investable universe looks like, and what the bear cases are” (which is the hard part) is approximately the size of the Grand Canyon. And most of what has been written to fill that gap is either too technical for generalists, too shallow for professionals, or too ideological to be useful for either.

We are going to fix that.

What follows is a first-principles walkthrough of nuclear power, from the physics of fission to the structure of the uranium market to the investable equities, written for the kind of person who wants to actually understand what they own and why they own it. We assume you are smart but not yet an expert. By the time you finish, you will be dangerous.

Part 2: How Nuclear Power Actually Works

2.1 Splitting the Unsplittable

The ancient Greeks called the atom “atomos,” meaning “uncuttable.” They were wrong. Spectacularly, violently, world-changingly wrong.

In December 1938, two German chemists, Otto Hahn and Fritz Strassmann, fired neutrons at uranium and found barium in the debris. Barium has no business being near uranium. It is a completely different element, roughly half the atomic weight. Their colleague Lise Meitner, who had fled Nazi Germany months earlier because she was Jewish, figured out what had happened: the neutron had split the uranium nucleus in two. She and her nephew Otto Frisch calculated the energy released using Einstein’s E=mc², and the number they got was enormous. Orders of magnitude larger than any chemical reaction. They called it “fission,” borrowing the term from biology, where it describes a cell dividing in two.

Within seven years, fission had ended a world war. Within fifteen, it was generating electricity. And within seventy, it was powering roughly 10% of the world’s grid while most people still could not tell you what it actually is.

So let us fix that.

2.2 The Bowling Ball and the Billiard Table

Here is the simplest possible explanation of nuclear fission, and I promise it is not dumbed down. It is just clear.

Imagine a bowling ball sitting on a billiard table. The bowling ball represents a uranium-235 nucleus. It is big. It is heavy. It has 92 protons and 143 neutrons crammed together in a space so small that if you scaled the atom up to the size of a football stadium, the nucleus would be a marble at the 50-yard line. All of the atom’s mass is in that marble. Everything else is empty space and electron clouds.

Now imagine you roll a cue ball at the bowling ball. The cue ball is a neutron. When it hits the bowling ball (gets absorbed by the U-235 nucleus), something extraordinary happens. The bowling ball does not just crack. It vibrates. The nucleus, which is held together by the strong nuclear force (the most powerful force in the universe at short range), becomes momentarily unstable. It elongates, like a water droplet about to split. And then it does split, violently, into two smaller nuclei (called “fission fragments”), plus two or three free neutrons, plus a burst of energy.

That burst of energy is the entire point. It is roughly 200 million electron volts per fission event, or about 200 MeV if you want to sound like you know what you are talking about at dinner. For comparison, the combustion of a single carbon atom with oxygen releases about 4 eV. Nuclear fission releases about 50 million times more energy per event than chemical combustion.

But here is the thing that makes fission uniquely powerful, and uniquely dangerous: those two or three free neutrons that fly out of the split nucleus? They can go hit other U-235 nuclei. And those nuclei split, releasing more neutrons. Which hit more nuclei. Which split, releasing more neutrons.

This is a chain reaction.

In a bomb, the chain reaction is uncontrolled. Each generation of fissions happens in nanoseconds, and the number of fissions doubles (or triples) with each generation. Within microseconds, you have a trillion trillion fissions happening simultaneously, and the resulting energy release is measured in kilotons of TNT equivalent. This is what happened over Hiroshima.

In a reactor, the chain reaction is controlled. The goal is to maintain what physicists call “criticality,” which means exactly one neutron from each fission event goes on to cause another fission event. Not two (which would cause the reaction to accelerate exponentially). Not zero (which would cause the reaction to die). Exactly one. This produces a steady, continuous release of heat, which is what we use to generate electricity.

The mechanism for control is beautifully simple: control rods. These are rods made of neutron-absorbing materials (boron, hafnium, cadmium, or silver-indium-cadmium alloys) that are inserted into the reactor core. Push them in further, they absorb more neutrons, the reaction slows down. Pull them out, more neutrons are available to cause fissions, the reaction speeds up. Want to shut the reactor down entirely? Drop all the rods in. Gravity does the work. In a modern pressurized water reactor, the control rods are held up by electromagnets. Lose electrical power, the magnets de-energize, and the rods fall into the core by their own weight. The reactor shuts itself down. This is called a “scram,” and it happens in seconds.

There is an elegant philosophical point buried in this engineering. A nuclear reactor is not a machine you “turn on.” It is a machine you prevent from shutting itself down. Left to its own devices, with the control rods dropped, it shuts down. Even without the rods, modern light water reactors are designed with negative temperature and void coefficients, meaning the physics actively resists power excursions: if the fuel heats up, the fission rate drops; if the coolant boils away, the chain reaction loses its moderator and dies.

You have to actively, deliberately maintain the precise conditions under which the chain reaction sustains itself. The default state of a well-designed reactor is “off.” The default state of a coal fire is “on until you run out of coal.” This distinction matters more than most people realize, and it is the reason why the catastrophic positive-feedback failure at Chernobyl was a design-specific pathology, not a feature of nuclear reactors in general.

2.3 Why U-235, and the 0.7% Problem

If fission is so powerful, why aren’t we fissioning everything? Why uranium specifically?

The answer is that not all atomic nuclei are willing to be split. Most are not. The strong nuclear force holds the nucleus together with extraordinary tenacity, and for most elements, a free neutron bouncing into the nucleus just gets absorbed without causing fission. The nucleus gets a little heavier, maybe emits some gamma radiation, and carries on with its life.

But a handful of very heavy, very specific isotopes have nuclei that are right on the edge of stability. They are so large, so packed with protons and neutrons, that the electrostatic repulsion between all those protons (positive charges repel each other) nearly overwhelms the strong force holding the nucleus together. These nuclei are primed. One extra neutron is all it takes to push them over the edge.

The isotope that matters most for commercial nuclear power is uranium-235. Why this one specifically? Two reasons.

First, U-235 is “fissile,” meaning it can be split by a slow-moving (thermal) neutron. This is crucial. Fast neutrons are hard to work with. They zip through material and are difficult to capture. Slow neutrons, the ones that have been decelerated by bouncing around in water or graphite (a process called “moderation,” which we will get to shortly), are much more likely to be absorbed by a U-235 nucleus and cause fission. This makes U-235 the practical workhorse of nuclear energy.

Second, U-235 exists in nature. You can dig it out of the ground. This sounds obvious, but it is not trivial. Other fissile materials, like plutonium-239 or uranium-233, must be manufactured in a reactor. They do not occur naturally in usable quantities. U-235 is the only fissile material you can mine.

Now here is the catch, and it is a big one.

Natural uranium is 99.3% uranium-238 and only 0.7% uranium-235.

Let that sink in. Of all the uranium you dig out of the ground, process, mill, and refine, only 7 atoms out of every 1,000 are the ones you actually want. The other 993 are U-238, which is not fissile. U-238 is “fertile,” meaning it can absorb a fast neutron and eventually transmute into plutonium-239 (which is fissile), but it will not sustain a chain reaction on its own.

This 0.7% problem is the reason the entire enrichment industry exists. It is the reason centrifuges are classified technology. It is the reason Iran’s nuclear program makes geopolitical headlines. It is the reason the nuclear fuel cycle is orders of magnitude more complex than “dig up coal, burn coal.” Everything downstream in the nuclear supply chain, conversion, enrichment, fabrication, exists because nature gave us uranium that is 99.3% wrong and 0.7% right, and we have to fix that ratio before we can use it.

Most commercial reactors need uranium enriched to between 3% and 5% U-235. This sounds modest until you consider the physics involved in separating two isotopes that are chemically identical and differ in mass by only three neutrons. You cannot do it with chemistry. You have to exploit that tiny mass difference mechanically, and the technology for doing so (gas centrifuges, spinning uranium hexafluoride gas at tens of thousands of RPM to separate the slightly lighter U-235 from the slightly heavier U-238) is one of the most technically demanding and strategically sensitive industrial processes on Earth.

We will cover enrichment in exhaustive detail in Part 3. For now, just understand this: the 0.7% figure is the first constraint in the entire nuclear value chain. Everything else flows from it.

2.4 From Fission to Electricity: The World’s Most Complicated Way to Boil Water

Here is the single most disappointing fact in all of energy technology:

A nuclear power plant generates electricity by boiling water.

That is it. That is the whole trick.

You use nuclear fission to generate heat. The heat boils water. The steam turns a turbine. The turbine drives a generator. The generator produces electricity. This is fundamentally identical to how a coal plant works, or a natural gas plant, or a concentrated solar thermal plant, or a geothermal plant. They all boil water. The only thing that changes is the heat source.

The reason this is disappointing is that fission releases energy at scales that could, in theory, do far more exotic things. Two hundred million electron volts per fission event, and we use it to make steam.

It is like using a Ferrari to deliver pizzas.

Technically, it works. Aesthetically, it is an affront to the engineering gods. A physicist’s footnote: this is a deliberate simplification. Some Gen IV designs abandon the steam cycle entirely. High-temperature gas-cooled reactors can drive a Brayton gas turbine cycle at higher efficiencies.

Molten salt designs can supply industrial process heat directly. And fission’s real untapped potential lies not in electricity generation but in high-temperature applications (hydrogen production, desalination, synthetic fuel synthesis) where the 600-900°C heat from advanced reactors becomes an industrial feedstock rather than a source of steam. The existing fleet boils water. The next generation may not.

But thermodynamics is thermodynamics. The Rankine cycle (heat water, make steam, spin turbine, condense steam, repeat) is the most proven, most efficient, most scalable method we have for converting heat to electricity.

Nuclear plants typically achieve thermal efficiencies of 32-34% for the existing PWR and BWR fleet, with newer Gen III+ designs like the EPR pushing toward 36-37%. Roughly a third of the heat energy from fission is converted to electricity.

The rest is waste heat, which is why nuclear plants need cooling, either from rivers, oceans, cooling towers, or (in some designs) air.

This is lower than a modern combined-cycle natural gas plant (which can hit 60%+ efficiency by using both a gas turbine and a steam turbine in sequence), but higher than most coal plants. The reason nuclear plants are less thermally efficient than gas plants has nothing to do with the fission itself and everything to do with the water temperature. Nuclear reactors operate at lower steam temperatures (roughly 290-330°C) than gas turbines (which operate at over 1,000°C), because the materials in the reactor have to withstand decades of neutron bombardment and cannot be pushed to extreme temperatures without degradation. Higher temperature = higher efficiency per the Carnot theorem, and nuclear plants just run cooler.

Some Gen IV designs (molten salt reactors, high-temperature gas-cooled reactors) aim to change this by operating at much higher temperatures, potentially pushing nuclear thermal efficiency into the 45-50% range. But those are still in development. The existing fleet, the roughly 440 reactors operable worldwide today (of which approximately 416 are actively generating, with the remainder in long-term outage or suspended operation, principally in Japan), essentially all boil water.

Now, how they boil water, that is where the engineering gets interesting.

2.5 The Reactor Zoo: PWR, BWR, and Why the Plumbing Matters

There are many types of nuclear reactor in the world, but the vast majority, roughly 85% of the global fleet, fall into two categories, both of which use ordinary water (called “light water” to distinguish it from heavy water, which we will get to) as both the coolant and the “moderator” (the material that slows down the neutrons so they can cause fission).

The Pressurized Water Reactor (PWR): The World’s Reactor

The PWR is the dominant reactor design on Earth. About 70% of all operating commercial reactors are PWRs. It was originally developed by the United States Navy in the 1950s for nuclear submarines (the USS Nautilus, the world’s first nuclear-powered submarine, used a PWR), and the design was then scaled up for commercial electricity generation by Westinghouse.

The defining feature of a PWR is that it uses two separate water loops, and the water in the reactor core never boils.

Loop one (the primary circuit): Water is pumped through the reactor core at very high pressure, around 155 bar (roughly 2,250 psi, for reference, your car tires are inflated to about 32 psi). This enormous pressure keeps the water liquid even though it is heated to over 320°C, well above water’s normal boiling point of 100°C. The superheated, pressurized water carries the heat from the fission reaction out of the core and into a steam generator.

Loop two (the secondary circuit): Inside the steam generator (essentially a giant heat exchanger), the hot water from the primary circuit transfers its heat to water in a separate, lower-pressure secondary circuit. This secondary water does boil, producing steam. The steam goes to the turbine, spins it, generates electricity, then gets condensed back into water and recycled.

The genius of this design is the separation. The water that touches the reactor core (and picks up some radioactivity from passing through a neutron field) never touches the turbine. The turbine sees only clean, non-radioactive steam from the secondary circuit. This means the turbine hall does not need heavy radiation shielding, maintenance workers can access it relatively freely, and if there is a leak in the turbine system, it does not release radioactive water.

The downside is complexity. Two full water circuits means more pumps, more piping, more things that can break, and the massive steam generators (which can be 20 meters tall and weigh 800 tons each, with thousands of tiny tubes inside) are among the most expensive and maintenance-intensive components in the plant.

France’s entire fleet of 56 operating reactors is PWR-based (from a historical build of 58, with Fessenheim 1 and 2 permanently shut in 2020). So are most of the US fleet. The AP1000, the EPR, China’s Hualong One, South Korea’s APR-1400, Russia’s VVER, these are all PWR variants. If you own a nuclear utility stock, you almost certainly own PWRs.

The Boiling Water Reactor (BWR): The Simpler Cousin

BWRs make up roughly 13-14% of the global fleet (down from 15% a few years ago, primarily due to permanent shutdowns of older Japanese BWR units post-Fukushima).

They were developed by General Electric (now GE Vernova Hitachi Nuclear Energy) at Argonne National Laboratory in the 1950s, and their design philosophy is: “Why bother with two loops when one will do?”

In a BWR, the water in the reactor core boils directly. The steam goes straight to the turbine. One loop. No steam generator. Simpler plumbing. Lower operating pressure (about 70 bar versus 155 bar for a PWR). In principle, cheaper and simpler to build.

The trade-off: the steam that drives the turbine has passed directly through the reactor core, picking up trace radioactivity (primarily nitrogen-16, half-life of about 7 seconds). The turbine hall needs radiation shielding, and maintenance workers need protocols for working around mildly activated components. In practice, this is manageable (radiation levels drop quickly once the reactor shuts down), but it adds operational complexity the PWR avoids.

BWRs are found primarily in the United States, Japan, Sweden, and Taiwan. Fukushima Daiichi was a BWR facility.

A Quick Note on the Moderator, Because This Matters

I mentioned “moderation” earlier, and it is one of those concepts that sounds simple but contains a key insight that most nuclear primers gloss over.

When a U-235 atom fissions, the neutrons that fly out are moving extremely fast, roughly 20,000 km/s, or about 7% of the speed of light. These are “fast neutrons.” And here is the counterintuitive part: fast neutrons are actually bad at causing fission in U-235. They zip right through other uranium nuclei without interacting. It is like throwing a bullet through a crowd; it goes too fast to shake hands.

To maximize the probability of a neutron being absorbed by another U-235 nucleus and causing fission, you need to slow it down. Way down. To “thermal” energies, where the neutron is moving at roughly the same speed as the molecules of the surrounding material (about 2 km/s, roughly the speed of a rifle bullet, versus the 20,000 km/s it started at).

You slow neutrons down by bouncing them off light nuclei. Think of it like pool: if a cue ball (the neutron) hits a bowling ball (a heavy nucleus), it bounces off with almost no loss of energy. But if it hits another cue ball (a nucleus of similar mass), it transfers a lot of energy in the collision. Hydrogen, the lightest element, is the best moderator. One good collision with a hydrogen nucleus transfers about half the neutron’s energy. A few dozen collisions, and the neutron is at thermal energies.

And what is full of hydrogen? Water. H₂O. Two hydrogen atoms per molecule. This is why light water reactors use water as both a coolant (to carry heat away from the core) and a moderator (to slow neutrons down so they can sustain the chain reaction). It is a beautifully efficient dual-purpose design.

It also contains a built-in safety feature that most people do not appreciate. If the water in a light water reactor boils away or leaks out, you lose your moderator. Without a moderator, the fast neutrons cannot sustain the chain reaction. The reactor shuts itself down. This is called a “negative void coefficient,” and it means that the most common failure mode (loss of coolant) inherently stops the nuclear reaction.

This is exactly the opposite of what happened at Chernobyl, and we need to talk about why.

2.6 The Outliers: CANDU and RBMK

Not every reactor uses light water. Two important designs use different moderators, and understanding them matters both for completeness and for understanding why Chernobyl happened.

The CANDU (Canada Deuterium Uranium)

Canada went a different route. Rather than enriching uranium (which Canada did not have the industrial capacity for in the 1950s), Canadian engineers designed a reactor that could use natural uranium, the stuff that comes straight out of the ground at 0.7% U-235, with no enrichment required.

The trick is the moderator. Light water absorbs too many neutrons, which means if you use natural uranium fuel (with its low U-235 content), there are not enough neutrons left over to sustain the chain reaction. Heavy water (deuterium oxide, D₂O, where the hydrogen atoms are replaced by deuterium, hydrogen with an extra neutron) absorbs far fewer neutrons, leaving enough to sustain fission even in natural uranium.

The CANDU reactor uses heavy water as both moderator and coolant. The fuel bundles are arranged horizontally in pressure tubes running through a large tank (the “calandria”) filled with heavy water. This design allows the reactor to be refueled without shutting down (you push new fuel bundles in one end and the spent ones come out the other), which is a significant operational advantage.

The downside: heavy water is expensive to produce (roughly $300-600 per kilogram), and you need hundreds of tons of it. The CANDU reactor makes up about 11% of the global fleet, operating primarily in Canada, India, Argentina, Romania, Pakistan, and China.

For investors, the CANDU matters because it represents a different fuel demand profile. CANDU reactors do not need enriched uranium, which means they do not contribute to enrichment demand (the bottleneck we will discuss later). But they do consume natural uranium in larger quantities per unit of energy produced.

The RBMK: Why Chernobyl Was a Design Problem, Not a Nuclear Problem

The RBMK (Reaktor Bolshoy Moshchnosti Kanalny, “High-Power Channel-type Reactor”) was a Soviet design, and understanding it matters because it explains why Chernobyl happened, and why Chernobyl could not happen in a Western reactor.

The RBMK used graphite as a moderator and light water as a coolant. This is the critical distinction. In a light water reactor (PWR or BWR), the water serves as both moderator and coolant. Lose the water, you lose both cooling and moderation, and the chain reaction stops. In an RBMK, the graphite moderator is separate from the water coolant. Lose the water, and you lose the coolant, but the graphite moderator is still there, still slowing down neutrons, still sustaining the chain reaction.

Worse, the RBMK had a large positive void coefficient of reactivity (approximately +4.7 beta, where beta is the delayed neutron fraction, the threshold above which a reactor becomes “prompt critical” and power rises on a timescale of microseconds rather than seconds). This positive void coefficient was always present, not just at certain power levels. At normal operating power, a separate effect, the negative Doppler fuel temperature coefficient (which reduces reactivity as fuel heats up), was strong enough to dominate, making the overall power coefficient negative. The reactor was manageable, if not safe, in its normal operating envelope. But below roughly 20% of rated power, where the Doppler effect weakened, the positive void coefficient overwhelmed all negative feedbacks.

The reactor became inherently unstable: any increase in void formation (steam bubbles in the coolant channels) increased reactivity, which increased power, which increased boiling, which created more voids. When the water coolant boiled (forming steam voids), it increased the reactor’s reactivity further.

The reactor got hotter, which caused more boiling, which caused more voids, which increased reactivity further. This is a positive feedback loop. In engineering, positive feedback loops are how you get explosions.

On April 26, 1986, the operators at Chernobyl Unit 4 were conducting a safety test at low power (which is ironically when the positive void coefficient was most dangerous), and they had disabled several safety systems to conduct the test. When the reactor experienced a sudden, massive power surge, the positive void coefficient amplified it exponentially. the reactor went from near-shutdown to an estimated 300,000 MW or more of thermal power (roughly 100 times its rated maximum) in less than a second. The last instrument reading before the sensors were destroyed registered approximately 33,000 MW, but computational models of the excursion indicate the actual peak was far higher.

The resulting steam explosion blew the 1,000-ton reactor lid off like a bottle cap, and the exposed graphite moderator caught fire, burning for 10 days and sending radioactive material into the atmosphere.

The point of this story is not to scare you. The point is that the RBMK design had known, fundamental flaws that do not exist in any Western reactor design, and the remaining RBMK reactors (which operate solely in Russia and now number approximately 10 units, making up roughly 2.3% of the global fleet following recent retirements including Kursk 2 in January 2024) have since been modified to eliminate the positive void coefficient at low power.

No PWR, no BWR, no CANDU, no Gen III+ reactor has the physics that enabled Chernobyl. It is not a question of better operators or better procedures. It is a question of fundamental reactor physics. The accident mechanism does not exist in these designs. Comparing Chernobyl to modern nuclear is like comparing the Hindenburg to modern aviation. Same word (”air travel”), completely different engineering reality.

2.7 Gen III+: The New Builds

The reactors operating today are mostly Gen II designs, built between the 1970s and 1990s. They work. They work extremely well, in fact (92%+ capacity factors, 40+ year lifetimes, declining maintenance costs). But when you hear about “new nuclear,” you are hearing about Gen III+ designs, which incorporate passive safety features and evolutionary improvements over the existing fleet.

There are two Gen III+ designs that dominate the current new-build conversation, and both have troubled histories.

The AP1000 (Westinghouse/Toshiba)

The AP1000 (”Advanced Passive 1000”) is a 1,100 MWe pressurized water reactor designed by Westinghouse. Its core innovation is passive safety: in an emergency, the reactor can cool itself for 72 hours without any operator action, without any AC power, and without any external water supply, using gravity-fed water tanks and natural convection. This is a direct response to Fukushima, where the failure was not the reactor physics but the loss of active cooling systems (the diesel backup generators that were flooded by the tsunami).

The problem with the AP1000 is not the design. It is the construction. The US built two AP1000 units at Plant Vogtle in Georgia. They were 7 years late and roughly 2.6 times over budget. Westinghouse went bankrupt during construction. A parallel project at V.C. Summer in South Carolina was abandoned entirely with $9 billion in sunk costs. China, meanwhile, built four AP1000s and completed them years faster, because China had a continuous construction workforce, a streamlined regulatory process, and the institutional coherence to learn from one unit and apply the lessons to the next in real time.

The AP1000 at Vogtle is now operating beautifully. The reactor is excellent. The construction management was catastrophic. We will dissect exactly why in Part 6, because the gap between “this is a good reactor” and “this reactor got built on time and on budget” is where the entire nuclear renaissance thesis lives or dies.

The EPR (European Pressurized Reactor, Framatome/EDF)

The EPR is a 1,600 MWe beast, the most powerful reactor design in commercial operation. It was designed by France’s Framatome (a subsidiary of EDF and Mitsubishi Heavy Industries) with an emphasis on safety redundancy: four independent safety systems (compared to two in most older designs), a “core catcher” to contain melted fuel in the event of a meltdown, and a double containment structure (a reinforced concrete inner shell plus a steel-lined outer shell designed to withstand the impact of a large aircraft).

The construction record is even worse than the AP1000’s.

Olkiluoto 3 in Finland: construction started 2005, originally due 2009, finally entered commercial operation in May 2023. Cost rose from roughly €3 billion (the fixed-price turnkey contract) to an estimated €11 billion all-in (combining TVO’s disclosed investment of €5.5 billion and Areva’s accumulated losses of €5.5 billion, which bankrupted the company and necessitated French government intervention). Eighteen years from first concrete to commercial operation.

Flamanville 3 in France: construction started 2007, originally due 2012, reached full power in late 2025. Cost rose from €3.3 billion to €13.2 billion on EDF’s books, or €23.7 billion including interest during construction according to the French Court of Auditors. That is four to seven times the original budget.

Hinkley Point C in the UK: construction started 2017, originally due 2025, now estimated for 2030 (base case per EDF’s February 2026 financial statement) with a cost of £35 billion in 2015 sterling, which translates to approximately £48.7 billion in 2026 prices (roughly $60+ billion at current exchange rates), up from the original £18 billion estimate. If it slips to 2031, add another £1 billion, bringing the worst case to £36 billion in 2015 prices.

The pattern across every Western new-build project is the same: the reactor works, the construction management does not. The reasons, which are institutional and financial rather than technical, are the subject of Part 6, and they are the single largest risk factor in the “nuclear renaissance” thesis.

A nuclear project ties up $15 to $35 billion for 8 to 15 years before generating a single dollar of revenue. During that entire period, the capital is accruing interest. This is interest during construction (IDC), and it is the mechanism that turns schedule delays into financial catastrophe.

At a 7% weighted average cost of capital, a $20 billion project that takes 7 years to build accumulates roughly $5 to $6 billion in IDC before commissioning. If that project slips an additional 5 years, IDC alone adds another $8 to $10 billion, and that compounding interest is itself accruing interest on the new, higher capital base. Vogtle’s final cost of roughly $36.8 billion includes an estimated $8 to $12 billion in IDC. Hinkley Point C’s escalation from £18 billion to roughly £49 billion nominal is substantially an IDC story. Flamanville’s explosion from €3.3 billion to €23.7 billion is an IDC story.

This is why nuclear is structurally different from every other generation technology. A gas plant that overruns by a year adds modest IDC on a $1 billion project with a 3-year build. A nuclear plant that overruns by five years is compounding interest on tens of billions across a decade-plus timeline. The implication for investors: construction schedule is the single most important variable in nuclear project economics. Overnight cost, cost per kilowatt, and LCOE estimates are all downstream of one question: how long does it actually take to build? Every 100 basis points on a $10 billion project adds roughly $100 million per year in interest expense during a 5-to-7-year construction period. That compounds into billions over the project lifecycle.

This is also why government-facilitated financing (DOE Loan Programs Office, regulated asset base models, sovereign credit guarantees) is not a nice-to-have but a structural prerequisite for Western nuclear construction. The UK’s Sizewell C learned from Hinkley Point C’s mistakes: it is financed under a Regulated Asset Base (RAB) model that allows construction costs to be recovered through consumer bills during construction, spreading IDC across ratepayers rather than concentrating it on the developer’s balance sheet. This is why Sizewell’s projected cost of £38 billion, roughly 20% cheaper than Hinkley for an identical design, is credible: the RAB model reduces the developer’s cost of capital, which reduces IDC, which breaks the compounding spiral. If Western nuclear is going to be built, the financing will look like RAB or Contract for Difference structures, not merchant market pricing. Investors evaluating new-build economics should be modeling these structures, not LCOE.

China, again, is the exception. It built two EPRs at Taishan and completed them roughly on time. South Korea builds its APR-1400 design reliably and at competitive cost. The problem is not nuclear physics. The problem is Western construction management, workforce continuity, and regulatory predictability. We will return to this theme repeatedly throughout this primer, because it is the single largest risk factor in the “nuclear renaissance” thesis.

2.8 From Physics to Logistics

We covered energy density in Part 1. Here is what it looks like when it hits the supply chain.

A uranium fuel pellet is a small ceramic cylinder of uranium dioxide (UO₂). It is about half an inch tall and half an inch in diameter. It weighs roughly 10 grams. It looks like a dark gray pencil eraser. We covered its energy content in Part 1, the “ten grams replaces a ton of coal” comparison. Here is what that energy density means when it hits the supply chain.

A single 1 GW nuclear reactor holds approximately 18 million of these pellets, arranged in fuel rods, bundled into fuel assemblies, loaded into the reactor core. The fuel stays in the reactor for about 4 to 5 years before being replaced. A typical refueling replaces about one-third of the fuel assemblies every 18 to 24 months.

Here is what that means for logistics. A 1 GW coal plant burns about 9,000 tons of coal per day. That is roughly 100 train cars, delivered every single day, year-round. The mine, the rail network, the unloading facility, the coal yards, the conveyor belts, the ash disposal: all of that infrastructure exists to feed the beast.

A 1 GW nuclear plant requires roughly 200 tonnes of natural uranium feed per year, which the enrichment process concentrates down to about 25 tonnes of enriched fuel actually loaded into the reactor. The annual fuel load fits in a handful of shipping containers.

There are no coal trains. No ash ponds. No conveyor belts. No daily deliveries. The fuel arrives in a truck a few times a year. That is it.

This is not a marginal difference. It is a categorical difference. It is the difference between a logistics nightmare and a logistics footnote. And it has a financial consequence that will matter enormously when we reach Part 8:because nuclear fuel is so energy-dense, fuel cost represents only about 10 to 15% of a nuclear plant’s total generation cost when measured on a levelized (new-build) basis, where capital costs dominate. For the existing operating fleet, where the capital is already sunk and depreciated, fuel’s share is somewhat higher, roughly 17% of total production cost per NEI’s 2023 data ($5.32/MWh out of $31.76/MWh total), because the denominator shrinks when you strip out capital recovery. Both figures are tiny relative to fossil fuels, where the fuel itself is 60-80% of the cost. The distinction matters for investors because the existing fleet thesis (Constellation, Kansai, CEZ) runs on operating cost economics, while the new-build thesis runs on levelized cost economics, and the fuel cost sensitivity is different in each case.

A $20/lb swing in the uranium price barely moves the needle for the utility that burns the fuel, even as it transforms the earnings of the company that mines it. That asymmetry between upstream and downstream sensitivity is one of the most important structural features of the nuclear investment landscape.

Part 3: The Nuclear Fuel Cycle

3.1 The Supply Chain Nobody Understands

Most people think nuclear fuel works like this: you dig up uranium, you put it in a reactor, electricity comes out. This is like saying oil works like this: you dig a hole, oil comes out, you put it in your car. Technically not wrong. Practically useless.

The nuclear fuel cycle is a five-stage industrial supply chain, and every single stage has its own economics, its own bottlenecks, its own geopolitics, and its own set of companies that control the choke points. Most nuclear primers treat this as a flat list of steps. That is why most nuclear primers are useless for investors. The fuel cycle is not a list. It is a sequential chain, and the tightest link in the chain sets the price for the entire system.

Here is the chain:

Mining & Milling → Conversion → Enrichment → Fuel Fabrication → Reactor → Back End (Storage/Reprocessing)

We are going to walk through each stage the way a supply chain analyst would, not the way a physics textbook would. At each stage, we care about three things: who controls the capacity, where the bottleneck is, and what happens if it breaks.

3.2 Stage 1: Mining and Milling (Digging Up the Yellowcake)

Uranium mining is, in principle, just mining. You find a deposit, you extract the ore, you process it into a concentrate called “yellowcake” (triuranium octoxide, U₃O₈), and you ship it to a converter. The yellowcake is a bright yellow powder, and yes, it looks exactly like it sounds. It is mildly radioactive, about as dangerous to handle as a bag of fertilizer (which, for the record, is not very dangerous if you are not eating it).

But uranium mining is not copper mining or gold mining. The economics are different, the geography is different, and the geopolitics are much different.

The Geography

In 2024, three countries produced 75% of the world’s uranium:

Kazakhstan: 39% Canada: 24% Namibia: 12%

Read that again. Nearly 40% of global uranium supply comes from a single landlocked Central Asian country that shares borders with Russia and China. If you have not thought about what Kazakhstan’s political alignment means for nuclear fuel security, you should start now.

Kazakhstan’s dominance is not an accident of geology. It is an accident of economics. Kazakhstan has large, low-grade uranium deposits in sandstone formations that are perfectly suited to a mining technique called in-situ recovery (ISR), which has dramatically lower costs than conventional mining.

ISR vs. Conventional Mining

Conventional uranium mining works like any other hard rock mining: you dig a pit or sink a shaft, haul out the ore, crush it, and leach it with acid to extract the uranium. It is capital-intensive, slow to permit, and environmentally intrusive. The Athabasca Basin deposits in Saskatchewan, Canada, are mined this way, and they are spectacular, with ore grades 100 times the world average (Cigar Lake runs at roughly 14% U₃O₈, compared to a global average around 0.1-0.3%). But they require hundreds of millions in upfront capital, years of permitting, and complex underground operations in challenging geological conditions.

ISR is completely different. Instead of bringing the rock to the surface, you bring the chemistry to the rock. You drill injection wells into a uranium-bearing sandstone aquifer, pump in a solution laced with an oxidant (sulfuric acid in Kazakhstan, alkaline carbonate solutions in the US), the solution dissolves the uranium from the sandstone grains, and then you pump the uranium-bearing solution up through production wells to a surface processing plant where the uranium is extracted via ion exchange.

No open pit. No underground shaft. No ore trucks. No tailings dam. The surface footprint looks like an oilfield, a grid of wellheads and piping, nothing more. The capital costs are a fraction of conventional mining, and the operating costs are even more favorable. This is why Kazakhstan, with its vast ISR-amenable deposits, went from a minor producer to the world’s largest in less than two decades.

The downside of ISR: it only works in specific geological settings (porous sandstone aquifers with uranium mineralization in a roll-front geometry), and it requires reliable supply of sulfuric acid, which has been a persistent problem for Kazakhstan. Kazatomprom, the state-owned national producer, has repeatedly flagged sulfuric acid shortages as a constraint on production, and in late 2023 it announced it would miss its production targets for 2024 and 2025, which was one of the catalysts that pushed uranium prices through $100/lb.

The Producers

The uranium mining industry is surprisingly concentrated. The top producers in rough order: Kazatomprom (Kazakhstan, state-owned, approximately 40% of global production on a 100% attributable basis across its JVs, or roughly 21-24% on a company-share basis), Cameco (Canada, publicly traded, ~15-18%), Orano (France, state-owned, mining in Kazakhstan, Canada, Niger), CGN Mining (China, state-affiliated), Uranium One (Rosatom subsidiary, operations in Kazakhstan), and a long tail of smaller producers and developers including Paladin Energy, Boss Energy, NexGen Energy, Denison Mines, and others.

For investors, the key dynamics are:

First, over half of global uranium mine production comes from state-owned entities. Kazatomprom, CGN, Orano, Uranium One. These companies do not always optimize for profit. They optimize for security of supply, strategic positioning, and sometimes just national prestige. This means the uranium market does not clear like a normal commodity market. Supply decisions are political as much as economic.

Second, the post-Fukushima bear market (2011-2020) destroyed the exploration and development pipeline. Mines were shuttered, exploration budgets were slashed, and an entire generation of mining engineers went into other commodities. The projects that were advancing were mostly ISR operations in Kazakhstan, which reinforced the geographic concentration.

The result is that the uranium market in 2025 faces a structural supply deficit that has been years in the making and will take years to resolve, because you cannot restart a conventional mine in 6 months. You need 5-10 years from discovery to production in most jurisdictions, and that assumes permitting goes smoothly (it rarely does).

A necessary counterpoint: the geographic concentration of uranium supply is frequently cited as a risk, and it is real, but it should not be confused with resource scarcity. Uranium is one of the most geologically abundant energy minerals on Earth. The concentration is economic and political, not geological. At sustained prices above $100/lb, deposits in Canada, Australia, the US, Namibia, and elsewhere become viable at scales that would meaningfully diversify supply within 5 to 10 years. The constraint is time, capital, and permitting, not the absence of uranium in the ground. The bull case does not depend on permanent scarcity. It depends on the lag between price signals and supply response, which in uranium is measured in years, not quarters.

Third, the Niger coup in July 2023 put a spotlight on jurisdictional risk. Orano’s large Somaïr and COMINAK operations in Niger were a meaningful share of global production. The military junta’s seizure of power introduced uncertainty about export approvals and operating permits that has not fully resolved. More broadly, the uranium mining map includes a number of politically unstable or strategically contested jurisdictions: Namibia, Uzbekistan, South Africa, Tanzania. The “friend-shoring” of uranium supply, redirecting procurement away from geopolitically risky sources toward Canada, Australia, and the US, is a live policy conversation that directly affects which mining companies benefit.

3.3 Stage 2: Conversion (The Bottleneck Nobody Talks About)

Once you have yellowcake (U₃O₈), it needs to be converted into uranium hexafluoride (UF₆), a chemical form that can be fed into enrichment centrifuges. UF₆ is a nasty compound, highly corrosive, reactive with moisture, and it sublimes directly from solid to gas at relatively low temperatures, which is exactly the property that makes it useful for enrichment (more on that in the next section).

Conversion sounds boring. It is not boring. It is one of the tightest bottlenecks in the entire nuclear fuel chain, and until recently, almost nobody outside the industry was paying attention to it.

The Players

There are effectively five conversion facilities in the world that matter:

Cameco’s Port Hope Conversion Facility (Ontario, Canada): approximately 12,500 tU/yr UF₆ nameplate capacity (with additional UO₂ and fuel bundle capacity bringing total facility throughput to roughly 14,000 tU/yr equivalent), the largest Western UF₆ converter. Cameco holds roughly 18% of world UF₆ conversion capacity and set a new production record in 2025 with 11.2 million kgU of UF₆ output.

Orano’s Malvési/Philippe Coste plants (Tricastin, France): approximately 15,000 tU/yr nameplate UF₆ capacity per Orano’s published figures. Orano’s new Philippe Coste plant was commissioned in 2018 and has been ramping up since, but has struggled with operational reliability. Production has hovered around 10,000-11,000 tU in recent years, well below nameplate.

ConverDyn/Honeywell Metropolis Works (Illinois, USA): approximately 15,000-17,600 tU/yr nameplate capacity (the higher figure reflecting a recent expansion). The only conversion facility in the United States.

It has had a troubled operational history, including labor disputes, extended shutdowns, NRC safety concerns, and a UF₆ release incident in 2023. ConverDyn resumed production in 2023 after a period offline, but its reliability remains a concern.

Rosatom/TVEL at Seversk (Russia): significant capacity, but largely serving Russia’s domestic market and politically inaccessible to Western utilities post-Ukraine.

CNNC (China): growing capacity, but reserved entirely for China’s domestic program.

That is it. For Western utilities, the accessible conversion supply comes from exactly three facilities: Port Hope, Philippe Coste/Malvési, and Metropolis. Three buildings. On two continents. Serving the entire non-Russian, non-Chinese nuclear fleet.

The Bottleneck

Conversion prices have increased roughly six-fold from end-2021 to mid-2024, reaching historic highs. This is not a speculative bubble. It is a supply-demand squeeze driven by the simple arithmetic of too few plants running at too-low utilization rates, with inventories drawn down and no new capacity coming online for years.

The Western conversion market operates at approximately 91% utilization. That leaves almost zero buffer for unplanned outages, and in an industry where “unplanned outages” are a regular occurrence (labor disputes at Metropolis, technical issues at Philippe Coste, maintenance requirements at Port Hope), a 91% utilization rate is essentially running on fumes.

For investors, conversion matters because it sits between two much more visible markets (uranium and enrichment) and amplifies price signals in both directions. A conversion shortage means utilities cannot feed their enrichment contracts, even if they have uranium. It is the coupling joint in the supply chain, and when it seizes up, everything downstream stops.

Cameco, as the operator of the world’s largest and most reliable Western conversion facility, captures disproportionate pricing power from this bottleneck. It is one of the underappreciated reasons Cameco trades at a premium to its uranium mining peers.

3.4 Stage 3: Enrichment (The Chokepoint That Keeps Generals Up at Night)

If you remember nothing else from this section, remember this: enrichment is the single most strategically sensitive, geopolitically consequential, and supply-constrained link in the nuclear fuel chain.

This is not hyperbole. The technology for enriching uranium is dual-use: the same centrifuges that enrich uranium to 4% for reactor fuel can, with additional cascading, enrich it to 90% for a weapon. This is why Iran’s enrichment program triggers international crises. It is why the Treaty on the Non-Proliferation of Nuclear Weapons exists. And it is why, of all the steps in the nuclear fuel cycle, enrichment is the one where the geopolitics matter most.

The Mixed Nuts Analogy

Imagine you have a 50-kilogram bag of mixed nuts. 99.3% of the nuts are almonds (U-238). 0.7% are cashews (U-235). You need the cashews, and you need the cashew concentration to go from 0.7% to about 4% of the total mix. The rest, the almonds, you discard (or store, because depleted uranium has some uses).

The problem: the almonds and the cashews look almost identical. Same color, same shape, same texture. The only difference is that the cashews are slightly lighter, about 1.3% lighter to be precise (uranium-235 has an atomic mass of 235 versus 238 for U-238, a difference of 3 atomic mass units out of 235).

You need a sorting machine that can separate nuts by a 1.3% weight difference, and you need it to run billions of nuts through the process, because to get the cashew concentration from 0.7% to 4%, you need to process roughly 8 to 10 kilograms of natural uranium for every kilogram of enriched product. The other 7-9 kilograms come out the other end as “tails,” depleted in U-235, and get stored in steel cylinders at the enrichment facility.

The sorting machine is a gas centrifuge. It spins UF₆ gas at extremely high speeds (50,000-70,000 RPM, generating forces of over a million Gs) inside a rotor the size of a trash can. The slightly heavier U-238 molecules migrate fractionally toward the outside of the rotor. The slightly lighter U-235 molecules concentrate fractionally toward the center. The enriched fraction is scooped off and fed into the next centrifuge. This is cascading: hundreds or thousands of centrifuges connected in series, each one making a tiny incremental separation, the cumulative effect producing usably enriched uranium.

SWU: The Currency of Enrichment

The unit of “sorting work” in the enrichment industry is the Separative Work Unit, or SWU (pronounced “swoo” by people in the industry and “S-W-U” by everyone else). A SWU does not measure the amount of enriched uranium produced. It measures the effort required to achieve a given degree of separation. The more separation you need (higher target enrichment, or lower tails assay), the more SWUs you need.

For a standard light water reactor fuel enrichment (from 0.711% natural to about 4.5% enriched, with tails at 0.25-0.30%), you need roughly 7-8 SWUs per kilogram of enriched product. At lower enrichments (3.5%, which is the bottom of the commercial range), the requirement drops to roughly 4.3-4.8 SWU/kg. The full range across commercial reactor fuels is approximately 4-9 SWU/kg depending on target enrichment and tails assay, which matters for fuel cost modeling.

SWU is priced independently of uranium: you can buy natural uranium from a miner and buy enrichment services (SWUs) from an enricher separately. The price of a SWU as of late 2024 was approximately $185-193 on the spot market, up from $47.50 in June 2020. That is a roughly 4x increase in four years.

Why this matters: the cost of nuclear fuel to a utility is a function of natural uranium prices, conversion prices, SWU prices, and fabrication costs. Enrichment (SWU) typically represents 30-40% of the total fuel cost. It is the largest single component. So when SWU prices quadruple, the economics of nuclear fuel change meaningfully, even if uranium prices stay flat.

The Market Structure: Russia’s Shadow

The global enrichment market is the most concentrated chokepoint in the entire nuclear supply chain. There are effectively four players:

Rosatom/TENEX (Russia): approximately 27 million SWU/year capacity across four facilities. That is roughly 40-44% of global capacity. Russia has the largest, most advanced, and cheapest enrichment infrastructure on Earth, a legacy of Cold War weapons production that was repurposed for commercial use.

Urenco (UK/Germany/Netherlands/USA joint venture): approximately 18 million SWU/year across four facilities (Capenhurst, Gronau, Almelo, and Eunice, New Mexico). Urenco is the principal Western enricher and is currently expanding capacity by 1.8 million SWU across all four sites, with the first new US cascades coming online in 2025.

Orano (France): approximately 7.5 million SWU/year at the Georges Besse II facility at Tricastin. Orano is studying a 30% capacity expansion and has selected Oak Ridge, Tennessee as the site for a potential new US enrichment facility.

CNNC (China): approximately 8-11 million SWU/year (estimates vary as Chinese capacity data is not publicly reported with precision) and growing rapidly, with upper-bound projections from Ux Consulting suggesting capacity could reach as high as 23 million SWU/year by 2030 under a maximum-build scenario, though most independent estimates cluster around 13-17 million SWU. But Chinese capacity is reserved for China’s domestic fleet and is functionally inaccessible to Western utilities.

Do the arithmetic. Total global capacity is roughly 60-65 million SWU/year. Russia controls about 27 million of that. China controls about 9 million, growing to 17 million. Neither is available to Western utilities going forward (Russia due to sanctions, China due to domestic reservation). That leaves Urenco and Orano with roughly 25-26 million SWU/year of capacity accessible to the West, against Western demand that currently runs around 35-40 million SWU/year.

The gap was historically filled by Russian supply. In 2024, Russia still supplied approximately 20% of total US enrichment needs (roughly 25% of foreign-origin SWU purchases), and about 23% of EU enrichment needs, down sharply from 31% in 2021 and 38% in 2023 per the Euratom Supply Agency.

The US banned imports of Russian enriched uranium effective August 2024 (with narrow waivers through January 2028), and Europe is moving toward disengagement under the REPowerEU framework, though without a hard ban.

This is the enrichment squeeze. Western utilities need to replace roughly 3.5-4 million SWU/year of Russian supply for the US market alone, and Urenco’s expansion adds only 1.8 million SWU across several years. Orano’s expansion is years from completion. Centrus Energy, the only other US-based enricher, is producing tiny HALEU (high-assay low-enriched uranium) quantities for the DOE, not commercial-scale LEU.

The market is responding. Urenco’s order book grew 27% in 2024 to €18.7 billion, extending into the 2040s. Utilities are signing longer-term contracts at higher prices. Spot SWU prices continue to rise. The structural deficit in Western enrichment capacity is not a forecast. It is a present-tense reality that will take until the early 2030s to meaningfully resolve, and that assumes all planned expansions execute on schedule (a generous assumption given the nuclear industry’s track record on construction timelines).

There is a second-order dynamic that the enrichment squeeze creates, and it is one that most uranium analysis misses entirely: overfeeding.

When SWU is expensive relative to uranium (as it is now, with spot SWU at roughly $193 versus historical norms of $50 to $100), utilities can adjust the enrichment process to use fewer SWUs per kilogram of product by raising the tails assay, the U-235 concentration left in the depleted uranium waste stream. Think of it this way: at a low tails assay (say, 0.20%), the centrifuges extract nearly every last atom of U-235, which requires more spinning (more SWU) but less raw uranium feed. At a higher tails assay (say, 0.30%), the centrifuges are less thorough, leaving more U-235 in the waste stream, which requires less spinning but more uranium to compensate.

The arithmetic is instructive. At a tails assay of 0.20% (historically standard when SWU was cheap), producing 1 kg of 4.5% LEU requires about 7.7 SWU and roughly 8.4 kg of natural uranium feed. At 0.30% tails (increasingly rational at current SWU prices), the same kilogram of LEU requires only about 6.2 SWU but needs roughly 10.2 kg of uranium feed. That is approximately 21% more uranium per kilogram of product.

Across a global fleet of 440 reactors, a widespread shift from 0.20% to 0.30% tails could add 10,000 to 15,000 tonnes of incremental annual uranium demand, roughly 15 to 20% above baseline consumption.

This is not theoretical. Utilities actively adjust tails assay in their enrichment contracts based on relative uranium/SWU pricing, and the current price relationship strongly incentivizes overfeeding. The investment implication is significant: the SWU shortage does not just tighten the enrichment market. It structurally inflates uranium demand, creating a second-order tailwind for uranium prices. This is one of the clearest examples of how pricing power flows backward through the fuel chain from the tightest bottleneck.

A note of self-interrogation is warranted here. Whenever an analytical finding perfectly supports the highest-conviction trade, the right instinct is suspicion. The counterargument is that utility inventories are larger and more resilient than public data suggests (utilities do not disclose fuel inventories in detail), that the shift to higher tails assays may be less uniform than the math implies, and that secondary supply from re-enrichment of existing tails could partially offset the incremental demand. The direction of the effect is robust. The magnitude is uncertain.

There is also a quiet, uncomfortable wrinkle that few analysts discuss publicly. Some evidence suggests that China is importing Russian enriched uranium to feed its domestic reactors, thereby freeing its own enrichment capacity for export to Western customers. If true, this means Western utilities are obtaining what is effectively Russian-origin enrichment through a Chinese intermediary. Western lawmakers are beginning to flag this as a potential backdoor channel, which could trigger future regulatory action and further tighten supply.

For investors, enrichment is the section of the fuel cycle with the strongest structural bull case and the fewest available equity proxies. Urenco is privately held (owned by the British, Dutch, and German governments). Orano is state-owned. Rosatom is sanctioned. That leaves Centrus Energy (NYSE: LEU) as essentially the only publicly traded pure-play enrichment company in the West, which goes a long way toward explaining why its stock has been one of the best performers in the nuclear sector.

3.5 Stage 4: Fuel Fabrication (The Relatively Boring One)

After enrichment, the enriched UF₆ is shipped to a fuel fabrication facility, where it is converted (de-converted, technically) from UF₆ gas back into uranium dioxide (UO₂) powder, pressed into those small ceramic pellets we talked about in Part 2, sintered (heated to about 1,750°C to densify them), ground to precise dimensions, stacked into metal tubes (zirconium alloy cladding, typically Zircaloy), sealed, and assembled into fuel assemblies. Each assembly is a precise, engineered component containing hundreds of fuel rods arranged in a specific geometry optimized for the reactor it is going into.

Fuel fabrication is more commoditized than the upstream stages, but it is still concentrated among a handful of players: Westinghouse (US/Sweden), Framatome (France, Orano subsidiary), TVEL (Russia, Rosatom subsidiary), CNNC/CJNF (China), and KEPCO Nuclear Fuel (South Korea). In the West, Westinghouse and Framatome dominate.

The strategic significance of fuel fabrication is more about lock-in than scarcity. Fuel assemblies are designed for specific reactor types, and switching fabricators requires re-engineering and regulatory re-qualification, which takes years. This means that once a utility is locked into a fabrication supplier, switching costs are high. It also means that Rosatom’s fuel fabrication business, particularly for VVER reactors (Russia’s export reactor design), creates a dependency chain: countries that bought Russian reactors are often locked into Russian fuel assemblies, which locks them into Russian enrichment, which locks them into a long-term strategic relationship with Moscow.

This is precisely why the transition of Eastern European VVER operators (Bulgaria, Czech Republic, Slovakia, Finland) away from Russian fuel assemblies and toward Westinghouse alternatives is such a significant strategic and commercial development. It is breaking Rosatom’s lock-in, one fuel assembly at a time.

For investors, fuel fabrication is not where the action is in the current cycle. The margins are relatively thin, the barriers to entry are high but stable, and the supply-demand dynamics are less acute than in enrichment or conversion. But watch Westinghouse (owned by Cameco and Brookfield) and Framatome (Orano subsidiary) as strategic assets in the broader nuclear renaissance.

3.6 Stage 5: The Back End (What Do You Do With the Waste?)

This is the section that anti-nuclear advocates want you to read first and pro-nuclear advocates want you to skip. We are going to do neither. We are going to lay out the facts, explain the two competing philosophies, and let you decide how much of a problem this actually is.

After about 4-5 years in the reactor, a fuel assembly is “spent.” It has used up enough of its U-235 to no longer sustain an efficient chain reaction (though it still contains about 95% of its original uranium, plus small amounts of plutonium and other transuranic elements created during irradiation). The spent fuel is intensely radioactive and thermally hot. It needs to be managed. How it gets managed depends entirely on which of two philosophical camps your country falls into.

Option A: Direct Disposal (The American Model)

In the United States, spent fuel is classified as high-level radioactive waste and is intended for permanent geological disposal, burial in a deep underground repository designed to isolate the waste for hundreds of thousands of years.

The keyword in that sentence is “intended.” The US has never built a permanent repository.

The saga of Yucca Mountain is a case study in how democracy, geology, and nuclear politics collide. In 1987, Congress designated Yucca Mountain in Nevada as the site for the nation’s permanent spent fuel repository. The DOE spent $15 billion studying the site. In 2008, it submitted a license application to the NRC. In 2010, the Obama administration defunded the project, largely due to opposition from Nevada senator Harry Reid. The license application was withdrawn. No alternative was designated.

As a result, spent fuel in the United States is stored “temporarily” at the reactor sites where it was generated. First in spent fuel pools (large pools of water that provide cooling and radiation shielding), then, once cooled sufficiently (5-10 years), transferred to dry cask storage, massive concrete and steel containers that sit on concrete pads at the reactor site.

There are currently over 90,000 metric tons of spent nuclear fuel stored at roughly 75 reactor sites across the United States. It is perfectly safe in its current configuration, dry casks are designed to withstand earthquakes, tornadoes, floods, and aircraft impacts, but it is not a permanent solution, and the legal and political status of the waste remains unresolved.

The cost is real but manageable. US utilities pay a fee of 1 mill (one-tenth of a cent) per kilowatt-hour of nuclear electricity into the Nuclear Waste Fund, which has accumulated over $50 billion (including interest), most of which the government has not spent because there is no repository to spend it on. The Department of Energy is legally liable for failing to take possession of the waste as required by the Nuclear Waste Policy Act, and has paid billions in damages to utilities for breach of contract.

Option B: Reprocessing (The French Model)

France took a fundamentally different approach. Rather than treating spent fuel as waste, France treats it as a resource to be recycled.

At the La Hague reprocessing plant in Normandy, operated by Orano, spent fuel assemblies are dissolved in nitric acid. The solution is then put through a chemical separation process (PUREX, for Plutonium Uranium Extraction) that separates three streams: recovered uranium (~95% of the original), plutonium (~1%), and high-level fission products (~4%).

The recovered uranium can be re-enriched and made into new fuel. The plutonium is mixed with depleted uranium to create MOX (Mixed Oxide) fuel, which can be used in standard PWRs. The high-level fission products are vitrified (mixed into glass) and stored in steel canisters for eventual deep geological disposal.

The result is dramatic waste volume reduction. France generates about 75% of its electricity from nuclear and has been doing so for decades, yet the total volume of high-level vitrified waste produced by the entire French nuclear program, all of it, fits into a single room beneath La Hague. That is not a metaphor. It is a literal room.

The trade-off is that reprocessing is expensive, it involves separating plutonium (a proliferation concern), and it does not eliminate the need for a geological repository (you still need to store the vitrified fission products somewhere permanent). France is building its deep geological repository, Cigéo, in clay formations in northeastern France, scheduled for operations starting in the 2030s.

The United Kingdom, Japan, and Russia also reprocess spent fuel. The US abandoned reprocessing in the 1970s under President Carter due to proliferation concerns, and has never revisited the policy in a serious way.

Putting the Waste “Problem” in Context

Here is the number that reframes the entire debate: all of the spent nuclear fuel ever generated by the entire US commercial nuclear fleet, more than 60 years of electricity generation providing roughly 20% of the country’s power, fits in a volume equivalent to a football field stacked about 10 yards high (though decay heat from spent assemblies requires wide spacing in any actual repository, making the facility footprint far larger than the raw volume implies). That is it

The total physical mass of the “waste problem” is roughly 90,000 metric tons, sitting on concrete pads at reactor sites, doing nothing harmful to anyone.

The contrast with coal ash, which is toxic, voluminous, and routinely leaches into groundwater from unlined disposal ponds, is categorical rather than incremental. Nuclear waste is compact, contained, solid, and has never harmed a single member of the public in the history of commercial nuclear power. The “waste problem” is a political problem, not a technical one, and it has been solved, technically, by multiple countries. The US just has not chosen to implement any of the solutions.

3.6.1 Decommissioning: The Back-End Cost Nobody Budgets For

There is a cost at the end of a reactor’s life that most nuclear investment analyses treat as a footnote: decommissioning. The process of safely dismantling a nuclear power plant, decontaminating the site, and managing the resulting waste takes 15-30 years and costs between $500 million and $1.5 billion per reactor, depending on size, design, and regulatory requirements. The NRC’s 2024 estimates range from $280 million to $612 million per unit in the most favorable cases, but actual completed decommissioning projects in the US (Yankee Rowe, Connecticut Yankee, Maine Yankee) have run $400 million to $1 billion per unit, and the UK’s Nuclear Decommissioning Authority estimates total lifetime decommissioning costs across the British nuclear legacy at £99-232 billion (though this includes military and early research facilities, not just commercial power stations).

US nuclear utilities are required to maintain decommissioning trust funds, funded by ratepayer contributions during the reactor’s operating life. As of 2024, the industry held approximately $95 billion in decommissioning trust assets, which the NRC deems adequate for current obligations. The good news: for operating reactors with decades of remaining life, the trust funds will continue to compound. The concern: utilities that shut reactors prematurely (like Entergy’s Vermont Yankee, Indian Point, and Pilgrim) sometimes found their trusts underfunded relative to actual costs, requiring corporate backstops.

For investors, decommissioning is a manageable but real liability that should be factored into any honest lifecycle cost comparison between nuclear and alternatives. It is also a growing business opportunity: companies like Holtec International and EnergySolutions are building decommissioning-as-a-service businesses, purchasing retired reactors from utilities and managing the dismantlement process. Holtec, which also owns the Palisades restart project, is effectively a decommissioning company that is pivoting to become a nuclear operator, an unusual but strategically logical evolution.

The analytical point: any LCOE comparison between nuclear and alternatives should include decommissioning costs (typically adding $1-3/MWh to nuclear’s levelized cost, depending on assumptions). Most published LCOE figures do include this, but investors should verify, because the ones that look suspiciously cheap sometimes do not.

3.7 The Supply Chain as an Investment Framework

Now that you understand the five stages, here is the framework that matters for investors:

The value chain flows left to right: Mining → Conversion → Enrichment → Fabrication → Reactor. But the pricing power flows right to left. The tightest bottleneck in the chain sets the clearing price for everything upstream.

Right now, enrichment is the tightest bottleneck. Western SWU capacity is structurally insufficient to replace Russian supply within this decade. Conversion is the second-tightest, with only three Western facilities operating at over 90% utilization. Mining is tight but responsive; higher prices are bringing mothballed capacity back online and incentivizing new development.

This means:

Enrichment companies (Centrus, and eventually Urenco/Orano if they ever go public or take investments) have the strongest structural pricing power. SWU prices have quadrupled and are not coming back down.

Conversion-advantaged producers (Cameco, which owns the world’s most reliable conversion facility) capture dual premiums: once on the uranium they mine, and again on the conversion services they provide. This is why Cameco is not “just a miner.”

Miners with near-term production in friendly jurisdictions (NexGen in Canada, Boss Energy in Australia, Paladin in Namibia, Denison in Canada) benefit from the overall uranium price environment, but their leverage to the enrichment/conversion bottleneck is indirect.

Physical uranium vehicles (Sprott Physical Uranium Trust, Yellow Cake plc) give exposure to the uranium price without mining risk, but they do not capture the conversion or enrichment premiums.

Here is the fuel cycle in one table:

Understanding this hierarchy, and where each investment sits within it, is the difference between a portfolio that captures the nuclear thesis and a portfolio that merely owns some uranium miners and hopes for the best.

Part 4: The Uranium Market

4.1 This Is Not a Normal Commodity Market

Uranium does not trade like copper. It does not trade like oil. It does not trade like gold. If you try to analyze the uranium market using the mental models you developed trading literally any other commodity, you will get it wrong. Not a little wrong. Fundamentally wrong.

Here is why.

There is no meaningful futures market for uranium. No NYMEX contract. No ICE benchmark. No pit traders in colored jackets screaming at each other. Uranium is not traded in meaningful quantities on any commodity exchange. The market is principally based on private, bilateral negotiations between a small number of producers and an even smaller number of utility buyers, conducted through intermediaries and consultants, with prices reported by two specialist firms (UxC and TradeTech) that survey the market weekly and monthly.

This means several things that matter enormously for investors:

Price discovery is opaque. You cannot pull up a real-time uranium chart on your Bloomberg Terminal the way you can for crude oil. The “spot price” you see reported is a survey-based indicator derived from a handful of transactions, sometimes just one or two per week, each involving a few hundred thousand pounds. The “market” in any given week might be three to five trades. In a thin week, it might be one.

The market is dominated by long-term contracts, not spot trading. Approximately 85-90% of all uranium is purchased under multi-year term contracts between producers and utilities, typically with delivery windows stretching 5-10 years. The spot market accounts for only 10-15% of total volumes. This means the spot price, the number that shows up on financial websites and in analyst reports, is a residual market, a sideshow. The term price is where the real business happens.

And here is the dynamic that most analysts, especially those who come from equity or macro backgrounds, consistently underestimate: utilities are the slowest buyers in any commodity market on Earth. A nuclear utility’s fuel procurement cycle operates on a timeline measured in years, not months. They plan fuel loads 2-3 years in advance. They negotiate contracts over quarters, sometimes years. They have internal committees, board approvals, regulatory filings. They are constitutionally incapable of acting quickly, and they are rewarded for delay because uranium fuel represents only about 5-10% of a nuclear plant’s total operating cost. If they pay 30% more for uranium, the impact on the cost of electricity to the consumer is measured in fractions of a cent per kilowatt-hour. They can afford to wait.

This means the uranium market has a persistent asymmetry: when prices fall, production gets curtailed quickly (miners shut mines at cash cost), but when prices rise, demand responds with a lag measured in years as utilities slowly, grudgingly, eventually sign contracts. The bull case in uranium is always a story of inevitability delayed. The bears always sound right for longer than they should, because they are observing the spot market while the real action happens in conversations between procurement officers and mining executives that will not become public for months.

4.2 A Brief History of Uranium Cycles (Or: How to Lose Money Being Right)

Understanding where we are requires understanding where we have been.

The First Cycle (2003-2007): Cigar Lake Breaks and the Market Goes Insane

In 2003, uranium was trading at roughly $10/lb. Nobody cared about it. The post-Cold War era had created a massive overhang of secondary supply: the US-Russia Megatons to Megawatts program was converting Russian weapons-grade highly enriched uranium (HEU) into commercial reactor fuel, flooding the market with roughly 24 million pounds of uranium equivalent per year. Mines were shuttering. Exploration was dead. The word “uranium” at an investment conference would empty the room.

Then two things happened. First, the Megatons to Megawatts program had a fixed end date (2013), and the market slowly began pricing in the eventual loss of that supply. Second, in October 2006, Cameco’s Cigar Lake mine, the largest undeveloped high-grade uranium deposit in the world, flooded during construction. The mine that was supposed to supply 18 million pounds per year, roughly 13% of global demand, was suddenly gone. The market panicked. Speculative capital rushed in. Uranium prices spiked from $40/lb in early 2006 to a peak of $136/lb in June 2007.

Then they crashed. The price fell back below $50 by late 2008 as the global financial crisis destroyed risk appetite, and speculators who had driven the price up vomited their positions. Cigar Lake eventually came online in 2014, eight years late.

The lesson: uranium price spikes driven by speculative capital and supply shocks can be violent but are not sustainable without a change in underlying demand fundamentals. The 2007 spike was real in its supply-side catalyst but synthetic in its magnitude.

The Second Cycle (2011-2020): Fukushima and the Lost Decade

On March 11, 2011, the Fukushima disaster destroyed not just three reactor cores but the entire demand thesis for uranium. Japan shut down 54 reactors. Germany accelerated its nuclear exit. Global uranium demand dropped roughly 10-15% almost overnight. Meanwhile, supply from Kazakhstan continued to ramp (Kazatomprom was aggressively growing ISR production regardless of price signals, because it was a state entity optimizing for market share, not profitability).

The result was a decade of oversupply. The spot price ground lower, from $70/lb pre-Fukushima to a bottom of roughly $18/lb in November 2016. At $18, uranium was below the all-in cost of production for virtually every mine on Earth except the lowest-cost ISR operations in Kazakhstan. Mines closed. Cameco idled McArthur River and Cigar Lake. Paladin Energy put Langer Heinrich into care and maintenance. Exploration budgets were annihilated. An entire generation of uranium projects was shelved.

This is the decade that created the conditions for the current bull market. A decade of underinvestment in supply, combined with the drawdown of above-ground inventories, the depletion of secondary supply sources, and the beginning of a demand recovery (Japanese restarts, Chinese construction, life extensions in the US and Europe), set up the structural deficit that we are now living through.

The Current Cycle (2020-Present): Structural Deficit Meets Geopolitical Disruption

The current uranium market is defined by a convergence of demand growth, supply constraints, and geopolitical disruption that is qualitatively different from prior cycles.

On the demand side: the global reactor fleet set a generation record in 2024 (2,667 TWh). China is building reactors at a pace of 6-10 per year. Over 30 countries pledged to triple nuclear capacity by 2050. Life extensions for existing reactors are adding decades of fuel demand. Data center PPAs from tech companies are creating a new category of demand that did not exist in prior cycles.

On the supply side: global primary uranium production met only about 90% of reactor demand in 2024, with the remaining 10% covered by drawdowns of inventory and secondary supplies. Kazatomprom missed its production targets in 2024 and cut nominal 2026 production guidance. The Niger coup disrupted Orano’s operations (zero production at SOMAÏR in 2025 under junta control). US ISR restarts have ramped more slowly than planned. Cigar Lake and McArthur River are producing but at lower rates than peak capacity.

On the geopolitical side: the US ban on Russian enriched uranium (effective August 2024) and the broader Western effort to disentangle from Russian fuel supply has disrupted the market’s historical logistics. Kazakhstan’s uranium, which accounts for 39% of global supply, must transit either through Russia or via the Trans-Caspian route through the Caspian Sea, Azerbaijan, Georgia, and Turkey, a long and geopolitically fraught supply chain that has already experienced delays.

The spot price in 2025 traded in a range of roughly $63 to $83/lb, frustratingly flat for investors who expected a runaway bull market, before surging above $100/lb in January 2026 on aggressive buying from Sprott and tightening physical availability.

But the spot price is, as we discussed, a sideshow. The term price, the one that actually matters, climbed steadily from $80 to $86 over the course of the year, and utilities began accepting higher base prices in new contracts. Conversion term pricing hit historic highs with a 27% yearly increase. SWU spot prices sat at roughly $193/SWU, up from $47.50 five years ago.

The market is in a stalemate. Producers are reluctant to sell forward at current prices, believing they will go higher. Utilities are reluctant to contract at current prices, hoping they will go lower. This standoff is familiar to anyone who has traded commodity cycles. It always breaks in the same direction: the buyer blinks, because the buyer is the one who has to fuel a reactor. You can choose not to mine uranium. You cannot choose not to fuel a reactor that 600,000 homes depend on for electricity.

In 2025, approximately 116 million pounds of uranium was placed under long-term contracts, below the industry’s estimated replacement rate of approximately 150 million pounds (and well below the roughly 185 million pounds of annual reactor demand). This means uncovered utility requirements are growing larger each year the stalemate persists. The pent-up demand is building. When it releases, whether in 2026 or 2027, the contracting wave will be substantial, and it will be at higher prices than today.

4.3 Secondary Supplies: The Buffer That’s Running Out

One of the most important and least understood dynamics in the uranium market is the role of secondary supplies, material that reaches the market through routes other than primary mine production.

Historically, secondary supplies have been enormous. During the 1990s and 2000s, they included:

Government stockpile releases (US DOE, Russian HEU downblending under Megatons to Megawatts). Utility inventory drawdowns (utilities ran down their fuel reserves, buying less new uranium than they consumed). Enricher underfeeding (enrichment plants can operate their centrifuges to extract more U-235 from the tails than the contract specifies, effectively creating “extra” uranium that can be sold back into the market). Reprocessed and re-enriched uranium (MOX fuel, RepU). Military surplus.

At their peak, these secondary supplies covered roughly one-third of global reactor demand. This was the elephant in the room that kept uranium prices suppressed for decades. Even as primary mine production was insufficient to meet demand, the market balanced because there was a massive buffer of above-ground material being drawn down.

That buffer is now largely depleted.

The Megatons to Megawatts program ended in 2013. US DOE stockpile sales have been curtailed by legislation. Utility inventories, after a decade of drawdowns following Fukushima, are at their lowest levels in years. Enricher underfeeding, which was substantial when SWU prices were low and uranium prices were low (making it cheap to “overfeed” the centrifuges and sell the excess natural uranium equivalent), has reversed as SWU prices have quadrupled, the economics now favor overfeeding (using more SWUs and less natural uranium), which increases uranium demand rather than substituting for it.

The transition from an inventory-driven market to a production-driven market is one of the most important structural shifts in uranium in two decades. It means that incremental demand must be met by incremental mine production, not by drawing down a stockpile. And incremental mine production takes 5-10 years to bring online.

This is why the “structural deficit” thesis has legs. The bears will point out (correctly) that the deficit has been smaller than the bulls predicted in some years, and that above-ground inventory acts as a shock absorber. But the shock absorber is getting thinner each year, and the demand trajectory is getting steeper. The math eventually overwhelms the buffer.

4.4 The Structural Deficit: Real, but More Complicated Than the Bulls Admit

Let me give you the bull case and the bear case honestly, because TSCS does not do advocacy.

The Bull Case

Global reactor demand is roughly 180-185 million pounds per year and growing. Primary mine production covers roughly 155-165 million pounds (UxC estimates approximately 164 million pounds for 2025). The gap of 25-35 million pounds per year is being filled by inventory drawdowns and secondary supplies that are depleting. New mine supply takes 5-10 years to bring online. Therefore, prices must rise to incentivize enough new production to close the gap, and in the interim, inventory drawdowns will tighten the market further.

This case is straightforward, well-supported by data, and directionally correct.

The Bear Case (or at least, the nuances the bulls skip)

First, the deficit is measured against requirements, not actual purchases. Utilities can and do undercontract, running down inventories, delaying procurement, and managing their fuel cycles to defer purchases. The “deficit” in any given year is partly real and partly a reflection of utility timing decisions. Utilities are not going to run out of fuel. They are going to run out of optionality, which is a different thing.

Second, supply elasticity is not zero. At $80+/lb, projects that were uneconomic at $40 are now viable. ISR operations in Wyoming and Texas are restarting. Boss Energy’s Honeymoon in Australia is ramping. Paladin’s Langer Heinrich in Namibia achieved commercial production in 2024. NexGen’s Arrow project in Canada is the most significant new high-grade deposit in a generation. The question is not whether new supply will come, it is whether it will come fast enough to match demand growth, given permitting timelines and construction schedules.

Third, China’s domestic uranium mining and enrichment buildout is wild card. If China successfully scales domestic capacity to self-sufficiency, it releases Kazakh and other supply back into the global market. China currently aims to source one-third of its fuel domestically, one-third from equity stakes in foreign mines, and one-third on the open market. If the domestic third grows faster than expected, the supply picture loosens.

Fourth, the demand projections assume that all announced reactor construction, life extensions, and restarts actually happen, on time. History suggests at least some of them will be delayed or cancelled. The demand side has optionality that the bulls sometimes assume away.

The Synthesis

The structural deficit is real in direction. Primary supply growth cannot match the trajectory of demand growth within this decade without significantly higher prices. The secondary supply buffer is thinning. The geopolitical disruption to Russian and Kazakh supply adds a risk premium that was not in the market five years ago.

But the magnitude and timing of the deficit are uncertain. It could manifest as a sharp price spike (if a major supply disruption coincides with a contracting wave), or as a slow grind higher (if supply restarts and new projects gradually close the gap). The base case for uranium is higher for longer, not a vertical spike. If you are positioning for $200/lb uranium on a specific timeline, you are making a timing bet, not a fundamental bet. The fundamentals say prices need to be higher. They do not say when.

4.5 What This Means for Investors

The uranium market’s unusual structure creates specific investment implications:

Physical uranium vehicles (Sprott Physical Uranium Trust, Yellow Cake plc) give you direct commodity exposure without mining risk. SPUT added 7.8 million pounds in 2025, growing its holdings to over 74 million pounds, making it one of the largest holders of physical uranium on Earth. These vehicles effectively remove supply from the market, tightening the spot balance. They are a way to own the “higher for longer” thesis without betting on any individual miner’s execution.

Producers with near-term cash flow (Cameco, Kazatomprom, Paladin, Boss Energy) benefit from the term price moving higher, as they roll legacy contracts signed at lower prices into new contracts at current market levels. Cameco is the best-positioned: it is a miner, a converter, and (through its Westinghouse JV) a fuel fabricator, capturing value at three stages of the fuel chain. The premium multiple Cameco trades at relative to its peers reflects this integration.

Developers (NexGen, Denison, Fission Uranium, Bannerman, Global Atomic, IsoEnergy, enCore) are leveraged bets on higher uranium prices and successful project execution. NexGen’s Arrow deposit in the Athabasca Basin is arguably the most important new uranium mine globally, with grades and scale that rival Cigar Lake. But it is still in permitting, and the road from discovery to production is long and full of risk. Developers are where you get the most upside if the cycle plays out, and the most downside if it doesn’t.

Enrichment proxies (Centrus Energy) capture the SWU pricing dynamic, which has a stronger structural case than uranium itself. As discussed in Part 3, Western enrichment capacity is the tightest bottleneck in the fuel chain, and Centrus is the only publicly traded Western enricher.

The key mental model for uranium investing is patience. Uranium cycles are long. The bull case took three years longer to play out than most analysts expected. Utilities will be the last actors to confirm the thesis through their contracting behavior. If you need to be right quickly, this is the wrong market. If you can be right slowly, the asymmetry is compelling.

Part 5: The Geopolitics of Nuclear Fuel

5.1 The Map Is the Trade

Every commodity has a geography. Oil has the Strait of Hormuz. Natural gas has the Yamal Peninsula. Copper has Chile. These are not footnotes in an investor presentation. They are the load-bearing walls of the global economy, and when they crack, everything built on top of them shifts.

Nuclear fuel has a geography too. And it is, to put it diplomatically, a disaster.

Forty percent of the world’s uranium comes from a landlocked Central Asian country sandwiched between Russia and China. Forty-four percent of the world’s enrichment capacity sits inside Russia. The only commercially operating conversion facility in the United States has a history of labor disputes, safety incidents, and extended shutdowns. The West’s newest major uranium mine, in Niger, was nationalized by a military junta that promptly invited Rosatom to the table.

If you designed a supply chain specifically to maximize geopolitical vulnerability, you could not do much better than the nuclear fuel cycle as it exists today. The question for investors is not whether this geography creates risk. It does. The question is whether the Western response to that risk, which is now funded, legislated, and accelerating, will succeed in time to matter. And who benefits while the transition unfolds.

5.2 Kazakhstan: The Saudi Arabia of Uranium, with a Russia Problem

Kazakhstan produces 39-43% of the world’s uranium. This is not a typo. A single country, with a GDP smaller than Norway’s, supplies nearly half of the fuel that powers 10% of global electricity. The Saudi Arabia comparison is not hyperbole. It is structural. And like Saudi Arabia’s relationship with oil, Kazakhstan’s relationship with uranium is complicated by the fact that its two largest neighbors are the two countries the West is most actively trying to decouple from.

Kazatomprom, the state-owned producer, is the world’s largest uranium mining company. It produced 23,270 tonnes in 2024, a 10% increase year-on-year but still 7-9% below its own target.

In August 2024, Kazatomprom slashed its initial 2025 production intention of 30,500-31,500 tonnes (100% basis) down to a formal guidance range of 25,000-26,500 tonnes, an approximately 17% reduction driven by sulfuric acid supply constraints and delayed infrastructure construction at JV Budenovskoye (which missed its subsoil use agreement target by roughly 65-80%).

Actual 2025 output of approximately 25,839 tonnes fell within this revised guidance, so the company technically delivered on its downgraded plans. But the key point for investors is that Kazatomprom voluntarily lowered the bar, not that it missed it. The original intention of 30,500-31,500 tonnes was a subsoil use agreement entitlement, not a production forecast, and the gap between entitlement and actual output reveals structural constraints (acid supply, infrastructure delays) that are not resolving quickly.

For 2026, Kazatomprom reduced its subsoil use agreement production level by roughly 10% to 29,697 tonnes and signalled it would exercise its 20% downflex option, meaning actual 2026 production could come in as low as approximately 23,750 tonnes. The company explicitly cited “insufficient uncovered demand” to justify restraint, a polite way of saying “we are not going to pump into a market where our biggest customers are not signing contracts.”

The persistent shortfalls trace to a mundane but critical bottleneck: sulfuric acid. Kazakhstan’s ISR mining method requires enormous quantities of the stuff, roughly 1.85 million tonnes per year, and supply has been disrupted by agricultural demand competition and regional logistics problems. Acid costs rose approximately 33.6% over the course of 2023-2024, per Kazatomprom’s financial disclosures.

Kazatomprom is building a new 800,000-tonne-per-year acid plant, but commissioning has slipped to late 2026 at the earliest. Until then, production will remain constrained by chemistry, not geology.

The Transit Problem

Here is a detail that should keep Western fuel procurement officers awake at night. Roughly 90% of Kazakh uranium exports have historically traveled by rail through Russia to the port of St. Petersburg, and then by ship to Western buyers. In a world where the West is actively sanctioning Russian enrichment and trying to decouple from Russian supply chains, the inconvenient fact is that most of the “non-Russian” uranium the West buys from Kazakhstan transits through Russia.

The alternative is the Trans-Caspian International Transport Route, the so-called “Middle Corridor,” which ships uranium from the Kazakh port of Aktau across the Caspian Sea to Azerbaijan, then by rail through Georgia and Turkey to European ports. Cameco, which operates the Inkai JV in Kazakhstan, has publicly called this route “neither reliable nor predictable.” A test shipment in 2022 took months. Total Middle Corridor trade volumes actually dropped 37% from 2022 to 2023 due to congestion, pushing traffic back to the Russian route.

This is the kind of detail that never appears in a sell-side uranium thesis but matters enormously for risk pricing. Kazakhstan’s uranium is only as “Western-friendly” as the route it travels. And right now, the dominant route goes through the country the West is trying to cut off.

China’s Quiet Takeover

While the West debates transit routes, China is solving the problem differently: it is buying Kazakhstan outright.

More than 50-60% of Kazatomprom’s output now flows to Chinese buyers. CGN holds 49% stakes in multiple mining JVs. CNNC has invested in additional deposits. And in December 2024, in a transaction that received remarkably little Western media attention, Uranium One (Rosatom’s mining subsidiary) sold its 49.98% stake in JV Zarechnoye and 30% stakes in JVs Khorasan-U and Kyzylkum to Chinese state entities. That single transaction transferred roughly 4,831 tonnes of annual production capacity from Russian to Chinese control.

In 2024, Kazatomprom signed a series of major long-term supply agreements with Chinese state entities, reportedly valued at a substantial fraction of the company’s total book assets. This is not a one-off supply contract. It is a strategic alignment, reinforced by CGN’s 49% stakes in multiple mining JVs and the December 2024 transfer of Uranium One equity to Chinese buyers.

Kazakhstan is doing what small countries between empires have always done: playing both sides. President Tokayev maintains a “permanent comprehensive strategic partnership” with Beijing, bilateral trade hit a record $44 billion in 2024, and China was selected to lead Kazakhstan’s second and third nuclear power plants. Simultaneously, Kazakhstan refused to join BRICS+, declined to recognize Russian-occupied Ukrainian territories, and welcomed Macron for the first French presidential visit since 1994. The country held a nuclear power referendum in October 2024, which passed with 71% approval, but awarded the first reactor contract to Rosatom anyway.

For Western utilities, the message is uncomfortable: Kazakhstan’s uranium is increasingly Chinese uranium. Not because Beijing owns the deposits, but because it owns the offtake contracts, the JV equity stakes, and the political relationship. The assumption that Kazakh supply is “available” to the West needs to be stress-tested against the reality that China is locking it up.

5.3 Russia: The Enrichment Chokehold

We covered the enrichment market structure in Part 3. Here is the geopolitical overlay.

Rosatom is not a company. It is a state within a state. It controls approximately 27 million SWU of enrichment capacity (40-44% of global total), supplies enrichment services to over 30 utilities across 16 countries, operates nuclear fuel fabrication for dozens of VVER reactors worldwide, builds reactors in Turkey, Egypt, Bangladesh, and India, and generates up to half of Russia’s foreign currency revenue outside hydrocarbons. It is, by any reasonable measure, Russia’s single most important strategic industrial asset after Gazprom.

The US Prohibiting Russian Uranium Imports Act, signed May 13, 2024, banned imports of Russian-origin low-enriched uranium effective August 11, 2024, with waiver provisions through January 1, 2028 and declining annual caps. The ban runs through 2040. Russia retaliated on November 15, 2024, imposing its own export ban on enriched uranium shipments to the US.

The European picture is messier. The EU has no hard ban on Russian nuclear fuel. Under the REPowerEU framework, Europe is “working toward” reducing dependence, and progress has been faster than expected: the share of EU enrichment sourced from Russia dropped from 38% in 2023 to 23% in 2024. But five EU member states still operate 19 Russian-designed VVER reactors that historically depended on specialized fuel assemblies made exclusively by Rosatom’s TVEL subsidiary. Bulgaria, Czech Republic, Finland, Slovakia, and Hungary all fall into this category.

The lock-in works like this. Rosatom does not just sell fuel. It sells an ecosystem. A country that buys a VVER reactor also buys VVER fuel assemblies (hexagonal geometry, incompatible with Western PWR fuel), which requires Russian enrichment, which is bundled with Russian conversion services, which creates a dependency chain that lasts the reactor’s entire operating lifetime (40-60 years). This is the nuclear equivalent of the razor-and-blade model, except the blades are enriched uranium and the switching cost is measured in billions.

Breaking that lock-in has been one of the most significant and underreported industrial achievements of the past two years. Westinghouse now manufactures VVER-1000 fuel at facilities in Sweden and the UK, and has been supplying Ukraine’s VVER fleet for over a decade. Bulgaria’s Kozloduy received its first Westinghouse VVER-1000 reload in April 2024. Czech Republic’s Temelín completed first full reload deliveries in June 2025. For the more challenging VVER-440 design, Westinghouse loaded first fuel at Finland’s Loviisa in late 2024, and Hungary signed a $114 million contract for VVER-440 fuel supply starting 2028. As of late 2024, 21 of 34 VVER reactors globally had fuel supply contracts outside Russia.

This is a quiet revolution. It does not make headlines like sanctions or tariffs. But it is systematically dismantling Rosatom’s most durable competitive moat, one fuel assembly at a time.

The China Backdoor

Here is the part that keeps compliance lawyers up at night.

Chinese shipments of enriched uranium to the United States surged from essentially zero in 2020-2022 to 242,990 kg in December 2023 alone (the same month the House passed the ban), with cumulative 2024 imports reaching approximately 123,894 kg by October 2024 per S&P Global reporting. The timing is difficult to dismiss as coincidental. Simultaneously, China’s imports of Russian enriched uranium tripled in the first ten months of 2024, reaching $849 million.

The pattern is obvious to anyone with a pulse. China imports Russian enriched uranium, feeds it to its own reactors, and frees up Chinese-enriched material for export to Western utilities at non-Russian prices. The DOE, Commerce Department, and National Security Council are investigating. The US ban includes anti-circumvention provisions that make such swaps illegal if designed to evade the law, but proving intent requires access to confidential fuel contracts, and good luck with that.

For investors, this has two implications. First, it means the effective Russia ban is leakier than it appears on paper, which partially explains why SWU prices have not spiked even more violently. Second, if the US government closes the backdoor (which it is actively exploring), the supply squeeze on Western enrichment tightens further, and SWU prices go higher.

What Russia Could Do Next

Russia’s November 2024 export ban on enriched uranium shipments to the US was a retaliatory gesture, not a strategic escalation. The US had already banned Russian imports; Russia simply formalized the reciprocal. But the question investors should be asking is not what Russia has done. It is what Russia could do if it chose to weaponize its enrichment dominance more aggressively.

Scenario one: VVER fuel embargo on non-aligned states. Russia continues supplying VVER fuel to Hungary (a political ally) but restricts or prices up fuel deliveries to Bulgaria, Czech Republic, and Slovakia, all of which have been actively transitioning to Westinghouse alternatives. The goal would be to punish defection and slow the fuel diversification campaign. The practical impact would be modest (Westinghouse VVER fuel supply is already ramping), but the signaling effect could accelerate European enrichment and conversion procurement, tightening those markets further.

Scenario two: Kazakhstan transit disruption. Russia does not need to formally block Kazakh uranium shipments transiting its territory. It simply needs to create “administrative delays,” customs complications, rail scheduling “challenges,” or inspection requirements that slow shipments by weeks or months. Given that roughly 90% of Kazakh uranium exports historically travel through Russia, even a moderate increase in transit friction would tighten Western spot availability and force utilities to draw down inventories faster. This is the scenario that is hardest to attribute, hardest to sanction, and most likely to actually happen. It may, in fact, already be happening; Cameco has reported transit delays from Inkai that are difficult to fully explain by logistics alone.

Scenario three: dumping enriched uranium into the Chinese market at below-cost pricing, effectively subsidizing China’s nuclear program while simultaneously flooding the channel through which Chinese-origin SWU reaches Western utilities. If Russia sells SWU to China at $50/SWU while Western spot is $193, China can re-export at $150 and still make a handsome margin, while Western enrichers face price competition from material that is, at root, Russian-subsidized. This would compress Western SWU pricing without resolving the underlying capacity deficit, squeezing Centrus and Urenco margins while leaving the strategic vulnerability intact.

None of these scenarios is certain. All are within Russia’s operational capability. The point is not to predict which one materializes but to stress-test the thesis against a world where Russia does not passively accept the loss of its most valuable non-hydrocarbon export franchise. Rosatom generated substantial foreign currency revenue from enrichment services. Losing that revenue stream to Western sanctions and EU disengagement gives Moscow an incentive to disrupt the transition, not facilitate it.

The investment implication: the geopolitical risk premium in Western enrichment and conversion pricing is not a temporary dislocation. It is a structural feature of a market where the largest supplier has both the capability and the incentive to create friction. Pricing that premium correctly is the difference between a good trade and a great one.

An honest note: the enrichment thesis is the section of this primer where my own analytical priors most perfectly align with the bullish conclusion, and that is precisely when rigorous self-interrogation is most necessary. The counterarguments deserve explicit airing. Utility inventories may be larger and more resilient than public data suggests; utilities do not disclose fuel inventories in detail, and most US utilities hold 2 to 3 years of fuel supply, providing more runway than the headline deficit implies. Third-country processing of Russian material through intermediaries could prove more substantial and more persistent than currently assumed. And demand growth projections for new reactors may slip right alongside their construction timelines, reducing the future SWU requirement. The enrichment gap is real. Whether it is as severe as the pure arithmetic suggests depends on assumptions about inventory levels that are, by their nature, opaque.

5.4 Niger: A Supply Source That Disappeared Overnight

Niger was never the largest uranium producer, but it was a meaningful one, roughly 4-5% of global supply, and critically important to France, which sourced roughly a quarter of its EU uranium imports from Nigerien mines.

Then the military coup happened on July 26, 2023. And the uranium supply went to zero.

The timeline is bleak. Orano halted production at its SOMAÏR mine on October 31, 2024, citing inability to export and losses exceeding $200 million. By December 2024, Orano confirmed it had lost all operational control. The junta formally nationalized SOMAÏR on June 20, 2025. An unauthorized shipment of roughly 1,000 tonnes of yellowcake left the mine without Orano’s knowledge in November 2025 and got stranded at Niamey’s international airport, unable to transit through hostile neighboring countries, where it became a sitting target for an IS Sahel Province attack in January 2026.

France’s relationship with Niger is effectively severed. All French troops withdrew by December 2023. The embassy is closed. Niger has joined the Alliance of Sahel States, withdrawn from ECOWAS, and signed an MoU with Rosatom and a mining cooperation agreement with Uranium One. CNNC has been studying the Azelik mine restart.

The Imouraren deposit, one of the world’s largest undeveloped uranium resources at roughly 174,000-200,000 tonnes, had its operating permit revoked. Canada’s GoviEx lost its Madaouela permit.

For the uranium market, Niger is a write-off until the political situation changes, and there is no indication that it will anytime soon. The practical impact has been manageable because Niger’s production was already declining and utility contracts provided buffer. But it removes optionality from the supply side at exactly the wrong time.

5.5 Australia: The Sleeping Giant

Australia holds 28-35% of the world’s recoverable uranium resources. It is, by a wide margin, the most resource-rich uranium jurisdiction on Earth. BHP’s Olympic Dam alone contains over 2.5 million tonnes, the single largest deposit anywhere.

Australia produced roughly 4,500-5,100 tonnes in 2024. That is about 8-9% of global output. From a country with a third of the world’s reserves.

The gap between reserves and production is entirely political. The May 2025 federal election was a landslide for Labor, which won 94 seats (a record for any single party) and comprehensively defeated Opposition Leader Peter Dutton, who lost his own seat partly on the back of his deeply unpopular proposal to build seven nuclear power plants. Nuclear power remains banned in Australia under the EPBC Act. Western Australia bans new uranium mines under state Labor policy.

Boss Energy’s Honeymoon ISR mine is the bright spot, having restarted in April 2024 and declared commercial production in January 2025 with cash costs of just $23/lb. But Australia will not be the swing producer that saves the Western supply chain. The political constraints are structural, not cyclical, and no amount of uranium bull market will change them within the investment horizon that matters.

For investors, the Australian resource base is a floor under the long-term uranium thesis (the material exists and can eventually be accessed), but it is not a near-term supply solution.

5.6 Canada: The West’s Last Best Hope

If you are looking for a jurisdiction that combines world-class geology, stable rule of law, friendly relations with every major Western ally, an experienced regulatory apparatus, and active political support for uranium mining, your list has one entry: Saskatchewan, Canada.

The Athabasca Basin hosts uranium grades 100 times the world average. Cameco’s McArthur River/Key Lake mill set a world record in 2024, packaging 20.3 million pounds of U₃O₈ on a 100% basis, the highest annual production from any uranium mill in history. Cigar Lake continues producing at planned levels. Cameco holds roughly 230 million pounds in committed long-term contracts and operates at an average realized sale price north of $79/lb.

Two transformative new projects are clearing final permitting hurdles:

NexGen Energy’s Arrow deposit contains 256.7 million pounds in measured and indicated resources. At projected average production of 28.8 million pounds per year in years 1-5, it would represent roughly 20% of current global fuel supply from a single mine. The Canadian Nuclear Safety Commission approved the construction licence in early 2026. NexGen holds over $800 million in cash and targets construction start in summer 2026 with a four-year build timeline.

Denison Mines’ Wheeler River (Phoenix ISR) received its construction licence in late 2025/early 2026, the first new uranium mine approved in Canada in approximately 20 years. The feasibility study shows an after-tax IRR of 90%, a 10-month payback, and operating costs of $6.28/lb. First production is targeted for mid-2028.

Canada is where the “friend-shoring” thesis becomes investable. The country is expanding both production (Cameco, NexGen, Denison) and fuel cycle services (Port Hope conversion, Cameco/Brookfield’s Westinghouse JV for fuel fabrication), and it is doing so in a regulatory environment that is actively scaling to accommodate the pipeline. Saskatchewan proclaimed its first Nuclear Science Week in October 2025 and released a nuclear-powered electricity strategy. The province wants to be the nuclear fuel capital of the West, and it is making the investments to back that ambition.

5.7 The Friend-Shoring Architecture: $5.6 Billion and Counting

The Western response to the Russia/China fuel supply challenge has moved from rhetoric to funded programs at a speed that is, by the standards of nuclear policy, genuinely impressive.

The core alliance is the Sapporo 5 (US, Canada, France, Japan, UK), formed at the April 2023 G7 Ministers’ Meeting, which has mobilized $5.6 billion as of September 2024 to establish a Russia-free commercial fuel supply chain. These five nations collectively control roughly half of world uranium conversion and enrichment capacity.

The US alone has committed $2.72 billion in appropriated funding for domestic enrichment. In January 2026, DOE awarded three major task orders: Centrus Energy ($900 million for HALEU enrichment at Piketon, Ohio, first capacity by 2029), General Matter ($900 million for a novel-technology facility at the former Paducah plant in Kentucky, targeting 2034), and Orano Federal Services ($900 million for “Project IKE,” a $5 billion LEU enrichment facility in Oak Ridge, Tennessee, targeting initial production by 2031).

Uranium was added to the USGS Final 2025 List of Critical Minerals, reversing its removal by the Biden administration. A Section 232 national security investigation on uranium and processed critical minerals was initiated in April 2025.

On the enrichment side, Orano inaugurated a €1.7 billion expansion of Georges Besse II in October 2024, boosting capacity by 30% to 10 million SWU/year. Urenco is adding 1.8 million SWU across its four plants, with new cascades online in 2025 and full completion by 2027. Urenco’s order book reached €20.1 billion as of mid-2025, extending through 2040.

The numbers are real. The funding is appropriated. The construction is underway. But there is a transition gap of 5-8 years before new Western enrichment and mining capacity can materially replace Russian SWU. During that gap, the market remains tight, prices remain elevated, and the companies positioned on the right side of the geopolitical divide capture disproportionate value.

5.8 China: The 800-Pound Gorilla

We have danced around China in every section of this primer. It is time to confront it directly.

China operates approximately 59-60 reactors with roughly 62 GW of installed capacity. Another 28-33 reactors are under construction, the largest pipeline on Earth. The country has approved 10 or more new reactor units annually for four consecutive years. The 15th Five-Year Plan targets 110 GW by 2030 and 150 GW by 2035, requiring a sustained build rate of roughly 10 GW per year.

China’s uranium demand, currently about 13,130 tonnes per year, is projected to reach roughly 21,000 tonnes by 2030 and 26,800 tonnes by 2035. Domestic mining covers only 13-15% of needs. The rest comes from imports (Kazakhstan supplies 66%), overseas equity mines (Namibia’s Husab and Rössing), and open market purchases. Beijing holds a strategic uranium stockpile estimated at 132,000 tonnes, roughly triple the US stockpile and sufficient for approximately 10 years of current fleet operations.

On enrichment, CNNC operates an estimated 8-11 million SWU of capacity with two new plants started in 2023. The upper-bound projection from Ux Consulting suggests capacity could reach 23 million SWU by 2030, though most independent analysts (IPFM, various consultancies) project 13-17 million SWU as more realistic. Even the lower range would make China largely self-sufficient; the higher range could position it as a net exporter. The uncertainty in this figure is material for the Western enrichment thesis, since it determines whether Chinese overcapacity eventually spills into global markets.

CNNC’s fuel cycle ambitions extend to reprocessing, MOX fabrication, and fast reactors. The world’s first thorium molten-salt reactor (TMSR-LF1) achieved first criticality in October 2023 and confirmed in-reactor breeding of U-233 from thorium in November 2025.

China’s reactor export program has underperformed relative to ambitions (actual completed exports are limited to Pakistan’s seven units, and the UK’s Bradwell B Hualong One project was cancelled). But where China’s nuclear program most affects global markets is not through exports. It is through sheer domestic demand. Chinese procurement is redirecting uranium flows eastward, Chinese enrichment expansion is absorbing SWU that might otherwise reach Western markets, and Chinese intermediation of Russian enriched material raises the specter of sanctions circumvention.

The 132,000-tonne strategic stockpile is the number that should give every Western fuel procurement officer pause. It means China can weather any supply disruption for a decade. It means Chinese utilities can afford to be patient in contract negotiations, waiting for Western desperation to drive prices up before releasing material. It is the uranium equivalent of the Strategic Petroleum Reserve, except China actually built one and the West did not.

5.8.1 India: The Thorium Wild Card

India is the analytical gap in most Western nuclear primers, and the reason is uncomfortable: India’s nuclear program does not fit neatly into either the Western friend-shoring narrative or the Chinese/Russian parallel ecosystem. It sits in between, strategically non-aligned, operationally constrained, and sitting on a geological asset that could, in theory, rewrite the nuclear fuel equation entirely.

India operates 23 reactors with roughly 8 GW of installed capacity. Another 8 reactors are under construction, and the government targets 22 GW by 2031-32 and 100 GW by 2047. These are ambitious numbers against a track record of persistent delays and chronic underperformance relative to plan. India has been “about to scale up nuclear” for roughly three decades, and the gap between announcement and execution has been consistently wide. The Kudankulam VVER-1000 units (built with Russian cooperation) took over a decade each. Indigenous PHWR (Pressurized Heavy Water Reactor, essentially a CANDU derivative) construction has improved but still averages 7-10 years per unit.

The strategic wrinkle is that India’s nuclear program was built outside the Non-Proliferation Treaty. India tested nuclear weapons in 1974 and 1998, and was excluded from the Nuclear Suppliers Group until the landmark US-India Civil Nuclear Agreement of 2008, which granted India access to international nuclear commerce in exchange for placing its civilian (but not military) reactors under IAEA safeguards. This means India’s fuel supply chain is bifurcated: civilian reactors can import uranium (from Canada, Kazakhstan, Australia, and others), while military reactors rely on domestic production. The country’s domestic uranium mining output covers only a fraction of civilian needs, roughly 25-30% by most estimates, and India imports the rest.

For the global uranium market, India’s expansion adds roughly 2,000-4,000 tonnes of incremental annual demand by 2035 under realistic (not aspirational) build-out scenarios. This is not transformative at the scale of China’s 26,800-tonne projected demand, but it is additive at exactly the wrong time for a supply-constrained market.

Now, the thorium question.

India holds a significant share of the world’s known thorium resources, commonly cited at 25-30% in Indian government and older IAEA estimates, though more recent compilations using stricter reserve-category definitions place the figure closer to the low-to-mid teens. Total Indian resources are estimated at roughly 1.1 million tonnes in the Keral, Jharkhand, and Rajasthan coastal sands.

Thorium is three to four times more abundant than uranium in the Earth’s crust. It is not, however, fissile. You cannot put thorium in a reactor and get a chain reaction. Thorium-232 must first absorb a neutron and transmute into uranium-233, which is fissile. This means thorium is a “fertile” material (like U-238), not a fuel in itself. To use it, you need a reactor that can breed U-233 from thorium, which requires either a fast neutron spectrum or a specialized thermal breeder design.

India designed its nuclear program around a three-stage strategy to exploit this thorium endowment. Stage one: pressurized heavy water reactors fueled by natural uranium, producing plutonium in the spent fuel. Stage two: fast breeder reactors fueled by plutonium, breeding U-233 from a thorium blanket. Stage three: advanced heavy water reactors fueled by thorium-U-233, achieving a self-sustaining thorium fuel cycle.

The program is decades behind schedule. The Prototype Fast Breeder Reactor (PFBR) at Kalpakkam, the linchpin of Stage two, has been under construction since 2004 and has been “nearing completion” for roughly a decade. First criticality is now targeted for 2025-2026, roughly 15 years late. Stage three remains theoretical.

China’s TMSR-LF1, the world’s first thorium molten-salt reactor, achieved first criticality in October 2023 and reported successful in-reactor conversion of thorium to U-233 in November 2025, though the conversion ratio and precise experimental parameters have not been independently published.

This is significant not because it makes thorium commercially viable tomorrow (it does not; TMSR-LF1 is a 2 MW thermal test reactor, not a power plant), but because it demonstrates the physics at reactor scale for the first time, and China, not India, achieved it first despite India having the larger thorium endowment and the longer head start.

For investors, the thorium question resolves simply: thorium is a 2040s-and-beyond technology. No commercial thorium power reactor exists or is under construction anywhere on Earth. India’s three-stage program has been running for 70 years and has not reached Stage two. China’s molten-salt program is a decade from commercial demonstration. Thorium does not affect the uranium supply-demand balance within any investment horizon that matters for this cycle.

What does matter is India’s conventional uranium demand, which is growing, and India’s geopolitical positioning, which is uniquely flexible. India buys Russian reactors (Kudankulam) and Russian enrichment. It also buys Canadian uranium (Cameco), Australian uranium, and French reactor technology (EDF has proposed EPR units at Jaitapur, which would be the world’s largest nuclear complex if built). India signed the COP28 nuclear tripling pledge. It is a member of the Quad (with the US, Japan, and Australia) but maintains deep economic ties with Russia.

For the friend-shoring thesis, India is neither friend nor adversary. It is an independent demand centre that will source fuel from whoever offers the best deal, which means it competes with Western utilities for the same Kazakh and Canadian supply while simultaneously absorbing Russian fuel that might otherwise need to find alternative markets. India’s nuclear expansion, if it actually executes at anything close to planned pace, tightens the global uranium and enrichment balance without contributing to Western supply chain diversification. It is additive to demand and neutral-to-negative for Western supply security.

5.9 The Parallel Ecosystem Risk: China as Nuclear Exporter

The primer has so far treated China as a demand centre and a supply chain variable. But there is a scenario that could reshape the geopolitical map of nuclear energy within the next decade: China becoming a full-spectrum nuclear exporter to the Global South, effectively creating a parallel nuclear ecosystem that competes with the Western friend-shoring architecture.

Russia perfected this model over two decades. Rosatom does not just sell reactors. It sells an integrated package: sovereign financing (Russian state loans covering up to 90% of construction costs), lifetime fuel supply (from Russian enrichment and fabrication, creating permanent dependency), training and workforce development, waste take-back arrangements, and strategic political alignment.

This model has been extraordinarily effective. Rosatom has reactor construction contracts or proposals in Turkey, Egypt, Bangladesh, India, Hungary, and several other countries. The lock-in mechanism (VVER fuel dependency) creates relationships that outlast individual political cycles. China now possesses every component needed to replicate this model, and in some areas it is already ahead of Russia.

On reactor technology, the Hualong One is a proven 1,000+ MWe PWR with 41 units operational or under construction and 100% domestic supply chain localization achieved in 2024. The Linglong One SMR, on track for 2026 completion, provides a smaller-scale export option for countries whose grids cannot absorb a large reactor. The Hualong Two, under development, targets a four-year build time and 25% cost reduction.

On financing, China Development Bank and Export-Import Bank of China have collectively financed hundreds of billions in Belt and Road infrastructure. Nuclear project finance, at $5-10 billion per reactor, is well within their capacity, and Beijing can offer concessional terms that Western export credit agencies cannot match.

On the fuel cycle, CNNC operates 8-11 million SWU of enrichment capacity targeting 23 million SWU by 2030. China has conversion, fabrication, and a reprocessing capability that enables it to offer a cradle-to-grave fuel cycle service. The 132,000-tonne strategic uranium stockpile provides buffer for fuel supply commitments to export customers. On construction workforce, China has maintained continuous construction (37 reactors in 10 years, 32 more underway) and has the only nuclear construction workforce in the world that is simultaneously expanding in both scale and capability.

The target market is countries that want nuclear power but cannot afford Western prices, cannot access Western financing, and are not geopolitically aligned with the US-led friend-shoring alliance. The pipeline is more concrete than most Western analysts appreciate.

The UAE is the proof of concept. Barakah’s four APR-1400 units are operational, generating roughly 40 TWh per year, approximately 25% of the UAE’s total electricity. This is the first new nuclear power program successfully completed in the developing world in decades, and it happened on something close to schedule and budget. The UAE now has a trained nuclear workforce, an operational regulator (FANR), and a demonstrated ability to integrate large-scale nuclear into a grid that was previously 100% gas-fired. Barakah’s fuel demand, roughly 400-500 tonnes of enriched uranium per year across four units, is fully contracted through KEPCO’s supply chain, but those contracts will eventually roll, adding to the global term market.

Saudi Arabia is further behind but moving with purpose. The Kingdom selected France’s EDF for its first large reactor project in early 2025, a two-unit EPR at a Red Sea coastal site, with target operations in the mid-2030s. Riyadh’s nuclear ambitions are intertwined with its broader energy transition strategy (Vision 2030) and its desire to free up domestically consumed oil and gas for export. Saudi Arabia has also engaged with China, South Korea, and the US on reactor technology, playing the same multi-vendor game that the UAE played before selecting KEPCO. If Saudi builds 4-6 large reactors by 2040 (the upper end of publicly discussed plans), that adds roughly 4,000-6,000 tonnes of annual enriched uranium demand.

Egypt’s El Dabaa project, four VVER-1200 reactors being built by Rosatom with $25 billion in Russian financing, is the live example of the Rosatom lock-in model in action. First unit targeted for 2028-2030. Turkey’s Akkuyu (also Rosatom) is under construction with first unit commissioning targeted for 2025-2026. Bangladesh’s Rooppur (Rosatom again) is advancing. Each of these projects adds roughly 200-250 tonnes of enriched uranium demand per reactor per year, locked into Rosatom’s fuel supply chain for the reactor’s lifetime.

Vietnam revived its nuclear plans in late 2024, reversing a 2016 cancellation. Indonesia and the Philippines are both exploring SMRs as a pathway to nuclear that avoids the massive capital commitment of large reactors. Kenya, Ghana, and Nigeria have all engaged with the IAEA on nuclear power feasibility studies. Argentina has cooperated with China on small reactor technology and operates two CANDU-derivative reactors.

Across Southeast Asia, Sub-Saharan Africa, the Middle East, and parts of Latin America, there are 15 to 20 countries where nuclear deployment is plausible within the next decade, and another 10-15 where it is possible within two decades. The aggregate demand implication is not trivial. If even half these programs execute at their most conservative projections, they add 15,000-25,000 tonnes of incremental annual uranium demand by 2040, roughly 10-15% above current global consumption. The question is not whether the Global South will go nuclear. It is who builds the reactors, who supplies the fuel, and whether that fuel comes from the Western ecosystem or the Chinese/Russian one.

The investment implications cut in several directions, not all of them bearish. Uranium demand increases regardless of who builds the reactor. Each Hualong One requires approximately 200 tonnes of enriched uranium per year. Twenty export reactors add roughly 10 million lbs of annual U3O8 demand. This is additive to the supply-demand balance. Enrichment demand bifurcates rather than loosens.

Chinese-exported reactors will use Chinese-enriched fuel, which means China’s enrichment expansion (from 9 million to 23 million SWU) may be consumed by export obligations rather than freeing up SWU for Western spot markets. This is actually bullish for Western enrichment pricing, because it removes the scenario where Chinese overcapacity spills westward. The friend-shoring premium widens. If the nuclear world bifurcates into a Western ecosystem and a Chinese/Russian ecosystem, the geopolitical premium on Western-aligned assets increases. Canadian uranium, Urenco/Orano enrichment, and Westinghouse/Framatome fabrication become even more strategically critical.

But SMR competition intensifies. If the Linglong One is successfully exported at Chinese construction costs (roughly $2.5-3 billion per unit), Western SMR developers face a competitor that can deliver at a fraction of the cost. The Western SMR thesis depends on achieving NOAK economics that may never be competitive with Chinese FOAK economics. This is the dynamic that killed Western solar panel manufacturing, and it could repeat in nuclear. The counter-argument is that China’s actual nuclear export track record is modest. Completed exports are limited to Pakistan’s seven units.

The UK cancelled CGN’s participation in Bradwell B on security grounds. Many countries are reluctant to accept Chinese infrastructure dependency after Belt and Road debt controversies. Nuclear exports require decades of regulatory engagement, safety case development, and workforce training that China has not yet demonstrated outside Pakistan. But the capability exists. The financing exists. The political will exists. Investors who assume the Western friend-shoring architecture faces no competition from an integrated Chinese nuclear export ecosystem are making an assumption that deserves explicit stress-testing.

5.10 The Geopolitical Scorecard

Let us summarize the board.

Kazakhstan: 39% of global production, increasingly locked into Chinese offtake, transit-dependent on Russia, persistent production shortfalls from acid constraints. Verdict: the swing producer, but swinging toward Beijing.

Russia: 40-44% of enrichment, sanctioned by the US, losing EU market share, retaliating with export bans, but still structurally irreplaceable until the early 2030s. Verdict: diminishing but dangerous.

Niger: nationalized, zero production for Western markets, aligned with Russia and potentially China. Verdict: write-off.

Australia: world’s largest reserves, politically paralyzed, modest production. Verdict: long-term option value, not near-term supply.

Canada: world-class deposits, stable jurisdiction, active regulatory support, two transformative projects in final permitting. Verdict: the West’s most important uranium supplier for the next two decades.

China: massive demand growth, expanding domestic fuel cycle, strategic stockpile of 132,000 tonnes, quietly consolidating control over Kazakh supply. Verdict: the biggest variable in the entire nuclear fuel equation.

The investment implication is straightforward. Companies with assets in politically aligned, geographically safe jurisdictions (Canada, Australia, to a lesser extent the US) will trade at structural premiums to companies with assets in contested or unstable jurisdictions (Niger, Uzbekistan, Kazakhstan). The “jurisdiction premium” in uranium equities is not a temporary risk-off trade. It is a structural repricing that reflects the new geopolitical reality of nuclear fuel supply.

The friend-shoring of nuclear fuel is not a slogan. It is a $5.6 billion funded program backed by five governments with legislation, appropriations, and construction timelines. The companies positioned to supply that program, Cameco, NexGen, Denison, Urenco, Orano, Centrus, Westinghouse, are not just commodity producers. They are strategic assets in a geopolitical realignment that will take a decade to play out and will define the economics of nuclear power for a generation.

Part 6: The New Build Landscape

6.1 Why Can’t the West Build Things Anymore?

This is the section where we stop being polite.

Between 2015 and 2024, China built 37 nuclear reactors with an average construction time of 6.3 years from first concrete to grid connection. The fastest took 4.1 years. The cost per unit averaged roughly $2.7 billion.

For context, the global average construction time over the same period was 9.4 years per IAEA PRIS data, meaning China was building roughly 33% faster than the rest of the world combined. This gap is not a function of lower safety standards (China’s National Nuclear Safety Administration applies IAEA requirements and conducts peer reviews). It is a function of continuous construction, design standardization, and an intact supply chain, which is precisely the institutional coherence argument we will develop in detail below.

China currently has 32 reactors under construction, approved 10 more in April 2025, and has 102 reactors either operating, under construction, or approved with a combined capacity of 113 GW. CNNC is already developing the Hualong Two, a second-generation variant of its flagship design that will reduce build time from five years to four and cut costs by a quarter.

Now let us look at the West. You have seen the headlines in Part 2. Here are the numbers that explain them.

The United States completed Vogtle Units 3 and 4: two AP1000 reactors, 7 years late, $36.8 billion (2.6x the original budget). Westinghouse went bankrupt during construction. A parallel project at V.C. Summer in South Carolina was abandoned with $9 billion in sunk costs.

France’s Flamanville 3: one EPR, 17 years from first concrete to full power, cost ballooned from €3.3 billion to €23.7 billion including interest. France had not opened a new reactor since 2002.

Finland’s Olkiluoto 3: one EPR, 17 years, cost rose from €3.2 billion to roughly €8.5 billion.

Hinkley Point C in the UK: two EPRs, enabling works from 2017 with first nuclear concrete in December 2018, now targeting 2030 at earliest, cost estimate risen to £35 billion ($46 billion) from an original £18 billion.

The pattern is not subtle. China builds a reactor in 5 years for $2.7 billion. The West builds one in 12-17 years for $15-25 billion. This is not a technology gap. The physics is the same. The concrete is the same. The steel is the same. The fuel is the same. The gap is institutional, regulatory, and organizational. And until you understand why this gap exists, you cannot assess whether the “nuclear renaissance” in the West is real or just another round of expensive disappointment.

6.2 The Coordination Freeze: Why Nobody Moves First

One of the most incisive analyses of this dysfunction appeared in an essay by John Arden in American Affairs in late 2025, titled “The Sovereign Signal: Nuclear Energy as Strategic Infrastructure.” Arden’s argument, stripped to its essentials, is this:

Nuclear energy in the United States is not suffering from a technology problem, a capital problem, or even a political will problem. It is suffering from a classification problem. The US has been trying to deploy nuclear power using the institutional framework designed for market-based energy commodities (solar, wind, natural gas), when nuclear is actually sovereign infrastructure, more analogous to the Interstate Highway System, missile defense, or the Federal Reserve. The frame itself produces paralysis.

Arden models the dysfunction as a multiplayer coordination game. Every actor in the nuclear deployment chain has a rational reason to wait. Developers hesitate without guaranteed demand. Fuel suppliers delay without volume commitments. Investors require anchor customers. Utilities demand regulatory certainty. Regulators resist political exposure. Each player sees the risk of going first more clearly than the reward, and so nobody moves. The Nash equilibrium is stasis.

In game theory terms, this is a stag hunt. The payoff-dominant outcome is mutual cooperation (everyone builds, everyone benefits). But the risk-dominant strategy is defection (everyone waits, nobody loses money). Without a credible signal that shifts expectations, the rational strategy for every individual actor is to stand still, even though collective action would make everyone better off.

The historical evidence supports this framing. Arden cites Delmas and Heiman’s comparative study of the French and American nuclear programs, which demonstrated that the US combination of fragmented power, activist access to the courts, and weak bureaucratic cohesion destroyed credible government commitment. Construction lead times in the US lengthened from roughly three years to thirteen, while France held steady near six. Average project delays reached thirty-four months versus six in France. Interest during construction climbed to roughly 30% of capital expenditure. And 134 US reactors were cancelled while France built fifty-six, eventually reaching 75% nuclear electricity by 1990.

The critical insight: France did not have better technology. It did not have smarter engineers. It had institutional coherence. A unified technocratic bureaucracy (EDF/CEA) that could issue credible commitments, standardize designs, maintain a continuous construction workforce, and iterate on a single reactor platform. The US had none of these things after the dissolution of the Atomic Energy Commission in 1974 and the fracturing of authority across DOE, NRC, EPA, and the courts.

Arden’s prescription is a “sovereign signal,” a visible, irreversible, and costly commitment from the executive branch that realigns expectations across all actors simultaneously. Not a subsidy. Not a tax credit. Not a pilot program. A structural declaration that nuclear is permanent national infrastructure, backed by institutional reorganization, binding procurement, and financial commitments that make retreat more expensive than advance.

This is not political commentary. It is an analytical framework for understanding why the West’s nuclear construction record is what it is, and what would need to change for it to improve. The framework matters for investors because it helps you calibrate the probability that any given Western new-build announcement actually results in a completed reactor.

6.3 Why China and Korea Can Do It

If the West’s problem is institutional fragmentation, the corollary is that countries with institutional coherence can build nuclear plants efficiently. The evidence is overwhelming.

China has built a nuclear construction machine that is, by any objective measure, the most productive in history. The State Council approves reactor batches annually (10 units in April 2025 alone, representing $27.4 billion in investment). The same design (Hualong One, with 41+ units operational or under construction) is replicated across sites, creating a manufacturing learning curve. The workforce moves from project to project without dispersing. Supply chain localization reached 100% in 2024, with 114 sets of domestically produced nuclear equipment delivered, double the 2023 figure. Each Hualong One unit is expected to reach power within 50-60 months of first concrete.

China has also achieved something no other country has: first-mover deployments across multiple advanced reactor types. The Shidaowan high-temperature gas-cooled reactor (the world’s first commercial Gen IV reactor) began operations in 2021. The Linglong One SMR (ACP100) is on track for completion by 2026. The CFR-600 sodium-cooled fast reactor began commissioning in 2023. And the world’s first thorium molten-salt reactor (TMSR-LF1) achieved first criticality in October 2023 and confirmed in-reactor breeding of U-233 from thorium in November 2025.

This is not because Chinese engineers are inherently superior. It is because China treats nuclear as sovereign infrastructure, exactly as Arden prescribes. The State Council provides the signal. CNNC and CGN execute. The National Nuclear Safety Administration licenses. The supply chain aligns because everyone knows the program is permanent. The stag hunt is resolved because the sovereign has committed.

South Korea tells the same story, slightly differently. Korea’s APR-1400 design (a 1,400 MWe PWR) was delivered at the four-unit Barakah plant in the UAE roughly on schedule, a remarkable achievement by the standards of any recent large reactor project. All four units were connected by 2024. The initial contract value was approximately $24 billion for 5,600 MW (~$4,300/kW at financial close), though subsequent reporting suggests the all-in cost including financing and scope changes may have reached $30-32 billion (~$5,400-5,700/kW). Even at the higher figure, this is roughly half the per-kW cost of Vogtle or Hinkley, and remains the strongest evidence that serial construction of a standardised design can achieve competitive nuclear economics.

Korea achieved this through design standardization (one design, replicated), a retained construction workforce (the same teams that built Korea’s domestic fleet), and a government that treated nuclear exports as a strategic national priority (the Korea Electric Power Corporation, KEPCO, is state-owned). After a brief flirtation with nuclear phase-out under President Moon Jae-in, Korea reversed course and now plans to increase nuclear’s share to 30%+ of the electricity mix by 2036.

The lesson from China and Korea is not “authoritarianism builds reactors.” It is “institutional coherence builds reactors.” France proved this in the 1970s-80s under a democratic system. The common thread is not political regime type. It is the presence of a sovereign commitment that resolves the coordination game and allows the supply chain to mobilize with confidence.

6.4 The Existing Fleet: Where the Near-Term Growth Actually Is

While everyone is talking about new build and SMRs, the near-term growth in nuclear capacity is coming from a much less glamorous source: life extensions and restarts of existing reactors.

The US nuclear fleet of 94 reactors is aging, with a median reactor age around 42 years. But the NRC has been granting 20-year license renewals that extend operating lifetimes from 40 to 60 years, and subsequent license renewals (SLRs) that take them to 80 years. As of 2025, several reactors have already received SLRs, and the industry expects the majority of the fleet to apply. At 80 years, a reactor that entered service in 1985 would operate until 2065. The fuel demand from these extensions is already baked into procurement plans but is rarely counted as “growth” because no new reactor is being built.

Restarts are the more dramatic story. Three Mile Island Unit 1 (the reactor that was not involved in the 1979 accident) is being restarted under a 20-year PPA with Microsoft for its data center needs. Palisades in Michigan, which was shut down in 2022, received a $1.5 billion DOE loan guarantee for restart and plans to return to service. Holtec, which owns Palisades, also plans to build an SMR-300 alongside the restarted large reactor.

Japan is restarting its idled fleet. After shutting down all 54 reactors post-Fukushima, Japan has gradually restarted 12 as of early 2026, with more in the pipeline. Each restart adds roughly 1 GW of demand for fuel services.

Life extensions and restarts represent the lowest-risk, fastest, and cheapest path to more nuclear capacity. You already have the site, the grid connection, the cooling water, the workforce, and (in many cases) the operating license. The fuel demand from these units is incremental to the market and largely uncovered by existing contracts, which is why utilities are being pushed into the term market.

6.5 SMRs: The Promise, the Hype, and the Reality

There is no topic in nuclear energy where the gap between narrative and reality is wider than small modular reactors. SMRs are Silicon Valley’s favorite nuclear talking point. They are the subject of breathless announcements from tech CEOs. They are invoked in every data center power strategy presentation. And as of March 2026, not a single SMR is producing commercial electricity anywhere in the Western world.

Let us be precise about what exists, what is under construction, and what is still a PowerPoint slide.

What actually exists (operating SMRs, globally):

Russia’s Akademik Lomonosov, a floating nuclear power plant with two 35 MWe reactors, has been operating in Pevek, Russia since 2020. China’s Shidaowan HTGR (a 210 MWe high-temperature gas-cooled reactor, arguably the first commercial Gen IV reactor) has been generating electricity since 2021. China’s Linglong One (ACP100, a 125 MWe light water SMR) is under construction in Hainan and on track for completion by 2026.

That is it. Russia and China have operating or near-operational SMRs. The West has zero.

What is under construction or in advanced permitting in the West:

Ontario Power Generation’s Darlington New Nuclear Project in Canada, using GE Hitachi’s BWRX-300 (300 MWe light water SMR). Construction began in May 2025. First unit expected late 2029. Four units total. This will be the Western world’s first operating SMR if it stays on schedule. The US DOE awarded a $400 million grant to TVA to accelerate deployment of the BWRX-300 at its Clinch River site in Tennessee, but that project is further behind, with commissioning targeted for 2032.

Kairos Power’s Hermes test reactor in Oak Ridge, Tennessee: a 35 MWt (non-power) demonstration of its fluoride salt-cooled high-temperature reactor. Construction permit granted December 2023. Scheduled for operation in 2026. This is a test reactor, not a commercial power plant. Google has partnered with Kairos to deploy 500 MW of commercial capacity, but those units are for the 2030s.

TerraPower’s Natrium (345 MWe sodium-cooled fast reactor with molten salt energy storage) in Kemmerer, Wyoming: the NRC issued a construction permit on March 4, 2026, the first commercial non-light-water reactor construction permit in over 40 years. Non-nuclear construction began in mid-2024, with the sodium test facility’s 167-foot steel frame already rising. Total project cost is approximately $4 billion, with DOE providing up to $2 billion via ARDP and Bill Gates contributing roughly $1 billion. Operations targeted for 2030 to 2031. The critical dependency is HALEU fuel at 19.75% enrichment, supply that barely exists.

X-energy’s Xe-100 (80 MWe helium-cooled pebble bed) at Dow Chemical’s Seadrift site in Texas: NRC accepted the construction permit application for review with an 18-month timeline. The critical milestone: TRISO-X received the first-ever NRC Part 70 HALEU fuel fabrication license in February 2026, a significant de-risking event for the advanced fuel supply chain. X-energy has raised over $1.1 billion. But the Dow partnership for four Xe-100 units (320 MWe) remains subject to Dow’s final capital commitment decision, expected around 2028.

NuScale remains the only SMR developer with NRC-approved designs: its original 50 MWe module received a full Design Certification (Part 52 rulemaking) in 2023, while the uprated 77 MWe version received a Standard Design Approval (SDA) in June 2025, a technically distinct but functionally equivalent NRC authorization.

But its first planned deployment (the Idaho CFPP project with UAMPS) was cancelled in November 2023 after costs nearly doubled. NuScale has since partnered with ENTRA1 Energy and TVA on a 6 GW deployment program announced September 2025, but no concrete construction timeline has been disclosed. NuScale has also signed deals in Ghana, Romania, and other countries, all in early planning stages.

What is still a press release:

Larry Ellison’s announcement of a 1 GW data center campus powered by three SMRs. Amazon’s nuclear data center plans. Meta’s 4 GW nuclear RFI. These are expressions of interest, not funded construction projects. They signal directional intent, which matters for market sentiment, but they do not represent committed capital or secured permits. Most of these are a decade away at best.

The honest assessment:

SMRs will contribute to the nuclear mix in the 2030s. The BWRX-300 at Darlington has the best chance of being the Western world’s first operating commercial SMR, probably in 2029-2030. The advanced designs (Kairos, TerraPower, X-energy) will follow in the early-to-mid 2030s. NuScale’s commercial timeline remains unclear.

But SMRs will not meaningfully contribute to the nuclear capacity or uranium demand picture within this decade. The IEA’s base case has the first SMRs coming online around 2030, with nuclear’s contribution to data center electricity supply growing “notably after 2030.” Goldman Sachs projects SMRs will be “generally cheaper to build” than traditional reactors once designs are “fully standardized,” but acknowledges this is contingent on achieving manufacturing scale that does not yet exist.

The hype cycle is running 5-10 years ahead of the engineering reality. This is not a criticism of the technology or the companies. It is a statement about timelines. If you are making investment decisions based on SMR deployment happening before 2030, you are making a bet against every nuclear construction timeline in history.

A technical wrinkle that SMR critics have raised and that proponents have not fully resolved: spent fuel characteristics.

A 2022 study published in the Proceedings of the National Academy of Sciences (Krall, Macfarlane, and Ewing) found that SMR designs may produce waste that is “more voluminous and chemically/physically reactive” per unit of electricity than conventional large reactors. The core physics reason is neutron leakage: smaller reactors have a higher surface-area-to-volume ratio, which means a greater fraction of neutrons escape the fuel and irradiate the surrounding structural steel rather than causing fission.

The result is more activated structural waste per MWh generated, and a requirement for higher initial fuel enrichment to compensate for the lost neutrons.

Argonne National Laboratory published a 2023 rebuttal finding SMR waste volumes “roughly comparable” to large reactors when measured on consistent assumptions. The debate is unresolved and matters for two reasons: first, waste volume directly affects disposal costs and public acceptance; second, if SMR waste profiles require different handling or disposal pathways, they could face additional regulatory burden that affects deployment timelines and economics. This does not invalidate the SMR thesis, but it introduces a variable that most SMR investment analyses do not account for.

6.6 The Construction Problem Is a Workforce Problem

One of the least discussed but most critical constraints on Western nuclear new build is the construction workforce. Or rather, the absence of one.

The United States has not sustained continuous nuclear construction since the 1980s. The skilled trades (nuclear-grade welders, pipe fitters, boilermakers, concrete specialists, I&C technicians) required for reactor construction have either retired, moved to other industries, or were never trained in the first place. When Vogtle tried to build two AP1000s after a 30-year construction gap, it discovered that the workforce simply did not exist. Workers had to be recruited from across the country, trained on site-specific requirements, and managed through a project management structure that had not built a reactor in a generation.

This is the problem Arden’s framework addresses at the institutional level. Without a “sovereign signal” that makes clear the nuclear construction program is permanent (not a one-off project that might be cancelled after the next election), the workforce has no reason to invest in nuclear-specific skills. A welder who trains for nuclear qualification (which takes years and is significantly more demanding than commercial welding certification) needs confidence that there will be a pipeline of projects to work on. Without that confidence, the rational decision is to stay in oil and gas, or LNG construction, or shipbuilding, where the project pipeline is visible.

China solved this by maintaining continuous construction (37 reactors in 10 years, with 32 more underway). France solved it in the 1970s-80s by building 58 reactors in roughly two decades with the same workforce moving from site to site. Korea solved it by maintaining both domestic and export construction programs.

The West has not solved it. And no amount of SMR announcements or federal funding changes that until someone builds a pipeline of projects large enough to justify workforce investment. This is why Arden calls for a “Nuclear Deployment Command” and an “Atomic Corps” to recruit, train, and deploy the construction workforce as permanent national infrastructure. Without it, each new project reinvents the wheel, retrains the workforce from scratch, and suffers the cost and schedule overruns that come with institutional amnesia.

6.7 What This Means for Investors

The new build landscape creates a tiered investment framework:

Tier 1: Existing fleet operators (Constellation, EDF, KEPCO, Exelon legacy units) benefit from life extensions and restarts with relatively low execution risk. These companies capture the nuclear thesis through fuel procurement, not construction. Constellation’s stock has been one of the best performers in US utilities precisely because the market recognizes the value of its 21 GW nuclear fleet in a power-constrained world.

Tier 2: Near-term fuel demand (Cameco, Kazatomprom, Urenco, Orano, Centrus) benefits regardless of whether new build succeeds or fails. The existing fleet plus life extensions plus restarts generate enough demand growth to support the bull case in uranium and enrichment. New build is upside, not the base case.

Tier 3: Construction-leveraged plays (Westinghouse via Cameco/Brookfield, Framatome via Orano, BWXT, Fluor, Bechtel, GE Vernova) benefit if and when the construction pipeline materializes. These are higher-risk, higher-reward, and dependent on the political and institutional variables that Arden’s framework highlights.

Tier 4: SMR developers (NuScale, Oklo, X-energy, Kairos, TerraPower) are venture-stage bets on technology commercialization with 5-10 year timelines. They carry execution risk, regulatory risk, and financing risk. The potential payoffs are enormous if they succeed, but the base rates for first-of-a-kind nuclear construction projects are not encouraging.

The key question for Tiers 3 and 4 is whether the “sovereign signal” Arden describes will be issued, and whether it will be credible enough to break the coordination freeze. The May 2025 executive orders from the Trump administration moved in this direction (streamlining NRC licensing, directing federal site deployments, announcing $80 billion in new reactor funding). But as Arden notes, these created “visibility without irreversibility.” The orders reintroduced momentum but did not resolve the stag hunt. Until the signal is structural (binding procurement, organizational restructuring, multi-decade commitments), the smart money remains cautious on Western new build and concentrated in Tiers 1 and 2.

The nuclear renaissance is real. But it is unevenly distributed. In China and Korea, it is an industrial program delivering results. In the West, it is a policy aspiration contending with institutional inertia. The investment strategy should reflect that asymmetry.

Part 7: Regulation, Safety, and Public Perception

7.1 The Regulator That Ate the Industry

Between 1954 and 1978, the United States authorized the construction of 133 nuclear reactors. Since 1978, two have entered commercial operation. Both were at Vogtle. Both were late. Both were over budget. Together they cost $36.8 billion.

That is the single most important statistic in this entire primer. Not because the technology changed (it improved). Not because the physics changed (it did not). But because the regulatory and institutional environment that governs nuclear construction in the United States was transformed in the late 1970s in ways that made building reactors functionally impossible for a generation.

Understanding how this happened, and whether it is changing, is essential for any investor attempting to price the “nuclear renaissance.”

7.2 A Brief History of How We Got Here

The Atomic Energy Commission (AEC), established in 1946, was the original US nuclear regulator, promoter, and fuel cycle custodian, all in one institution. This unified mandate was coherent but carried an obvious conflict of interest: the same body that promoted nuclear energy was responsible for ensuring its safety. After the environmental movement of the early 1970s and growing public concern about nuclear safety, Congress dissolved the AEC in 1974 and split its functions between the Department of Energy (promotion, fuel cycle, weapons) and the newly created Nuclear Regulatory Commission (safety and licensing).

This split was well-intentioned. It was also the moment nuclear deployment in the United States began to die.

Not because the NRC was incompetent. It was and is staffed by serious, technically capable professionals. But because the NRC was created in an era of adversarial legalism, a term coined by political scientist Robert Kagan to describe the distinctly American regulatory style in which policy disputes are resolved not through bureaucratic consensus but through litigation, public hearings, and procedural challenges.

The practical effect on nuclear licensing was devastating. Environmental statutes, particularly NEPA (the National Environmental Policy Act), introduced layers of mandatory review that could be challenged in court at every stage. Standing rules were relaxed, allowing virtually any interested party to intervene in licensing proceedings. Citizen suits multiplied. Each reactor became a bespoke legal battle. And because every challenge added months or years to the licensing timeline, and because interest during construction (IDC) compounds on a multi-billion-dollar project at rates that can reach 30% of total capital expenditure, each delay made the reactor more expensive, which made it harder to justify, which invited more legal challenges, which created more delays.

This was not a conspiracy against nuclear. It was a systemic outcome of a regulatory philosophy that prioritized procedural rights over institutional coherence. The same legal and administrative framework that gummed up nuclear reactors also slowed highway construction, pipeline permitting, and transmission line development. Nuclear just suffered the most because its capital costs are the highest and its construction timelines are the longest. In a system where any intervenor can add three years to a licensing review, a $5 billion project becomes a $10 billion project through accumulated IDC alone.

Delmas and Heiman’s comparative study, which we referenced in Part 6, quantified the result. US construction lead times tripled from roughly three years in the 1960s to thirteen years by the 1980s. France’s held steady near six. The US cancelled 134 reactors. France built 56. Same decade. Same technology. Different institutional architecture.

7.3 The NRC Today: Reform Is Real, but the Stag Hunt Persists

The NRC is in the middle of the most significant reform effort in its 50-year history, driven by three overlapping mandates: the ADVANCE Act of 2024 (bipartisan legislation that explicitly added “enabling the deployment of nuclear power for the benefit of society” to the NRC’s mission statement), NEIMA (the 2019 Nuclear Energy Innovation and Modernization Act, which required modernized fee structures and performance metrics), and Executive Order 14300 (President Trump’s May 2025 order titled “Ordering the Reform of the Nuclear Regulatory Commission”).

EO 14300, signed May 23, 2025 and published in the Federal Register on May 29, is the most aggressive federal intervention in nuclear regulatory policy in decades. It mandates binding licensing deadlines: no more than 18 months for new reactor applications and no more than 12 months for license renewals. It directs NRC to reorganize staff in consultation with DOGE, expand capacity in new reactor licensing, establish a 20-member team to draft new regulations, and reduce the Advisory Committee on Reactor Safeguards to its statutory minimum. Proposed rules must be issued within nine months; final rules within eighteen months.

The NRC has responded with measurable improvements.In 2024, it completed review of the Hermes 2 construction permit (a Kairos Power test reactor) in approximately 16 months from application submission to permit issuance (July 2023 to November 2024), making it one of the fastest NRC construction permit reviews in the agency’s history. NRC Chair Hanson described the review as taking “less than 18 months.”

It renewed 13 reactor licenses (each for an additional 20 years, preserving 12,000 MW on the grid). It authorized the first-ever return-to-operation of a decommissioned reactor (Palisades). It established an 18-month-or-shorter review schedule for X-energy’s Xe-100 construction permit at Dow Chemical’s Seadrift site. It extended AP1000 design certifications from 15-20 years to 40 years. And it realized up to a 50% reduction in resources and schedules for license renewals, with 35% schedule reductions and 40% cost reductions for new reactor reviews.

The Part 53 rulemaking, a technology-inclusive licensing framework designed for advanced (non-light-water) reactors, has been in development for five years and must be finalized by law before the end of 2027. If implemented well, it could reduce licensing costs per applicant by $53-68 million and provide a regulatory pathway that does not require dozens of exemptions from LWR-specific rules. But as of early 2026, most developers (including TerraPower) are filing under the existing Part 50 framework because Part 53 is not yet available.

Here is the honest assessment: the NRC reforms are real, directionally positive, and more aggressive than anything attempted in the past two decades. But they do not, by themselves, resolve the coordination freeze that Arden describes. Faster licensing is necessary but not sufficient. A utility that can get a permit in 18 months instead of 60 months still needs to solve the financing problem, the construction workforce problem, the fuel supply problem, and the political sustainability problem. The NRC is one node in a multi-player coordination game, and reforming one node does not unfreeze the game.

This is the nuance that most nuclear commentary misses. Pro-nuclear advocates celebrate NRC reform as if it were the binding constraint, and once removed, reactors will spring from the ground like mushrooms after rain. Anti-nuclear advocates warn that reforms are “rubber-stamping” safety, as if the only thing standing between America and a nuclear accident is a lengthy review timeline. Both are wrong. The binding constraint is systemic, not regulatory. And the safety case for nuclear is independent of how long the NRC takes to issue a permit.

7.4 The Safety Record: Numbers That Nobody Wants to Hear

This is the section that kills arguments at dinner parties.

Nuclear power, measured by deaths per unit of energy produced, is the safest large-scale energy source humanity has ever deployed. It is safer than solar (rooftop installation falls). It is safer than wind (turbine maintenance fatalities). It is dramatically, overwhelmingly safer than coal, oil, and natural gas (air pollution, mining accidents, pipeline explosions).

The numbers below draw on two separate studies compiled by Our World in Data: Markandya and Wilkinson (2007) for fossil fuels (covering air pollution and accidents, European data) and Sovacool et al. (2016) for renewables (covering accidents only, global data). The methodologies differ, the geographic scope differs, and the fossil fuel figures are likely conservative for global averages (Our World in Data’s global estimates for coal range from 93 to 224 deaths per TWh).

But even acknowledging these caveats, the relative ordering is obvious and the magnitude of the gap between nuclear and fossil fuels is not in dispute:

Nuclear and solar are statistically tied for the safest energy sources on Earth. Coal kills roughly 800 times more people per unit of energy than nuclear. Natural gas kills roughly 90 times more. These are not abstract statistical constructs. They represent real human beings who die from particulate matter inhalation, mercury poisoning, black lung disease, and methane explosions, versus the vanishingly small number who have died from radiation exposure in the entire history of commercial nuclear power.

The total confirmed death toll from radiation exposure across all commercial nuclear accidents in 70 years is fewer than 50, all from Chernobyl. The WHO’s most aggressive long-term estimate projects approximately 4,000 additional cancer deaths over the lifetimes of the most exposed Chernobyl populations. Three Mile Island caused zero radiation deaths and zero measurable increase in cancer rates in the surrounding population (confirmed by multiple epidemiological studies over 40 years). Fukushima caused zero deaths from radiation (the deaths associated with the disaster, approximately 2,300, were caused by the evacuation itself, not by radiation exposure, a fact that has prompted significant debate about whether the evacuation was proportionate to the actual radiological risk).

These numbers are not disputed by the scientific community. They are disputed by the public imagination, which is a very different thing.

7.5 Radiation: The Thing Everyone Gets Wrong

The public perception of radiation risk is, to put it charitably, untethered from empirical reality. This is not because people are stupid. It is because radiation is invisible, the word “nuclear” is culturally fused with “weapon,” and the human brain processes risk through narrative, not statistics.

Here is what the science actually says.

Radiation is everywhere. You are being irradiated right now. The average American receives approximately 6.2 millisieverts (mSv) of radiation per year from natural and medical sources. Of this, roughly 3 mSv comes from natural background radiation (cosmic rays, radon gas in your basement, potassium-40 in your bananas and your own body) and roughly 3 mSv comes from medical procedures (CT scans, X-rays, nuclear medicine).

A chest X-ray delivers about 0.02 mSv. A CT scan delivers about 7 mSv. A transatlantic flight delivers about 0.08 mSv (cosmic ray exposure increases with altitude). Living within 50 miles of a nuclear power plant adds approximately 0.0001 mSv per year to your exposure (the NRC’s figure is 0.01 millirem, not millisieverts), which is roughly 70,000 times less than a single CT scan.

The “banana equivalent dose” is a popular science communication tool: a banana contains enough potassium-40 to deliver about 0.1 microsieverts per banana eaten. You would need to eat roughly 10 million bananas in rapid succession to receive a lethal dose. This is obviously absurd, but it makes the point: radiation is not binary (safe/lethal). It is a continuum, and the doses associated with normal nuclear power plant operations are negligible compared to the radiation you voluntarily accept from medical imaging, air travel, and living in a stone house.

The health effects of low-dose radiation (below roughly 100 mSv) are the subject of genuine scientific debate.

The “linear no-threshold” (LNT) model, which underpins most radiation protection regulations worldwide, assumes that any amount of radiation carries some cancer risk, proportional to dose, with no safe threshold. LNT remains the consensus regulatory standard, endorsed by the ICRP, UNSCEAR, the IAEA, the US NRC, and the National Academies (BEIR VII, 2006). The NRC rejected a formal petition to abandon LNT as recently as 2021. It is not a fringe position. It is the institutional default, and for good reason: it provides a conservative, precautionary framework for occupational and public dose limits that has served the industry well for decades.

That said, there is genuine scientific debate at the low-dose frontier. A comprehensive 2024 review in Health Physics found that roughly 62% of animal and cellular studies support non-linear or hormetic effects at low doses, while human epidemiological data broadly supports LNT (or at least cannot rule it out, given the statistical difficulty of detecting tiny excess cancer rates against large natural background rates). The “hormesis” hypothesis, which suggests that very low doses may actually stimulate protective biological responses, remains controversial and has not achieved regulatory acceptance.

For investors, the practical consequence of LNT matters more than the biological debate. LNT drives the evacuation thresholds, cleanup standards, and decommissioning costs that affect the economics of every nuclear facility. Fukushima’s 154,000-person evacuation, which caused approximately 2,300 deaths from the dislocation itself while radiation caused zero confirmed fatalities, was ordered under LNT-derived dose limits. If regulators eventually adopt a threshold model (a Trump-era executive order has directed the NRC to reconsider LNT), it would reduce decommissioning costs, narrow evacuation zones, and improve public acceptance. That regulatory shift is speculative but worth monitoring.

What is not controversial: the health risk from low-dose radiation exposure near nuclear power plants is unmeasurably small. It is lost in the statistical noise of background radiation variation, dietary differences, and other environmental factors. The NRC’s regulatory limit for public dose from nuclear plant operations is 1 mSv/year, and actual doses to members of the public are typically less than 1% of that limit.

7.5.1 Water: The Operational Constraint Nobody Talks About

Here is an operational risk that nuclear’s advocates consistently underweight and that climate change is making progressively more material: nuclear power plants need enormous quantities of water for cooling, and they are vulnerable to the same heatwaves and droughts that drive peak electricity demand.

Nuclear is the most water-intensive major electricity generation technology. A typical 1 GW PWR with once-through cooling withdraws approximately 25,000-60,000 gallons per MWh (though most of this is returned to the source at elevated temperature). Even with cooling towers (which consume rather than just withdraw water), consumption runs roughly 500-800 gallons per MWh, compared to roughly 200-300 for a combined cycle gas plant and near-zero for solar PV and wind.

This is not a theoretical concern.

Peer-reviewed research by Linnerud, Mideksa, and Eskeland (2011, The Energy Journal) found that a 1 degree Celsius increase in ambient water temperature reduces nuclear power output by approximately 0.5%, a finding corroborated by Ibrahim et al. (2014) at 0.44% per degree. Losses can exceed 2% per degree during extreme heat events. A separate study in Nature Energy (Ahmad, 2021) documented increased outage frequency and duration at nuclear plants during heatwaves, reinforcing the operational concern from a different angle.

France, the poster child for nuclear power, has repeatedly been forced to curtail reactor output during summer heatwaves because the rivers that supply cooling water (the Rhone, the Garonne, the Loire) get too warm to absorb the waste heat without exceeding environmental discharge limits. In the scorching summer of 2022, EDF curtailed output from multiple plants while simultaneously France was experiencing record electricity demand from air conditioning. Germany and Switzerland have experienced similar constraints.

For the existing US fleet, most plants are sited on large rivers, lakes, or coastlines with adequate water supply, and the operational impact has been modest. But several plants in the southeastern US (where water availability is declining and summer temperatures are rising) may face increasing curtailment risk over their remaining operating lifetimes, which could extend to the 2060s and 2070s under 80-year license renewals.

For new build, water availability is a binding site selection constraint. This is one reason SMR developers are exploring air-cooled designs (which eliminate the water dependency entirely but at a thermal efficiency penalty) and why several proposed SMR sites are at existing nuclear facilities or former coal plants that already have cooling water infrastructure.

The investment implication: water risk does not change the fundamental thesis for nuclear, but it does affect which specific plants will maintain their 92%+ capacity factors over the coming decades of climate warming, and which may experience declining performance. Investors in nuclear fleet operators should assess the cooling water vulnerability of the specific plants in the portfolio, not just the fleet-average statistics.

7.6 Three Mile Island, Chernobyl, Fukushima: What Actually Happened

These three accidents form the trinity of public nuclear fear. Each is invoked as evidence that nuclear power is inherently dangerous. Each, properly understood, tells a story that is more nuanced, more instructive, and frankly more reassuring than the popular narrative suggests.

Three Mile Island (1979): The Accident That Proved the Containment Worked

A sequence of equipment failures and operator errors caused a partial meltdown of the reactor core at TMI Unit 2 in Pennsylvania. About half the core melted. And then the containment building, the massive reinforced concrete and steel structure that surrounds every Western reactor, did its job. It contained the vast majority of the radioactive material.

The average dose to the roughly 2 million people living within 50 miles was approximately 0.08 mSv, comparable to a chest X-ray. The maximum individual offsite dose was estimated at approximately 1 mSv, roughly equivalent to a few chest X-rays and still well below any threshold for measurable health effects.

No injuries. No deaths. No measurable health effects in 40+ years of follow-up studies.

TMI was a design validation event disguised as a disaster. The containment performed as engineered. The public was not harmed. But the psychological and political damage was catastrophic, because The China Syndrome had been released twelve days earlier, and the media narrative framed TMI as a narrowly averted Armageddon rather than a demonstration of defense-in-depth working exactly as intended.

Chernobyl (1986): The Reactor That Should Never Have Existed

We covered this in Part 2, but it bears repeating in the safety context. Chernobyl was an RBMK reactor with a positive void coefficient, no containment building, and operators who disabled safety systems during a poorly designed test. The explosion and graphite fire released massive amounts of radioactive material. Two plant workers died in the explosion. 28 emergency responders died of acute radiation syndrome within months. The long-term WHO estimate is approximately 4,000 additional cancer deaths among the most exposed populations.

Chernobyl was a Soviet-specific failure that could not occur in any Western reactor design. No PWR, BWR, CANDU, or Gen III+ reactor has the physics that produced Chernobyl. The comparison is technically illiterate, and yet it dominates public discourse about nuclear safety to this day.

Fukushima (2011): When the Backup to the Backup Failed

The Fukushima accident was caused not by a reactor failure but by a beyond-design-basis natural disaster: a magnitude 9.0 earthquake (the fourth largest ever recorded) followed by a 14-meter tsunami that overwhelmed the seawall and flooded the backup diesel generators. Without power for cooling, three reactor cores melted.

The radiological release was significant but the direct health impact was minimal. The UN Scientific Committee on the Effects of Atomic Radiation (UNSCEAR) concluded that “no discernible increased incidence of radiation-related health effects are expected among exposed members of the public or their descendants.” Zero deaths from radiation exposure. The World Health Organization’s assessment found that the risk of additional cancers above natural rates was too small to be statistically detectable.

The human cost of Fukushima was real, but it came overwhelmingly from the evacuation, not the radiation. Approximately 2,300 people, mostly elderly, died from the stress, dislocation, and interrupted medical care caused by the forced evacuation of 154,000 people. A growing body of research, including a major 2024 study in the journal Risk Analysis, suggests that the evacuation of low-contamination zones was not justified by the radiological risk and may have caused more harm than the radiation would have.

This is not to minimize the accident. Fukushima was a massive industrial disaster that contaminated land, disrupted communities, and will take decades to fully decommission. But the radiation health effects, the thing the public fears most, were negligible. The lesson of Fukushima is not “nuclear is dangerous.” It is “tsunamis are dangerous, and you should not put your backup generators in the basement of a plant on a coastline in the most seismically active region on Earth.”

7.7 The Waste Question: A Political Problem in a Technical Costume

We covered this in Part 3, but given that waste is the anti-nuclear movement’s favorite argument, it is worth addressing frontally in the safety context.

The total volume of high-level spent nuclear fuel generated by the entire US commercial nuclear fleet in 60+ years of operation fits on a single football field stacked about 10 yards high. It sits in dry cask storage at reactor sites, in concrete and steel containers designed to withstand earthquakes, tornadoes, floods, and aircraft impacts. It has never harmed a single member of the public.

We have already covered the coal comparison in detail (Part 1 and Part 3), but one additional fact belongs here because it is genuinely surprising: coal plants release more radioactivity into the environment per kWh than nuclear plants do. The naturally occurring thorium and uranium in coal concentrate in the fly ash, and because that ash is dispersed into the atmosphere and surrounding land (even with modern capture systems recovering over 99% of particulates), the total radiological dose delivered to the public from coal combustion exceeds the dose from a properly operating nuclear plant. Spent nuclear fuel is far more radioactive per unit mass, but it is contained in engineered storage rather than dispersed.

The volume comparison is even more damning: 300,000 tons of toxic ash per year from a single coal plant, disposed of in unlined ponds, versus 90,000 metric tons total from the entire US nuclear fleet across 60 years of operation, sitting safely in engineered dry cask storage.

Finland is building Onkalo, the world’s first permanent deep geological repository for spent nuclear fuel, carved into 1.8-billion-year-old granite bedrock on the island of Olkiluoto. It is designed to isolate waste for 100,000 years. Sweden has approved a similar repository at Forsmark. France’s Cigéo repository in clay formations is scheduled for operations in the 2030s.

The United States has none of these things, not because the technology does not exist, but because Yucca Mountain became a political football and no politician has had the courage (or the institutional mandate) to resolve it. This is Arden’s classification problem in miniature. Waste custody is a sovereign function. No market mechanism, no private company, and no state government can solve it. Only the federal government can declare custodianship, fund a repository, and implement it against local opposition. The refusal to do so is not a technical failure. It is a failure of sovereign posture.

For investors, the waste question is a negligible risk factor. The existing fleet has operated for decades with on-site storage, which is safe, licensed, and unconstrained by capacity limits within the investment horizon that matters. The political unresolvability of permanent disposal is a talking point for anti-nuclear advocates, not a binding constraint on nuclear operations or investment returns.

7.7.1 The Price-Anderson Act: The Liability Architecture Nobody Discusses

One element of the nuclear regulatory framework that belongs in any honest assessment, and that most nuclear primers omit because it complicates the narrative, is the Price-Anderson Nuclear Industries Indemnity Act.

Price-Anderson, originally enacted in 1957 and most recently extended through 2045 by the ADVANCE Act, establishes a two-tier insurance system for nuclear accidents. Each reactor operator must carry roughly $500 million in primary insurance. If damages exceed that, all US reactor operators contribute to a secondary pool totaling roughly $15 billion (currently about $166 million per reactor, payable in annual installments). Claims exceeding this combined pool would require Congressional action. The total coverage, after the 5% surcharge and post-2023 inflation adjustments to retrospective premiums, now exceeds $16 billion (the Congressional Research Service estimates approximately $16.6 billion; the NRC’s current published figure states “more than $16 billion”).

This is, functionally, a liability cap. And it is a subsidy. If a nuclear accident caused damages exceeding $15.5 billion (which a severe accident plausibly could, depending on the modeling assumptions), the excess falls on the federal government and, ultimately, taxpayers. No private insurer would cover the tail risk of a nuclear accident at any commercially viable premium, which is why Price-Anderson exists: without it, no utility would build or operate a reactor.

This matters for investors not because it changes the operational thesis (Price-Anderson has been in place for nearly 70 years and is deeply embedded in the industry’s financial structure) but because anti-nuclear critics correctly identify it as a hidden subsidy, and because any honest LCOE comparison between nuclear and alternatives should, in principle, include the actuarial value of the liability cap. It also matters for the new-build thesis: Price-Anderson’s extension through 2045 was a prerequisite for any utility considering a construction commitment.

7.8 Public Perception: Why the Vibes Are Finally Shifting

For decades, nuclear energy had the worst brand in the energy business. Mushroom clouds. Geiger counters. Hazmat suits. Homer Simpson. The cultural firmware was set to “fear,” and no amount of safety data could overwrite it.

Something has changed. Not gradually. Abruptly.

A 2024 Pew Research Center survey found that 56% of Americans now favor expanding nuclear power, up from 43% in 2020. Gallup found similar trends. Among younger demographics (18-29), support has grown even faster, driven partly by climate concern and partly by the tech industry’s enthusiastic embrace of nuclear for data center power.

The shift is not driven by better messaging or public education campaigns. It is driven by perceived purpose. When Microsoft signs a 20-year PPA to restart Three Mile Island for its data center, that signals to the public that nuclear energy is not an artifact of the Cold War but a technology of the future. When Google partners with Kairos Power to deploy advanced reactors, that reframes nuclear from “scary legacy technology” to “cutting-edge clean energy.” When Sam Altman invests in Oklo, nuclear gets Silicon Valley credibility.

Arden’s framework predicts this. He argues that public legitimacy is not primarily a function of information but of perceived purpose. “When a technology is seen as essential to national survival, as spaceflight was during the Cold War, or the internet during the post-9/11 cybersecurity build-out, public ambivalence generally yields to alignment.” Nuclear’s wavering public legitimacy reflects its institutional ambiguity. Until it is declared, visibly and authoritatively, as a permanent pillar of national infrastructure, the public will mirror the system’s uncertainty.

The tech industry’s embrace of nuclear is doing something that decades of industry PR could not: it is giving nuclear a narrative of inevitability. When the richest, most innovative companies on Earth choose nuclear, it signals that nuclear has been chosen. That signal matters more than any safety statistic.

But let us not overstate the shift. Opposition remains concentrated and politically powerful in specific geographies (Nevada, parts of New England, Germany, Austria, Australia). NIMBY dynamics around waste storage and new reactor siting are real and will not disappear because Bill Gates tweets about Natrium. The permitting and litigation infrastructure that Arden describes as the legacy of 1970s adversarial legalism is still fully operational, and any new reactor project will face legal challenges from well-funded environmental organizations.

The difference now is that the political wind has shifted from headwind to tailwind. Bipartisan support for nuclear in the US Congress (the ADVANCE Act passed with overwhelming bipartisan margins). State-level support in key jurisdictions (Tennessee, Wyoming, Michigan, Ontario, Saskatchewan). International support (30+ countries pledging to triple capacity at COP28). And the perception that opposing nuclear is no longer the “green” position but the “obstructionist” position, at least among the centrist and center-left voters who increasingly see nuclear as essential to decarbonization.

For investors, the public perception shift matters because it reduces the political risk premium on nuclear investments. A reactor project that would have been politically toxic in 2015 is politically viable in 2025. That does not mean it will be built on time or on budget. But it means the political kill switch, which destroyed nuclear construction in the 1980s and could always be pulled again, is less likely to be pulled in the current environment than at any point in the past 40 years.

7.9 The Regulatory Edge: What TSCS Readers Need to Know That Others Don’t

Most nuclear primers cover regulation as a boring procedural topic and move on. Here is what actually matters for positioning:

The 18-month mandate is a regime change, not a tweak. EO 14300’s binding 18-month review deadline for new reactor applications, if actually enforced, compresses a process that historically took 5-7 years. This does not just save time. It changes the financial math. Interest during construction on a $5 billion project at 7% is roughly $350 million per year. Cutting 3-4 years off the pre-construction licensing phase saves $1-1.4 billion in IDC alone. That is the difference between a project that pencils and one that does not.

Part 53 is the sleeper catalyst. The technology-inclusive licensing framework, which must be finalized by end-2027, would allow advanced reactor developers to file applications without dozens of exemptions from LWR-specific rules. The NRC estimates savings of $53-68 million per applicant. But the real value is in standardization: once a design is licensed under Part 53, subsequent deployments of the same design at different sites require only site-specific reviews, not full design re-reviews. This is how France built 58 reactors in 20 years, one design reviewed once and then replicated.

The ADVANCE Act made deployment part of the NRC’s statutory mission. Before 2024, the NRC’s mandate was purely safety-focused. The ADVANCE Act explicitly added language requiring the NRC to “enable the deployment of nuclear power for the benefit of society.” This is not symbolic. It changes the legal framework within which NRC staff make discretionary decisions. A regulator whose mission is solely to prevent harm optimizes for delay (you cannot be blamed for approving too slowly, but you can be blamed for approving too quickly). A regulator whose mission includes enabling deployment has institutional permission to balance speed and safety. The cultural shift this implies within the NRC bureaucracy is significant and will compound over years.

The reduced hourly rate for advanced reactor applicants is a direct subsidy of regulatory risk. The ADVANCE Act required NRC to establish a reduced fee for advanced reactor pre-applicants and applicants: $148/hour versus the standard $318/hour (a 53% reduction). For a multi-year licensing review that might consume thousands of staff-hours, this reduces the regulatory cost burden by millions. More importantly, it signals that the federal government wants these applications to happen.

The 40-year design certification extension is a quiet game-changer. NRC extended AP1000 design certifications from 15-20 years to 40 years. This means Westinghouse does not need to periodically re-certify its design, which removes a recurring regulatory risk that hung over the AP1000 program and, by extension, over every utility that might consider building one. For Korean APR-1400 or other standardized designs seeking US certification, this precedent matters enormously.

The sum of these reforms does not guarantee a nuclear construction renaissance. As Arden argues, regulatory reform is necessary but not sufficient in a multiplayer coordination game. But it removes one of the most frequently cited excuses for inaction, and it shifts the calculus for utilities, developers, and investors who were previously deterred by regulatory uncertainty.

The question is no longer “will the NRC let you build a reactor?” The question is “can you finance it, staff it, fuel it, and sustain the political will to complete it?” Those are harder questions. But they are the right questions, and the fact that we have moved past the regulatory blocker is itself a form of progress.

Part 8: The Investable Universe

8.1 The Map Before the Territory

You have read seven parts of this primer. You understand the physics. You understand the fuel cycle. You understand the bottlenecks, the geopolitics, the regulatory terrain, and the construction dynamics. Now we deploy all of it.

This section is not a stock-picking exercise. TSCS does not give price targets, and anyone who gives you a price target on a uranium miner is either lying about their confidence interval or does not understand what a confidence interval is. What we are going to do is something more valuable: give you a decision architecture. A framework for evaluating every nuclear investment on the same axes, so you can assess where each one sits in the risk-return spectrum and, critically, why it trades where it does.

The nuclear investment landscape is not one market. It is five distinct markets with different drivers, different risk profiles, and different relationships to the underlying thesis. Most investors treat it as a monolith (”I’m long nuclear”) and wonder why their portfolio does not perform the way the thesis suggests it should. The answer is almost always that they are in the wrong part of the chain.

Here are the five markets:

1. Physical uranium (commodity exposure, no operational risk)

2. Uranium producers (mining execution + commodity leverage)

3. Fuel cycle services (conversion, enrichment, fabrication, structural bottleneck exposure)

4. Nuclear utilities and fleet operators (power price leverage, political stability)

5. New build and SMR developers (venture-stage, binary outcomes, long duration)

Each demands a different analytical framework, a different time horizon, and a different tolerance for execution risk. We will walk through each one and identify the names that matter, the drivers that move them, and the traps that catch the unwary.

8.2 Physical Uranium: The Cleanest Exposure

If you believe the uranium price is going higher but do not want to underwrite mine execution risk, geological risk, permitting risk, or management risk, physical uranium vehicles are the purest play.

Sprott Physical Uranium Trust (TSX: U.U / OTCQX: SRUUF)

SPUT is the largest physical uranium holding vehicle in the world. It is effectively a closed-end trust that buys and holds physical uranium, storing it in licensed facilities. When SPUT trades at a premium to its net asset value, it issues new units and uses the proceeds to buy more uranium. When it trades at a discount, it stops buying.

This mechanism is important because SPUT is not just a passive vehicle. It is an active demand participant in the spot market. Its buying removes physical pounds from the market, tightening the spot balance. In weeks when SPUT is actively purchasing, spot volumes and prices both tend to move higher. In weeks when it is sidelined (trading at a discount), the spot market can soften. Understanding SPUT’s premium/discount dynamics is essential for anyone trading uranium equities short-term.

The bull case: SPUT gives you 1:1 exposure to the uranium spot price with zero execution risk. No mine needs to work. No permit needs to be granted. No CEO needs to execute.

The bear case: SPUT is tied to the spot price, which we established in Part 4 is the less important of the two uranium markets. The term price, where the real business happens, can diverge from spot for extended periods. In 2025, spot traded $63-83 while the term price climbed to $86 and continued rising. SPUT captures the spot, not the term. Also, SPUT charges a management fee (roughly 0.35% annually) and you are paying for physical storage of a commodity that generates no yield.

Yellow Cake plc (LSE: YCA)

The UK-listed equivalent, Yellow Cake holds physical uranium purchased under a long-term supply agreement with Kazatomprom. Smaller than SPUT but with a structural advantage: its Kazatomprom agreement provides a mechanism for purchasing uranium at spot prices regardless of market conditions, which gives it a reliable acquisition channel.

The framework: Physical uranium is a macro trade. You are betting on the direction of the commodity, full stop. It is the lowest-risk way to express the nuclear fuel thesis, and for investors who want nuclear exposure without the idiosyncratic risks of individual companies, it is the right starting point.

8.3 Uranium Producers: The Heart of the Sector

This is where most nuclear investors live, and where the analytical complexity is highest. Not all miners are created equal, and the axes of differentiation are: jurisdiction, stage of development, cost curve position, contract book, and integration with downstream fuel cycle services.

Tier 1: The Majors

Cameco Corporation (TSX: CCO / NYSE: CCJ)

Cameco is the Western world’s vertically integrated nuclear fuel company. This distinction matters enormously and is the primary reason Cameco trades at a premium multiple that occasionally frustrates value-oriented investors.

The business has three segments: uranium mining (McArthur River/Key Lake, Cigar Lake, Inkai JV), fuel services (Port Hope conversion, fuel fabrication), and a 49% equity stake in Westinghouse Electric Company (acquired in late 2023 with Brookfield for an enterprise value of $8.2 billion).

McArthur River/Key Lake has persistently underproduced relative to its 25 million pound licensed capacity due to ground freezing challenges and development delays in new mining areas, a recurring operational constraint that tempers the bullish production narrative even as Cigar Lake and Inkai continue delivering. Port Hope conversion has been running at record output levels and climbing.

The company holds a contract book of roughly 220 million committed pounds, rolling legacy pricing into progressively higher term rates as older contracts expire. Cameco’s 49% share of Westinghouse is contributing growing EBITDA at a mid-to-high single digit compound annual growth rate, with additional upside from Westinghouse’s role in the Czech Republic’s Dukovany new-build project.

Then there is the $80 billion US government partnership announced in October 2025, which requires careful framing because it is the single most consequential development for Cameco’s Westinghouse stake and also the one most susceptible to over-reading. The binding term sheet was still being finalized at announcement. This is a framework agreement with aspirational scale, not a funded construction order. Under the framework, at least $80 billion of new reactors are to be constructed using Westinghouse AP1000 and AP300 technology, with the Department of Commerce arranging financing and facilitating permitting. The government receives a participation interest entitling it to 20% of cash distributions exceeding $17.5 billion, though this vests only upon a final investment decision and execution of definitive agreements.

It represents the most significant federal commitment to commercial nuclear construction since the Atomic Energy Act, and it is, in Arden’s framework, the closest thing to a “sovereign signal” for nuclear construction that the US has issued in decades.

Why Cameco matters for the thesis: it captures value at three points in the fuel chain (mining, conversion, and downstream services/fabrication via Westinghouse). When the uranium price rises, Cameco’s mining segment benefits. When conversion prices rise (they are up 6x since 2021), Port Hope benefits. When new reactors are ordered, Westinghouse benefits from reactor services, fuel fabrication, and new-build construction management. This triple exposure is why Cameco trades at roughly 17x forward EBITDA versus 5-8x for pure-play miners. The premium is structural, not speculative.

The risk: Cameco is a Canadian company selling into US markets, and the Trump administration’s tariffs on Canadian goods could add approximately 10% to costs. Most US customers are locked into long-term contracts through 2026, cushioning the near-term impact, but the tariff overhang creates volatility. Also, Cameco’s Inkai JV production in Kazakhstan remains vulnerable to the acid supply and transit risks we discussed in Part 5.

Kazatomprom (LSE: KAP / OTC: NATKY)

The world’s largest uranium producer by volume, the lowest-cost major producer on the planet, and the single name that most perfectly illustrates why jurisdiction matters more than geology in this cycle. We covered the operational details extensively in Part 5, so here is the investment summary.

Kazatomprom has latent capacity well above current output. It has been persistently underproducing relative to its subsoil use agreement ceilings, constrained by sulfuric acid supply and infrastructure bottlenecks rather than by geology or capital. In theory, it is the closest thing to a “central bank of uranium,” with the ability to move the global supply dial by simply choosing to produce more.

In practice, it is a state-owned entity whose output is increasingly locked into Chinese offtake agreements (north of half its production now flows east), whose transit logistics depend on Russian rail and port infrastructure, whose JV equity stakes are being transferred from Russian to Chinese ownership, and whose geopolitical positioning makes it a fundamentally different risk than a Canadian or Australian miner. The jurisdiction discount Kazatomprom trades at relative to Cameco is not a market inefficiency. It is the market correctly pricing the fact that this company’s uranium is being redirected away from the West, one long-term contract at a time. Western investors who own Kazatomprom are, indirectly, financing a supply chain realignment that works against Western energy security. That may not bother you. But you should know what you own.

Tier 2: The Developers (High Conviction, Pre-Revenue)

NexGen Energy (TSX: NXE / NYSE: NXE)

NexGen’s Arrow deposit in the Athabasca Basin is, by any measure, the most important undeveloped uranium project in the world. 256.7 million pounds of measured and indicated resource. Projected average production of 28.8 million pounds per year in years 1-5, which would represent roughly 20% of current global fuel supply from a single mine. The Canadian Nuclear Safety Commission approved the construction licence in early 2026, the first new uranium mine approved in Canada in approximately 20 years.

NexGen is well-capitalized and targeting a four-year build timeline from construction start to first production. If it executes, Arrow will be a generational asset, one of the lowest-cost, highest-grade uranium mines on Earth, located in the safest jurisdiction on the planet.

The analytical question for NexGen is always the same: at what uranium price and discount rate does Arrow’s NPV justify the market cap? The feasibility study economics are compelling, but NexGen is pre-revenue, and the stock trades on NAV expectations driven by three variables: the assumed long-term uranium price, the discount rate, and the execution probability. The spread between “Arrow on time at $100+ uranium” and “Arrow two years late at $70 uranium” is the entire investment debate in one sentence.

Denison Mines (TSX: DML / NYSE: DNN)

Denison’s Wheeler River project (Phoenix ISR deposit) is the second most exciting new development in the Athabasca Basin. The feasibility study shows an after-tax IRR of 90%, a 10-month payback, and operating costs of $6.28/lb, among the lowest globally. The CNSC approved the construction licence in late 2025/early 2026, and first production is targeted for mid-2028.

Denison is smaller than NexGen in scale but arguably lower-risk in execution because ISR mining is simpler and cheaper than the underground mining NexGen will employ at Arrow. For investors who want Athabasca Basin exposure with faster time-to-production, Denison is the more conservative bet.

Tier 3: The Restarters and Mid-Caps

Paladin Energy (ASX: PDN / OTC: PALAF): Restarted the Langer Heinrich mine in Namibia. Commercial production achieved in early 2024. Paladin offers near-term production growth in a jurisdiction that is politically stable but geographically distant from Western fuel cycle infrastructure.

Boss Energy (ASX: BOE): Restarted Honeymoon ISR mine in South Australia at a cash cost of roughly $23/lb, making it the lowest-cost ISR producer in the Western world. Small scale but high margins, with dual-jurisdiction exposure through a 30% interest in the Alta Mesa ISR project in Texas. Boss is the proof of concept that ISR uranium mining can be restarted quickly and cheaply when price signals justify it.

Uranium Energy Corp (NYSE: UEC): The largest US-based uranium mining company. Owns ISR projects in Texas and Wyoming plus the high-grade Roughrider project in the Athabasca Basin (acquired from Rio Tinto). UEC benefits disproportionately from US-centric policies (Section 232 investigations, critical minerals designation, domestic procurement preferences) but has yet to demonstrate sustained production at meaningful volumes.

enCore Energy (TSX: EU / NYSE: EU): Another US ISR producer with multiple projects in Texas. Small scale, early-stage production ramp. Similar policy-driven thesis to UEC but smaller.

The framework for producers: The key variables are jurisdiction (Canada and Australia command premiums over Kazakhstan and Africa), stage (producing assets are valued differently than development assets), cost curve position (all-in sustaining cost relative to spot/term price), and contract book (contracted pounds at locked prices vs. uncontracted production exposed to spot). The best producers combine friendly jurisdictions with near-term production, low costs, and a growing contract book. The worst combine hostile jurisdictions with long development timelines, high costs, and no contractual visibility.

8.4 Fuel Cycle Services: The Structural Bottleneck Plays

This is where TSCS’s earlier analysis (Parts 3 and 5) converts most directly into investment positioning. The thesis is simple: the tightest bottleneck in the fuel chain captures the most pricing power, and the companies positioned at those bottlenecks trade at structural premiums.

Enrichment

Centrus Energy (NYSE: LEU)

Centrus is the only publicly traded Western enrichment company. Full stop. Urenco is privately held by three governments. Orano is state-owned. Rosatom is sanctioned. CNNC is Chinese and inaccessible. If you want equity exposure to the enrichment bottleneck, your list has one name.

Centrus operates two businesses: a legacy SWU and uranium brokering business (buying enrichment services from Urenco, Orano, and others and reselling to US utilities) and a nascent HALEU production program at its American Centrifuge Plant in Piketon, Ohio. The DOE awarded Centrus $900 million in January 2026 to expand HALEU enrichment, with first new capacity targeted for 2029.

The bull case: SWU spot prices have quadrupled from $47.50 to $193 in five years, and the structural deficit in Western enrichment capacity will take until the early 2030s to resolve. Centrus’s brokering margins expand as the SWU price rises. The HALEU contract provides a multi-year revenue floor. And if the US government decides to rebuild commercial-scale domestic LEU enrichment (which the $2.7 billion DOE program is designed to enable), Centrus’s centrifuge technology and NRC-licensed facility give it a first-mover advantage.

The bear case requires understanding the business model’s central vulnerability with precision. Centrus’s LEU segment functions primarily as a middleman: it purchases SWU from enrichers (historically dominated by Russia’s TENEX at favorable pricing) and resells to US utilities at a spread. That spread, applied across millions of SWU annually, generates the bulk of segment earnings. The problem is structural: the Prohibiting Russian Uranium Imports Act phases out Russian LEU imports by January 2028. If Centrus cannot replace Russian volumes with Urenco or Orano supply at comparable input prices (and both are themselves capacity-constrained, with order books extending deep into the 2030s), the brokerage margin compresses or the backlog becomes undeliverable. The backlog is an asset only if the supply side holds. If it does not, those contracts become liabilities.

The HALEU program is a demonstration project, not a commercial enrichment plant, and the transition from brokerage to domestic production is a multi-billion-dollar, multi-year execution challenge that has not begun at commercial scale. The stock prices in the destination; the risk lives in the journey. If SWU prices soften (unlikely near-term but possible if the China backdoor persists and Western capacity additions accelerate faster than expected), the entire margin structure compresses simultaneously.

Conversion

Conversion does not have a clean publicly traded equity proxy. Cameco’s Port Hope facility is the closest thing (it is Cameco’s fuel services segment, reported separately). Conversion prices have risen 6x since 2021 and the Western market operates at 91%+ utilization with no buffer. If you own Cameco, you own conversion exposure. There is no standalone play.

Fuel Fabrication

Westinghouse (via Cameco’s 49% stake) and Framatome (via Orano, state-owned) dominate Western fuel fabrication. BWXT Technologies (NYSE: BWXT) is the US leader in naval nuclear reactor fuel and components, with growing exposure to commercial nuclear via the eVinci microreactor program and DOE contracts. BWXT trades at a premium (roughly 30x+ forward earnings) that reflects its defense-industrial positioning as much as its nuclear exposure.

8.5 Nuclear Utilities: The Power Price Play

If you believe that nuclear power plants are the most valuable generation assets in a power-constrained world, the obvious investment is the companies that own and operate them.

Constellation Energy (NASDAQ: CEG)

Constellation is the single largest nuclear fleet operator in the United States with approximately 22 GW of nuclear capacity across 12 stations (21 reactors). It was created from the 2022 separation of Exelon’s competitive generation business and is, effectively, the US nuclear pure-play.

Constellation has been one of the best-performing utility stocks in recent memory, driven by the recognition that nuclear baseload power is increasingly scarce and increasingly valuable in a world of AI-driven electricity demand growth. Microsoft’s 20-year PPA to restart Three Mile Island Unit 1 (a Constellation station) was the landmark deal that signaled to the market what TSCS readers already know: tech companies will pay whatever it takes for reliable, carbon-free baseload.

The bull case for Constellation is essentially the power price thesis. In deregulated markets (PJM, ERCOT, NYISO, ISO-NE), nuclear plants earn the market clearing price for electricity, which has been rising as demand growth outpaces supply additions. Nuclear’s 92%+ capacity factor and zero marginal fuel cost means nearly every MWh sold is profit above fixed costs. As gas prices rise, as renewables hit curtailment walls, and as data center demand pushes load growth to levels not seen in decades, the scarcity premium for reliable baseload increases.

There is a regulatory risk here that most nuclear commentary ignores and that TSCS readers need on their radar. In November 2024, FERC rejected Talen Energy’s proposed behind-the-meter arrangement with Amazon at the Susquehanna nuclear plant, ruling that directly serving a 960 MW data center from a grid-connected nuclear plant without going through the transmission system raised reliability concerns and could shift costs to other ratepayers. This was not a one-off ruling. FERC subsequently opened a Notice of Inquiry (Docket No. RM25-3-000) on large load co-location at generating facilities, signaling that the commission is evaluating whether to impose broader constraints on how nuclear plants can contract directly with hyperscalers.

If FERC determines that behind-the-meter co-location at grid-connected generators creates systemic reliability risk or cost-shifting, it could force all hyperscaler nuclear arrangements into front-of-meter retail structures. Talen’s restructured Amazon deal already uses this workaround (a retail power contract that bypasses FERC wholesale jurisdiction). But front-of-meter retail deals face state-level regulatory scrutiny, including public utility commission review, interconnection studies, and potential ratepayer advocacy challenges. The path from signed PPA to delivered electrons gets longer and less certain. The proceeding is ongoing. No final rule exists. But investors treating the utility re-contracting wave as a regulatory-free trade are underpricing a live risk.

The bear case requires quantifying what the current valuation already embeds. Constellation’s roughly 17 GW fleet at 92% capacity factor generates roughly 137 TWh per year. However, not all of that capacity is available for premium re-contracting. A significant portion serves regulated load-serving obligations, existing bilateral contracts at legacy rates, and merchant sales into PJM and other markets that cannot simply be redirected to hyperscalers. The re-contractable capacity, the megawatts that can realistically migrate from merchant wholesale to premium bilateral PPAs as existing commitments roll off, is probably closer to 8 to 12 GW over the next 3 to 7 years, depending on contract expiration schedules and the FERC co-location outcome.

The re-contracting math is straightforward. Of Constellation’s roughly 22 GW fleet, only a portion is re-contractable at any given time (the rest serves regulated load-serving obligations, existing bilateral deals, and merchant commitments that cannot simply be redirected). As existing commitments roll off over the next several years, each GW that migrates from legacy merchant pricing to premium PPA rates represents hundreds of millions in incremental annual revenue. The analytical question is not whether re-contracting happens (it will) but how much of the eventual re-contracting economics the market has already priced in. The risk is that investors are paying for the destination rather than the journey, and any recession-driven demand destruction, unexpected gas price collapse, or adverse FERC ruling that slows co-location could compress the multiple significantly against an elevated starting valuation.

8.5.1 The Economics That Separate Upstream from Downstream

Before we continue through the utility landscape, two sector-level dynamics need to be made explicit, because they determine how every name in this section should be valued relative to the producers and enrichers in the sections above.

It is also worth making explicit what most nuclear commentary leaves implicit: nuclear utility PPAs are, structurally, a short on long-term natural gas prices. Here is why.

US wholesale electricity prices are set by the merit order, and in most hours the marginal generator setting the clearing price is a gas-fired combined cycle plant. At $2.50/MMBtu gas (roughly where Henry Hub traded through much of 2024 to 2025), the marginal cost of gas generation is approximately $25 to $30/MWh. This means wholesale power clears at $30 to $45/MWh, and a nuclear PPA reportedly structured at $80 to $100/MWh represents a $45 to $60 premium the hyperscaler pays for carbon-free, firm, 24/7 attributes. That premium looks reasonable if you believe carbon regulation is tightening and grid reliability has value, and it looks expensive if you think gas stays cheap indefinitely.

But the sensitivity runs both directions. At $5 to $6/MMBtu gas (which the US experienced in 2022 and which most long-term forecasts project as plausible under LNG export growth and declining domestic production), marginal gas generation costs rise to $50 to $65/MWh, and wholesale power follows. At that gas price, a $90/MWh nuclear PPA is barely above market clearing, meaning the hyperscaler locked in near-market pricing with the bonus of carbon-free attributes and baseload reliability. At $7+/MMBtu gas (a tail scenario, but one that occurred during the 2022 European energy crisis), wholesale power could reach $80 to $100/MWh, and the nuclear PPAs become below-market contracts.

Given the 20-year time horizon of these PPAs, the bet that $90/MWh for firm, carbon-free power will look cheap before the contracts expire is probably right. But it is a bet on gas fundamentals over two decades, and your position in Constellation should reflect your conviction on that variable.

To make the fuel cost economics concrete, and to illustrate why Cameco and Constellation are fundamentally different bets despite both being “nuclear,” consider the annual fuel cost for a single 1 GW PWR at 92% capacity factor (generating roughly 8,060 GWh per year). The reactor consumes roughly 500,000 lbs of U3O8 per year (purchased under long-term contracts, not at spot). At a blended contract price of $70/lb, that is roughly $35 million. It requires roughly 18,000 kgU of conversion services at roughly $65/kgU, adding roughly $1.2 million. It needs approximately 120,000 SWU of enrichment at a blended contract rate of roughly $140/SWU (well below spot), totaling roughly $17 million. Fuel fabrication adds another $10 to $15 million per annual reload equivalent. Total fuel cost: roughly $65 to $70 million per year, or about $8 to $9/MWh. Uranium feedstock is roughly 50% of fuel cost. Enrichment is roughly 25%. Fabrication and conversion split the rest.

The critical insight: a $20/lb swing in uranium spot price (say, from $80 to $100) changes this reactor’s annual fuel bill by roughly $10 million, against total generation revenue of $400 to $800 million. That is noise for the utility. It is everything for the miner. A reader who buys Cameco and Constellation as a “nuclear basket” is actually making two very different bets: Cameco is a leveraged commodity play on uranium prices and fuel cycle bottlenecks. Constellation is a contracted infrastructure play on power prices and PPA terms. Their correlation to the “nuclear theme” is high. Their correlation to the same financial drivers is low. Structure your portfolio accordingly.

8.5.2 The Rest of the Utility Landscape

The nuclear utility story is not a US-only story, and investors who treat it as one are leaving exposure on the table.

Korea Electric Power Corporation (NYSE: KEP / KRX: 015760)

KEPCO is the state-owned Korean utility that operates Korea’s nuclear fleet and, through its subsidiaries, builds and exports reactors (APR-1400). KEPCO has been a value trap for years (burdened by regulated electricity prices that did not cover costs), but the reversal of Korea’s nuclear phase-out policy and the potential for APR-1400 exports (following the successful Barakah project in the UAE) have improved the outlook.

For investors who want exposure to the nuclear construction thesis in a jurisdiction where reactors actually get built on time and on budget, KEPCO is the most direct play. The Barakah project in the UAE, four APR-1400 units delivered and operational, is the strongest evidence in the world that serial construction of a standardized design achieves competitive economics.

But the state-owned structure, the regulated pricing dynamics that historically suppressed returns, and the Korean market’s structural discount to Western utilities make KEPCO a more complex investment than Constellation. It trades like a regulated utility with construction optionality, not like a growth company, and investors need to accept that framing.

Électricité de France (EPA: EDF) was delisted in 2023 after the French government’s full nationalization. It is no longer investable in public markets but remains the world’s largest nuclear fleet operator (57 reactors) and the key player in every European nuclear initiative.

Kansai Electric Power (TYO: 9503) is the most nuclear-leveraged utility in Japan. It operates the Takahama and Ohi stations, which were among the first to clear Japan’s post-Fukushima restart process, and has the highest nuclear share of generation among the major Japanese utilities. As Japan continues restarting its idled fleet (12 units back online as of early 2026, with more in the pipeline), Kansai captures the restart thesis more directly than any other listed Japanese name. Kyushu Electric Power (TYO: 9508) was the first Japanese utility to restart a reactor post-Fukushima (Sendai Units 1 and 2) and has the operational track record to prove the restart model works. Both trade at valuations that reflect Japan’s historically suppressed utility returns rather than the scarcity value of their nuclear fleets.

CEZ Group (PRG: CEZ) is arguably the most interesting nuclear utility in Europe that almost nobody in the Anglophone investment world is watching. CEZ operates the Dukovany and Temelin nuclear stations in the Czech Republic, both VVER designs that historically depended on Rosatom fuel assemblies. The transition to Westinghouse fuel, which we covered in Part 5, is live and underway. More importantly, CEZ is the developer of the Dukovany 5 new-build project, a 1,200 MWe unit for which the Czech government selected Korea’s KHNP in a closely contested tender. CEZ sits at the intersection of three themes: the European VVER-to-Westinghouse fuel diversification story, the Central European nuclear new-build story, and the broader European energy security repricing. It trades at a fraction of Constellation’s multiples for a fleet that occupies a similarly irreplaceable position in its domestic grid.

Fortum (HEL: FORTUM) owns the Loviisa VVER-440 units in Finland, the same reactor type that received its first Westinghouse fuel loading in late 2024. Smaller nuclear exposure than Kansai or CEZ, but relevant for the VVER diversification thesis and as a liquid European utility with nuclear optionality

8.6 New Build and SMR Developers: Venture Nuclear

This is the part of the market where the narrative is strongest and the fundamentals are weakest. SMR developers are, at this point, pre-revenue venture-stage companies that happen to be publicly listed. They should be sized accordingly.

NuScale Power (NYSE: SMR): The only SMR with full NRC design certification. The Idaho pilot project was cancelled. NuScale has since partnered with ENTRA1 and TVA on a 6 GW deployment program and signed international deals (Ghana, Romania, Czech Republic). The 77 MWe design received its uprated approval in June 2025. Revenue is minimal and the path to profitability is multi-year and execution-dependent. NuScale’s edge is regulatory (no other Western SMR has full NRC design certification), but regulatory approval without a construction site is a credential, not a business.

Oklo (NYSE: OKLO): Sam Altman-backed fast microreactor developer. Plans for demonstration units at Idaho National Laboratory. No NRC license yet (a previous application was denied for insufficient information; Oklo is re-applying). The stock has been a meme-grade performer, driven almost entirely by the Altman association and AI-nuclear narrative.

Nano Nuclear Energy (NASDAQ: NNE): Developing ultra-small portable microreactors. Extremely early stage. More of a concept stock than an operating company.

The framework for SMR/new build developers: These are call options on nuclear construction, priced as such. The implied probability of success embedded in current valuations varies wildly (NuScale’s market cap implies significant near-term commercialization; Oklo’s implies moonshot potential). If you want exposure, size it like a venture bet: small enough that a total loss does not impair the portfolio, large enough that a 5-10x return is meaningful.

8.7 Adjacent and Overlooked Plays

The nuclear supply chain does not end at uranium miners and reactor operators. There is an entire ecosystem of companies that make the components, build the structures, and provide the services without which no reactor operates. These names are underrepresented in most nuclear portfolios because they are not “pure play,” but several of them are better businesses than the pure plays, with higher margins, more defensible moats, and less binary risk profiles.

Curtiss-Wright Corporation (NYSE: CW) is the name most nuclear-focused institutional investors own that never appears in retail-oriented nuclear primers. Curtiss-Wright manufactures reactor coolant pumps, control rod drive mechanisms, naval nuclear components, spent fuel management systems, and instrumentation for essentially every Western reactor type. Their equipment sits inside PWRs, BWRs, and naval propulsion reactors across the US fleet and allied navies. The investment logic is elegant: regardless of which reactor design wins the next decade, regardless of whether SMRs succeed or fail, regardless of whether new build happens on schedule or stalls, Curtiss-Wright supplies the existing fleet and every plausible future fleet. Life extensions, uprates, maintenance, and refurbishment all flow through their order book. It is the “sells ammunition to both sides” play in nuclear, and it trades accordingly.

BWX Technologies (NYSE: BWXT) is the US monopoly supplier of naval nuclear reactor components and fuel. Every US Navy aircraft carrier and submarine runs on BWXT fuel. This is a defense-industrial franchise that happens to have growing commercial nuclear exposure through the eVinci microreactor program, DOE contracts, and medical isotope production. BWXT trades at a defense-sector premium that reflects the naval nuclear monopoly; the commercial nuclear optionality is essentially free in the current stock price.

Doosan Enerbility (KRX: 034020) manufactures the reactor pressure vessels, steam generators, and heavy forgings for Korea’s APR-1400. If you believe the Korean construction thesis (and the Barakah evidence says you should), Doosan is the company that physically builds the hardware. It is the most direct play on Korean nuclear export success. For investors who can access the Korean exchange, Doosan offers exposure to the part of the nuclear value chain where execution risk has actually been demonstrated to be manageable.

Rolls-Royce Holdings (LSE: RR) has two nuclear angles. The legacy business builds and services submarine reactor plants for the Royal Navy. The growth bet is Rolls-Royce SMR, a consortium targeting deployment of a 470 MWe small modular reactor in the UK, with the first unit targeting operations in the early 2030s. The UK government has backed the programme with significant development funding. Rolls-Royce is an enormous company and the SMR programme is a small fraction of revenue, but for investors who want UK nuclear exposure without concentrating in a single-product SMR developer, it provides optionality within a diversified industrial.

GE Vernova (NYSE: GEV) is the parent of GE Hitachi Nuclear Energy, which makes the BWRX-300, arguably the Western SMR closest to commercial deployment (construction started at Darlington, DOE funding for TVA deployment). GE Vernova is a massive power and energy company; nuclear is a modest contributor to overall revenues. But the BWRX-300’s progress makes GEV the lowest-risk way to own a piece of the SMR story, because even if the SMR programme disappoints, the rest of the business (gas turbines, grid equipment, wind) provides a floor. You are not paying a venture multiple for an unproven reactor design. You are paying an industrial multiple with SMR upside attached.

Fluor Corporation (NYSE: FLR) is one of the few Western EPC contractors with demonstrated nuclear construction capability. If the Western new-build pipeline materializes at anything close to the $80 billion framework scale, Fluor is one of a handful of companies that can actually manage the construction. Bechtel (private) is the other. The constraint is that nuclear EPC is a small and historically painful part of Fluor’s business (Vogtle was not a happy experience for anyone involved), but the scarcity of nuclear-capable contractors means Fluor’s optionality here is real.

The Additional Developers

Beyond the Tier 2 names (NexGen, Denison), there is a long tail of uranium developers at various stages of advancement. Most will never become mines. Some will become very good mines. The challenge for investors is distinguishing geology from execution.

IsoEnergy (TSX: ISO) holds the Hurricane deposit in the Athabasca Basin, the highest-grade uranium discovery since NexGen’s Arrow. Hurricane is still at an early stage relative to Arrow or Wheeler River, but the grade (33.9% U₃O₈ in the summer 2022 discovery holes) is extraordinary. IsoEnergy merged with Consolidated Uranium to build a broader portfolio, but Hurricane is the asset that matters.

Ur-Energy (TSX: URE / NYSE: URG) operates Lost Creek, one of the few producing ISR uranium mines in the United States. Small scale, but directly leveraged to US domestic procurement policy (Section 232, critical minerals designation, DOE reserve purchases).

Bannerman Energy (ASX: BMN) holds the Etango project in Namibia, one of the largest undeveloped uranium deposits outside Canada and Kazakhstan. Bannerman is a scale play on higher uranium prices in a stable African jurisdiction.

Deep Yellow (ASX: DYL) merged with Vimy Resources, giving it the Tumas project in Namibia and the Mulga Rock project in Western Australia. Dual-jurisdiction exposure across two of the more politically stable uranium-hosting countries.

Lotus Resources (ASX: LOT) acquired the Kayelekera mine in Malawi from Paladin. Restart economics are leveraged to sustained higher uranium prices.

A note on the developer tail: these names are leveraged bets on higher uranium prices, full stop. They do not have the resource quality of NexGen, the jurisdictional premium of Denison, or the integrated value chain of Cameco. What they have is optionality. In a world where uranium prices sustain above $80 for years and Western utilities are scrambling for supply diversity, some of these developers will be acquired, partnered, or funded into production. In a world where uranium stalls or retreats, most of them will remain exploration companies with nice drill results and no revenue. Size accordingly.

ETFs and Basket Vehicles

For investors who agree with the nuclear thesis but do not want to pick individual names, or who want a liquid core position to complement specific high-conviction bets, three ETFs offer meaningfully different expressions of the same theme.

Sprott Uranium Miners ETF (NYSE: URNM) is the most concentrated. It holds uranium miners and physical uranium vehicles (SPUT is typically the largest holding), giving you direct commodity and producer exposure. This is Proposition 1 (uranium price higher) in ETF form.

Global X Uranium ETF (NYSE: URA) is broader. It includes miners, developers, and some nuclear technology companies, providing more diversified nuclear exposure with less commodity concentration than URNM. It is a blend of Propositions 1 and 4.

VanEck Uranium+Nuclear Energy ETF (NYSE: NLR) is utility-heavy, with significant holdings in Constellation, utilities, and nuclear technology companies alongside miners. This maps most closely to Proposition 3 (fleet operators mispriced) with some commodity exposure layered in.

The compositional differences matter. If you own URNM, you are making a uranium commodity bet. If you own NLR, you are making a nuclear power generation bet. If you own URA, you are somewhere in between. Most investors who buy “a nuclear ETF” do not check which one they own. They should.`

8.8 The TSCS Nuclear Portfolio Framework

The Tier Structure

Rather than prescribing specific allocations (which would go stale before you finish reading this sentence), here is how to think about the tiers of nuclear exposure. The tiers are ranked by the probability-weighted robustness of the thesis, not by expected return. The highest-expected-return names are in Tier 3 and 4 precisely because they carry the most execution risk.

Tier 1: Own the Existing Machine. Physical uranium, the integrated major (Cameco), and the fleet operators (Constellation, Kansai, CEZ). These names do not require a single new reactor to be built anywhere in the West. They require the existing fleet to keep running (it will), fuel demand to grow (it is), and supply bottlenecks to persist (they will). This is the base case, and the base case is enough.

Tier 2: Own the Bottleneck. Centrus for enrichment. Cameco again for conversion (Port Hope). The structural deficit in Western SWU and conversion capacity is the highest-conviction call in this primer, and these are the only public equities that capture it directly. If you own nothing else in nuclear, you should understand why these bottlenecks exist and what they mean for pricing power.

Tier 3: Own the Next Mine. NexGen, Denison, and the better-positioned developers (IsoEnergy, Boss, Ur-Energy). These are leveraged bets on higher uranium prices AND successful project execution. The “AND” is the key word. Uranium can go to $120 and a developer with permitting problems still goes to zero. Geology is necessary. Execution is sufficient.

Tier 4: Own the Future. SMR developers (NuScale, Oklo, Kairos), advanced reactor plays (TerraPower, X-energy), and construction-leveraged names (GE Vernova for BWRX-300, BWXT for eVinci, Fluor and Curtiss-Wright for EPC and components). These are call options on nuclear construction with 5-10 year expiry dates. Size them like options. Define your maximum loss before you enter. Do not average down on hope.

What to avoid:

Jurisdiction traps. A uranium deposit in Niger is a geological fact and a commercial fiction. A mine in Kazakhstan whose offtake is locked into Chinese contracts and whose product transits through Russian rail infrastructure is not a “Western supply” play, no matter what the investor presentation says. The jurisdiction premium is not a temporary risk-off trade. It is a structural repricing of where nuclear fuel can and cannot flow, and it is widening, not narrowing. If you cannot trace the supply chain from pit to reactor through friendly jurisdictions, you do not own what you think you own.

Binary event stocks. A company whose entire valuation depends on a single catalyst (this specific licence, this specific contract, this specific government decision) is not an investment. It is a lottery ticket with a narrative attached. The nuclear sector is full of these because permitting timelines create natural “catalyst dates” that attract speculative capital. Some of them will pay off. Most will not. If you take binary bets, size them as binary bets.

Nuclear cosplay. The nuclear theme has attracted a constellation of companies that rebrand themselves as nuclear-adjacent to capture retail enthusiasm. New ticker, new investor deck, same empty revenue line. The test is simple: open the most recent quarterly filing and look at the revenue number. If it says zero, or if the revenue comes entirely from government grants rather than commercial operations, you are not investing in a nuclear company. You are investing in a nuclear story. Stories can be worth owning at the right price. Just do not confuse them with businesses.

The “I’m long nuclear” basket. Buying one of everything in the nuclear sector is not a thesis. It is a confession that you do not have one. Cameco and Oklo are both “nuclear” in the same way that ExxonMobil and a wildcat driller are both “oil.” They share a theme. They share almost nothing else: not risk profile, not time horizon, not the variables that drive their returns. If you cannot articulate which of the five propositions each name in your portfolio maps to, you have a basket, not a portfolio. Baskets are for ETFs. Portfolios are for investors who did the work.

8.9 The Meta-Framework: What Are You Actually Betting On?

Every nuclear investment is, at root, a bet on one or more of the following propositions. Understanding which proposition each name maps to is the difference between a coherent portfolio and a grab bag of tickers.

Proposition 1: Uranium prices go higher. If this is your primary thesis, own the commodity (SPUT, Yellow Cake) and the lowest-cost producers (Cameco, Kazatomprom, Boss, Denison). Enrichment and utility plays are secondary.

Proposition 2: The fuel cycle bottleneck persists. If you believe enrichment and conversion scarcity is the binding constraint, own Centrus and Cameco (for Port Hope). This thesis does not require uranium prices to spike; it requires SWU and conversion prices to stay elevated, which is a higher-probability bet.

Proposition 3: Nuclear fleet operators are mispriced in a power-constrained world. If you believe data center demand and grid stress will push power prices structurally higher, own Constellation. This is a power market bet that happens to express through nuclear assets.

Proposition 4: Western governments will actually build new reactors. If you believe the “sovereign signal” has been issued and will be sustained, own Cameco (for Westinghouse), BWXT, Fluor, and the SMR developers. This is the highest-conviction, highest-risk proposition, the one most dependent on political variables that are, by definition, unpredictable, and the one where the $80 billion framework’s conversion from aspirational to binding will be the key signal.

Proposition 5: The nuclear theme attracts persistent capital flows regardless of near-term fundamentals. If you believe the narrative momentum (AI + data centers + energy security + climate) will continue attracting generalist capital into nuclear equities, own the liquid large-caps (Cameco, Constellation, NuScale, Centrus) and the physical vehicles (SPUT). This is a momentum/flows thesis, not a fundamental thesis, and it should be traded, not held.

Most thoughtful nuclear investors are betting on some combination of Propositions 1, 2, and 3, with Proposition 4 as upside optionality. The mistake most commonly made is over-indexing to Proposition 4 (new build) at the expense of Propositions 1-3 (commodity, bottleneck, fleet value), because the new-build narrative is more exciting but the commodity and bottleneck theses are more probable and more investable today.

The nuclear renaissance does not require a single new reactor to be built in the West for the investment thesis to work. It requires the existing fleet to keep running (it will), life extensions to proceed (they are), fuel demand to grow (it is), and supply bottlenecks to persist (they will). Everything else is upside.

Build your portfolio from that base, not from the press releases.

Part 2 of this primer takes each of these propositions and stress-tests it against live market data, specific equity positioning, scenario analysis, and the quantitative framework you need to actually deploy capital rather than just understand the sector. That is next.

Part 9: Risks and Bear Cases

9.1 The Section the Bulls Don’t Want You to Read

Every serious bull case is the one that has survived its own bear case. If you cannot articulate exactly what would make you wrong, you do not have a thesis. You have a religion. TSCS does not publish religions.

What follows is an honest attempt to enumerate everything that could go wrong with the nuclear investment thesis, ranked roughly by probability and severity. Some of these risks are real and underappreciated. Some are loudly debated but ultimately manageable. And some are tail risks that would redefine the sector if they materialized. Your job, as an investor, is to assign probabilities and position accordingly.

9.2 Risk 1: Natural Gas Stays Cheap and Kills the Economic Case

Probability: Moderate. Severity: High.

This is the bear case that nuclear bulls dismiss too quickly.

The entire power-price thesis for nuclear (which underpins Constellation’s valuation, the restart economics of Three Mile Island and Palisades, and the new-build business case) depends on electricity prices staying elevated relative to the operating costs of nuclear plants. The single largest determinant of wholesale electricity prices in the United States is the price of natural gas, because gas-fired combined cycle plants are the marginal price setters in most deregulated markets.

If the US natural gas market enters a sustained period of oversupply (which is not implausible given the massive Permian associated gas production, the Haynesville ramp, and LNG export terminal buildout that could be curtailed by demand weakness), Henry Hub prices could stay in the $2-3/MMBtu range for years. At those prices, combined cycle gas plants produce electricity at roughly $25-35/MWh. Nuclear’s operating cost advantage narrows considerably, and the scarcity premium that Constellation captures evaporates.

The counter-argument: gas prices have historically been volatile, and any sustained increase in demand (AI load growth, LNG exports, coal retirements) would tighten the gas market. Also, nuclear’s value is not purely economic; its 92%+ capacity factor, zero carbon emissions, and baseload reliability carry premiums that gas cannot replicate, especially as carbon pricing and clean energy standards become more stringent. But if your thesis depends on power prices, you need to stress-test it against a world where gas stays cheap.

9.3 Risk 2: The Cost of Capital Is the Hidden Variable in Every Nuclear LCOE Comparison

Probability: High (rates are already elevated). Severity: High for new build, Low for existing fleet.

This is the bear case that most nuclear investment theses simply ignore, and it is arguably the most important variable in the entire new-build economic analysis.

Nuclear power is the most capital-intensive energy technology in commercial deployment. A 1 GW AP1000 has an all-in construction cost of $8-12 billion assuming NOAK economics (which, for the record, have never been demonstrated in the West). A comparable 1 GW solar plus 4-hour battery installation costs roughly $1.5-2.5 billion.

The operational economics favor nuclear (near-zero marginal fuel cost, 92% capacity factor, 60-80 year lifetime), but the upfront capital requirement is 3-5 times larger. This means nuclear’s levelized cost of electricity is extraordinarily sensitive to the discount rate.

Here is the math that most nuclear advocates prefer you not see.

The sensitivity is not symmetrical. Moving from 3% to 5% adds roughly $20-25/MWh. Moving from 5% to 7% adds another $25-30/MWh. Moving from 7% to 10% adds $40-50/MWh. The relationship between cost of capital and LCOE is convex, not linear, because IDC compounds on a larger and larger base as the project takes longer and costs more.

Without at least one of these, no private developer will break ground. The $80 billion Westinghouse partnership framework explicitly includes “the US Government arranging financing,” which is an acknowledgment of this reality, not an incidental detail. If interest rates stay elevated (the 10-year Treasury yield has been above 4% since 2023), the government’s cost of facilitating that financing increases, reducing the political appetite for nuclear subsidies and increasing the effective cost to taxpayers.

Every 100 basis points on a $10 billion project adds roughly $100 million per year in interest expense during a 5-7 year construction period. That compounds into billions over the project lifecycle.

For investors, this risk is most acute for Tier 3 and Tier 4 plays (construction-leveraged companies and SMR developers). The existing fleet is unaffected: those plants are already built, already depreciated, and generate electricity at $25-30/MWh operating cost regardless of where the 10-year Treasury sits. Constellation’s fleet is actually a beneficiary of higher rates, because elevated rates make it harder for competitors to build new capacity, reinforcing the scarcity premium on existing nuclear baseload.

The interest rate bear case is a new-build bear case, not a sector bear case, and your portfolio construction should reflect that distinction.

The counter-argument: government-facilitated financing (DOE LPO loans, sovereign guarantees) effectively removes rate sensitivity for the projects that receive support. The Inflation Reduction Act’s nuclear PTC ($15/MWh) and the DOE’s expanded lending authority provide a floor. And in countries like China and Korea, where the government is the developer, the cost of capital is whatever the sovereign decides it is. But for anyone modeling Western new-build economics, stress-testing against a range of discount rates (3%, 5%, 7%, 10%) is not optional. It is the single most important variable in the analysis.

9.3 Risk 3: SMR Cost Overruns Replicate Large Reactor Problems at Smaller Scale

Probability: High (for first-of-a-kind projects). Severity: Moderate to High.

The entire promise of SMRs is that they will be cheaper, faster, and more financeable than large reactors. The theory is compelling: factory fabrication enables manufacturing learning curves, smaller modules reduce on-site construction risk, standardized designs eliminate bespoke engineering, and shorter build times reduce interest during construction.

The problem is that NuScale’s Idaho pilot, the most advanced Western SMR project, was cancelled in November 2023 after cost estimates nearly doubled, from roughly $5.3 billion to $9.3 billion, largely due to inflation, rising interest rates, and the inherent challenge of costing a first-of-a-kind project. The cost per MWe for the NuScale CFPP had reached roughly $20,000/kW, which is not competitive with anything.

Every SMR project currently advancing is a first-of-a-kind deployment. FOAK projects in nuclear have a 100% historical track record of cost overruns. Not 80%. Not 90%. One hundred percent. Vogtle, Flamanville, Olkiluoto, Hinkley Point C, even the relatively well-managed Barakah project in the UAE experienced schedule adjustments. The learning curve that makes Nth-of-a-kind (NOAK) units cheaper requires surviving the FOAK pain, and there is no guarantee that investors and utilities will have the patience or capital to push through FOAK disappointments to reach NOAK economics.

If the BWRX-300 at Darlington comes in late and over budget (which is the base case for any honest analyst given historical precedent), the narrative damage to the SMR thesis could be severe, even if the long-term economics remain sound. Markets do not distinguish between “temporary FOAK growing pains” and “SMRs don’t work.” They react to the headline.

The counter-argument: China’s Linglong One and Shidaowan HTGR are demonstrating that advanced and small reactor designs can be built on time, and the BWRX-300 is an evolutionary design based on 65 years of BWR operating experience, not a speculative Gen IV concept. The risk is real but not existential; it delays the timeline rather than killing the thesis.

9.4 Risk 4: A Nuclear Accident Resets Public Sentiment

Probability: Low. Severity: Extreme.

This is the tail risk that haunts the sector. The global nuclear fleet has operated for over 18,000 reactor-years with three significant accidents (TMI, Chernobyl, Fukushima), two of which involved reactor designs (RBMK and early BWR) that are either extinct or dramatically improved. The probability of a major accident at a modern PWR or BWR is estimated at roughly 1 in 10 million reactor-years by probabilistic risk assessment. This is an extraordinarily low probability.

But the consequences of a single accident, anywhere in the world, would be catastrophic for the sector regardless of the actual radiological harm. Fukushima killed zero people from radiation, and it still shut down Japan’s entire nuclear fleet and catalyzed Germany’s nuclear exit. The public does not process nuclear risk through probability distributions. It processes nuclear risk through narrative and fear. One mushroom cloud photograph, even metaphorical, can destroy a decade of positive momentum.

The countries most vulnerable to an accident-driven sentiment reversal are those with large, aging fleets (US, France) and those with new, untested designs entering service (China, which is deploying more novel reactor types simultaneously than any country in history). An accident at a Chinese Gen IV reactor, for instance, would not just affect China; it would trigger a global policy reassessment.

The counter-argument: modern reactor designs have multiple, independent, passive safety systems that make severe accidents orders of magnitude less likely than at TMI, Chernobyl, or Fukushima. The defense-in-depth principle (multiple barriers between radioactive material and the environment) has been validated across 70 years of commercial operation. And the post-Fukushima regulatory response (seismic re-evaluations, flood protection upgrades, portable emergency equipment) has further reduced residual risk. But “low probability” is not “zero probability,” and anyone who tells you nuclear is risk-free is selling something.

9.5 Risk 5: The “Structural Deficit” Is Smaller Than the Bulls Claim

Probability: Moderate. Severity: Moderate.

We covered this honestly in Part 4, but it bears repeating in the risk section because it is the most common source of disappointment for uranium investors.

The bull case rests on a structural gap between primary uranium production (~150-160 million lbs/year) and reactor demand (~185 million lbs/year), with secondary supplies depleting. But the deficit is measured against requirements, not actual purchases. Utilities routinely undercontract, defer procurement, manage inventory cycles, and optimize fuel assemblies to stretch the life of existing fuel loads. The “deficit” in any given year is partly a real physical gap and partly a reflection of timing decisions by sophisticated buyers who have years of fuel inventory and no urgency.

Additionally, supply elasticity is not zero. At $80+/lb, projects that were uneconomic at $40 are restarting. ISR operations in Wyoming and Texas are ramping. Australia’s Honeymoon is producing at $23/lb cash cost. Paladin’s Langer Heinrich is back. NexGen and Denison are advancing toward production. Kazatomprom has latent capacity it could bring online (if acid supply and political will permit). The question is not whether new supply responds to price signals but whether it responds fast enough to prevent price spikes, and history suggests that commodity supply responses always lag the cycle by several years, which creates the volatility that makes uranium both attractive and treacherous.

The specific scenarios where the deficit thesis disappoints:

China’s reactor construction decelerates. The demand projections throughout this primer assume China’s reactor program executes at planned pace: 10 units approved per year, 110 GW by 2030, 150 GW by 2035. But China’s economy is navigating a property crisis, demographic decline, potential deflation, and industrial overcapacity. If electricity demand growth undershoots plan, the reactor approval pipeline slows, and Chinese uranium/SWU demand misses projections, the global supply-demand picture loosens materially. Given that China is responsible for essentially all growth in global nuclear generation outside the existing Western fleet, a Chinese demand miss is not a tail risk. It is a plausible scenario that deserves explicit weight in any supply-demand model. The 132,000-tonne strategic stockpile provides additional buffer: if Beijing draws it down rather than procuring on the open market, it removes a major demand source from the global balance for years.

Kazakhstan resolves its acid supply problems and ramps to full subsoil use agreement capacity (possible by 2027-2028). A major undiscovered secondary supply source emerges (government stockpile release, accelerated reactor decommissioning freeing fuel). Japan’s restarts stall again (reducing demand growth from the second-largest historical nuclear fleet).

The counter-argument: even accounting for all of these factors, the directionality of the deficit is robust. The demand trajectory (reactor life extensions, Chinese construction, data center PPAs) is growing faster than supply can respond. The secondary supply buffer is thinner than at any point in the last two decades. And the geopolitical bifurcation of the market (Western vs. Eastern supply chains) means Western utilities are competing for a shrinking pool of accessible uranium, even if the global aggregate looks more balanced.

9.6 Risk 6: The Western Coordination Freeze Never Actually Breaks

Probability: Moderate. Severity: High for new-build plays, low for commodity plays.

This is the Arden risk, and it is the most intellectually interesting bear case.

Suppose the US government issues all the right signals (executive orders, $80 billion partnership frameworks, NRC reform, HALEU funding) but fails to follow through with the institutional restructuring that Arden argues is necessary. Suppose the “sovereign signal” turns out to be visible but not irreversible: a subsequent administration reverses the executive orders, defunds the DOE contracts, and the nuclear construction program dies in its cradle. Suppose the coordination freeze is not a temporary political condition but a permanent feature of American institutional architecture, baked into the separation of powers, adversarial legalism, and fragmented regulatory structure that Delmas and Heiman diagnosed two decades ago.

In this scenario, the new-build thesis fails. The $80 billion AP1000 program does not deliver. SMR deployments slip from the 2030s to the 2040s. The US remains a country that operates its existing nuclear fleet (well) but cannot build new reactors (at all). The investment implications: Tiers 3 and 4 from Part 8 (construction plays and SMR developers) underperform, while Tiers 1 and 2 (commodity, producers, fleet operators) continue to benefit from the existing fleet’s insatiable fuel demand and irreplaceable power generation.

This is why we structured the portfolio framework the way we did. The core positions (Cameco, Constellation, SPUT) do not require a single new Western reactor to be built. They require the existing fleet to keep running, fuel demand to grow, and supply bottlenecks to persist. All of those conditions are already true. New build is upside optionality, not the base case.

The counter-argument: China and Korea will build regardless of what the US does, and their construction programs drive the marginal demand for uranium, enrichment, and conversion. The thesis does not depend on Washington solving its coordination problem. It depends on global nuclear capacity growing, which the IAEA has projected upward for five consecutive years.

9.7 Risk 7: Fusion Disrupts the Thesis (Eventually)

Probability: Low within investment horizon. Severity: Theoretically extreme.

Fusion is the permanent “30 years away” punchline of the energy industry. But genuine progress is being made. The NIF achieved ignition (more energy out than laser energy in, though not more than total facility energy) in December 2022. Commonwealth Fusion Systems has raised approximately $3 billion and is targeting Q>1 demonstration with its SPARC tokamak by 2027. Helion Energy demonstrated the first private deuterium-tritium fusion in February 2026 with its Polaris prototype, achieving plasma temperatures of 150 million degrees Celsius. China’s EAST tokamak has achieved increasingly long plasma confinement times.

If commercial fusion arrives, it renders fission obsolete. Unlimited fuel (deuterium from seawater), no long-lived radioactive waste, no meltdown risk, and energy densities that make even uranium look quaint.

But “arrives” means “delivers electricity to the grid at competitive cost,” which is an engineering and economic challenge of staggering complexity. No fusion device has ever produced net electricity. The most optimistic commercial timelines (Commonwealth Fusion Systems targets the early 2030s for a demonstration plant) imply grid-scale commercial deployment no earlier than the 2040s, and that assumes no setbacks.

For any investment with a 5-10 year horizon, fusion is not a relevant risk factor. For a 20-30 year horizon, it is the most important variable in the energy landscape. Nuclear investors should be aware of it, monitor it, and not lose sleep over it this decade.

9.8 Risk 8: Tariffs, Trade Wars, and Policy Reversals

Probability: High (they are already happening). Severity: Moderate.

The Trump administration’s tariffs on Canadian goods directly affect Cameco, the largest Western uranium producer, by adding approximately 10% to costs for US customers. The Section 232 investigation on critical minerals (including uranium) could result in further trade restrictions. The ongoing uncertainty about whether uranium, conversion, and enrichment services will be included in future tariff regimes has already suppressed utility contracting activity through most of 2025 (as we documented in Part 4).

Beyond tariffs, the nuclear fuel sector is uniquely vulnerable to policy whiplash. The US ban on Russian uranium imports (effective 2024) could theoretically be modified or waived under a future administration that prioritizes engagement with Russia. DOE funding commitments ($2.7 billion for enrichment, $400 million for BWRX-300 deployment) could be rescinded if Congress claws back appropriations. The IRA’s nuclear production tax credits, which provide $15/MWh for existing nuclear plants, could be modified or eliminated.

The counter-argument: bipartisan support for nuclear is stronger than at any point in 40 years. The ADVANCE Act passed with overwhelming margins. Nuclear is one of the few energy topics where the progressive climate caucus and the conservative energy security caucus agree. Policy reversals are possible but would require a political environment that does not currently exist.

9.9 Risk 9: Proliferation Concerns Constrain Enrichment Expansion

Probability: Low to Moderate. Severity: Moderate.

Enrichment is dual-use technology. The same centrifuges that produce 4% LEU for reactors can, with additional cascading, produce 90% HEU for weapons. As the West expands enrichment capacity (Urenco, Orano, Centrus, General Matter), and as HALEU production (5-19.75% enrichment) grows for advanced reactors, proliferation concerns will inevitably intensify.

The Nuclear Suppliers Group’s debate over export controls for carrier and coolant salts (used in advanced reactor designs) has already delayed a US rulemaking on advanced reactor exports. If proliferation hawks succeed in imposing tighter export controls, it could constrain the international market for enrichment services and slow the deployment of advanced reactors that require HALEU fuel.

This risk is most relevant for Centrus (which is building HALEU production capacity) and for the advanced reactor developers (Kairos, X-energy, TerraPower) whose designs require fuel enriched above the 5% threshold that has been the commercial standard for decades.

9.10 Risk 10: Long-Duration Storage Is Closer Than the Bulls Acknowledge

Probability: Moderate (technology is real; commercial scale is unproven). Severity: Gradual erosion of new-build economics, not acute disruption.

The standard nuclear bull rebuttal to the “renewables plus storage” bear case goes like this: batteries can only store 4 hours, therefore you need nuclear for baseload. This was accurate in 2023. It is becoming less accurate in 2026, and it may be materially wrong by 2030. Solar panel costs have fallen 99% since 1976. Lithium-ion battery costs have fallen 97% since 1991. These cost curves show no sign of flattening. Solar plus 4-hour battery storage is approaching $30/MWh all-in in the sunniest regions.

On a pure LCOE basis, new-build nuclear is already two to three times more expensive than utility-scale solar in most geographies. If the gap widens further, the political will to subsidize new nuclear construction could erode. But the real threat is not 4-hour lithium-ion. It is 100-hour iron-air. Form Energy, backed by $405 million in recent funding from GE Vernova, Breakthrough Energy Ventures, and Energy Impact Partners, has developed an iron-air battery capable of 100 hours of discharge. The chemistry is elegantly simple: the battery “rusts” iron during discharge (absorbing oxygen from the air) and reverses the process during charging. The raw materials are iron, water, and air. Form’s target cost is $20/kWh, roughly one-fifth of current lithium-ion on a per-kWh basis. At 100 hours of duration, this translates to approximately $2,000/kW of firm capacity, which, if achieved, would be cost-competitive with new-build nuclear on a firmed-capacity basis.

The deployment timeline is no longer theoretical. Form’s first commercial pilot (1.5 MW / 150 MWh with Great River Energy in Minnesota) broke ground in August 2024 and targets late 2025 commissioning. Three GWh-scale deployments are targeted for 2026: two projects with Xcel Energy (10 MW / 1 GWh each) and a 15 MW / 1.5 GWh facility for Georgia Power. In February 2026, Google and Xcel Energy announced a 300 MW / 30 GWh iron-air deployment, described as “the largest battery project by gigawatt-hour energy capacity announced to date in the world.” Form’s largest planned installation, an 85 MW / 8.5 GWh project in rural Maine funded by a $147 million DOE grant, targets construction start in 2027 and operations in 2028.

Form’s Weirton, West Virginia factory is producing batteries now, with 14 GWh of projects announced. The nuclear bulls are correct on two technical limitations. First, round-trip efficiency is low, roughly 40-50% compared to 90% for lithium-ion. You lose half the energy in the charge-discharge cycle. For a grid with abundant cheap solar where curtailed energy is effectively free, this penalty is tolerable.

For a grid where every MWh costs money, it is significant. Second, the manufacturing scale needed to meaningfully affect grid-scale storage (tens of GWh per year) has not yet been demonstrated. The factory exists. The production rate does not. But it would be analytically dishonest to dismiss iron-air as irrelevant when it has $405 million in funding, contracts with six major utilities, a DOE grant, a collaboration with GE Vernova, a 30 GWh project with Google, and a cost target that, if achieved, fundamentally changes the firmed-renewable cost calculation.

The scenario where this matters for the nuclear thesis is not “iron-air kills nuclear.” It is “iron-air reduces the economic justification for new-build nuclear in regions with abundant solar, by providing multi-day storage at costs below nuclear’s LCOE.” This does not affect the existing fleet (already built, operating at $25-30/MWh). It does not affect fuel demand (driven by the existing fleet plus life extensions). It affects Tier 3 and Tier 4 plays, the new-build and SMR developers whose business case depends on nuclear being the only option for firm clean power. If iron-air delivers on its cost and performance targets by 2030-2032, the addressable market for new nuclear narrows from “all new baseload” to “baseload in land-constrained, low-solar geographies,” which is a smaller but still enormous market.

The LCOE comparison between nuclear and renewables is also misleading without accounting for system costs (grid integration, transmission buildout, curtailment, backup capacity). The “firmed” cost of renewables, meaning solar plus sufficient storage to provide baseload-equivalent reliability, is significantly higher than the headline LCOE, and in some analyses still approaches or exceeds nuclear’s cost even with iron-air. The existing nuclear fleet’s operating costs remain highly competitive with any alternative regardless of what happens to new-build economics. The threat from long-duration storage is to the new-build thesis, not to the existing fleet thesis, and your portfolio construction should reflect that asymmetry.

9.11 Risk 11: You Are Early, and Early Is the Same as Wrong

Probability: High. Severity: It depends on your patience.

This is the meta-risk that subsumes all others. The uranium bull thesis has been “right” for three years and has delivered frustrating returns to spot-price-focused investors. Spot uranium was $63-83 in 2025. Utilities undercontracted for the third consecutive year. SMR timelines slipped. New build remains a press release, not a construction site.

The thesis is directionally correct. The fundamentals are tightening. The policy support is real. But commodity cycles operate on timescales measured in years, not quarters, and the uranium cycle in particular moves slowly because utilities are constitutionally incapable of urgency. If you are a quarterly-reporting hedge fund, the uranium thesis may not deliver within your performance measurement period. If you are an endowment or a patient individual investor, the asymmetry is compelling but the holding period is indefinite.

The greatest risk in nuclear investing is not being wrong. It is being right too early, running out of patience, selling at the bottom of a sideways consolidation, and watching the thesis play out in someone else’s portfolio. The nuclear industry operates on timelines measured in decades; capital markets operate on timelines measured in quarters. That mismatch is where both the opportunity and the pain live. Size accordingly. Define your time horizon before you enter. And do not confuse short-term price action with the soundness of the underlying analysis.

9.12 The Synthesis: What the Bear Cases Tell You

If you have read this far, you now hold both the bull case and the bear case in your head simultaneously. That is the point. The strongest investment conviction comes not from ignoring the risks but from pricing them.

Here is what the bear cases, taken together, tell you about portfolio construction:

The existing fleet is more defensible than new build. Risks 1, 2, 5, 6, and 9 all hit new-build and SMR plays harder than they hit commodity and fleet-operator plays. The existing fleet is already built, already licensed, already fueled, and already generating cash. It is insulated from construction risk, SMR execution risk, and regulatory uncertainty. If you want nuclear exposure with the narrowest risk profile, own the fleet (Constellation) and the fuel (Cameco, SPUT).

The fuel cycle bottleneck is the highest-probability thesis. Risks 4 and 5 could soften the uranium commodity thesis at the margins, but the enrichment and conversion bottlenecks (covered in Parts 3 and 5) are structural, funded by $5.6 billion in government commitments, and will take until the early 2030s to resolve regardless of what happens to the uranium price. Centrus and Cameco’s conversion business are positioned at these bottlenecks.

New build is real upside but should be sized as optionality, not conviction. Risks 2, 5, 7, and 8 all introduce uncertainty around Western construction timelines that no amount of executive orders can eliminate. If new build delivers, the payoff is enormous (for Westinghouse, for EPC contractors, for fuel demand). If it stalls, the commodity and fleet theses still work. Structure accordingly.

Time is the binding variable. Risk 10 is the one that actually costs people money. Everything else is analytical. Patience is operational. The thesis will either be validated or invalidated, but it will not be validated quickly. If you cannot hold for 3-5 years, the nuclear sector is not for you, regardless of how compelling the fundamentals look on paper.

9.13 Final Words

If there is a single unifying framework for everything in this primer, it is that the unit of the nuclear industry is the decade. Construction takes a decade. Licensing takes a decade. Fuel contracts run for decades. Reactor lifespans stretch across eight decades. Enrichment capacity takes a decade to build. Workforce pipelines require a decade to fill. The geopolitical dependencies embedded in Russian fuel supply chains will take decades to unwind.

Capital markets, structured around quarterly reporting and 3-to-5-year investment horizons, are fundamentally mismatched to an industry where the relevant planning unit is 10 years. Understanding this mismatch is the single most important edge you can have as a nuclear investor.

The answer to why nuclear is suddenly everywhere, in government policy, in tech company strategy, in investment flows, and in the public imagination, is structural. The world needs more electricity. It needs that electricity to be reliable, available 24/7, and low-carbon. The list of technologies that meet all three criteria is exactly one item long. The fuel that powers it is structurally undersupplied. The supply chain that processes it is concentrated in adversarial or contested jurisdictions. And the Western response, while real and funded, will take the better part of this decade to materialize.

That is the thesis. It is not dependent on hype. It is not dependent on any single SMR working. It is not dependent on any single government keeping its promises. It is dependent on the physics of energy demand, the chemistry of the fuel cycle, and the arithmetic of supply and demand. Those are the hardest things to argue with.

Whether you invest a dollar in this sector or not, you now understand it. That was the point.

Share TSCS

Share

That is China’s share of global rare earth chemical separation. Not mining. Processing.

The step that turns worthless rock into the separated oxides and metals inside every EV motor, every wind turbine, every fighter jet. Sources vary from 87% to 93% depending on how you count Chinese-backed operations in Myanmar and Vietnam, but the USGS, IEA, and European Parliament all converge on approximately 90%.

I have looked at a lot of commodity supply chains. I have never seen a single-country concentration at a critical downstream processing step that approaches this level.

Semiconductors come close (Taiwan at 90%+ for leading-edge logic), and we have spent three years and $280 billion trying to fix that. Rare earths are the semiconductor problem, but for atoms instead of transistors.

The financial markets are not paying attention.

This is partly forgivable: global rare earth oxide is a $15–20 billion market, roughly one tenth the size of the lithium market and invisible next to oil. It does not screen on any quantitative fund’s radar. There is no liquid futures contract to attract speculative capital. The sell side covers it thinly because the banking fees are small. And so a chokepoint that matters more to Western industrial capacity than any single commodity except oil sits in the analytical blind spot between “too small to model” and “too important to ignore.”

China controls 69% of global rare earth mining, approximately 90% of chemical separation, and 94% of permanent magnet manufacturing. It has demonstrated, in real time across multiple escalation cycles in 2025, that it will weaponise that control. New non-China supply is 3 to 5 years from meaningful scale. Demand from EVs, wind, defense, and robotics is growing at 8–11% annually and is price-inelastic in the current range (though not at extreme prices, as the 2011 cycle proved).

The handful of Western companies positioned to fill this gap are in the early innings of a repricing the market has only partially absorbed.

This is not a geopolitical essay dressed up as investment research. The war in Iran, the defense spending ramp, the trade war: those are catalysts, not theses. They accelerate a repricing that was already underway. The thesis is structural, and it rests on arithmetic that most of the market has not done because rare earths are small, obscure, and boring.

Until they are not.

This piece is 10,000 words. It contains a reconstructed marginal cost curve for non-China rare earth production that does not, to my knowledge, exist elsewhere in public research, a probability-weighted scenario framework with explicit equity outcomes, and the bear case that most bulls refuse to write. It draws on USGS data, Adamas Intelligence, Fastmarkets assessed pricing, Bloomberg Terminal data, Capital IQ consensus models, and ISP-Myanmar satellite analysis.

It is free because the thesis is too important to gate. TSCS publishes institutional-grade equity research and commodity deep dives for a readership that includes portfolio managers, analysts, and founders at over 60 investment firms. If this is what we publish for free, you already know what the paid research looks like.

Subscribe now

Let’s get into it.

Seven Links in a Chain China Owns

The first thing most people get wrong about rare earths is treating them as a mining story. It is a processing story, and it always has been. Mining is where China is weakest, and it still controls 69%. What happens downstream is where the thesis lives.

(Chart: Global rare earth mine production by country, 2020–2024. Source: USGS Mineral Commodity Summaries 2025.)

Global rare earth mine production reached approximately 390,000 metric tonnes of rare earth oxide equivalent in 2024. China’s production quota: 270,000 MT (69%). The US produced 45,000 MT from MP Materials’ Mountain Pass mine (11.5%), followed by Myanmar at 31,000 MT (8%), Australia at 13,000 MT (3.3%), and newcomers Thailand and Nigeria at 13,000 MT each.

Those numbers look concerning. They are not the concerning numbers.

The rare earth supply chain has seven distinct stages: mining, concentration, chemical separation, metal reduction, alloying, magnet manufacturing, and magnetization. China’s dominance increases at each downstream step. Mining: 69%. Chemical separation: ~90%. Heavy rare earth separation (dysprosium, terbium): ~99%. Sintered NdFeB permanent magnets, the actual product that goes into an F-35 actuator, an EV traction motor, or an offshore wind turbine: 94%.

(Chart: China’s share at each supply chain stage. Source: USGS, IEA, Adamas Intelligence.)

The entire non-China world produces perhaps 20,000–25,000 tonnes of NdFeB magnets. Japan’s Proterial, Shin-Etsu, and TDK collectively produce roughly 10,000–15,000 tonnes. Europe adds perhaps 8,000–10,000 tonnes. The United States had essentially zero large-scale magnet manufacturing until MP Materials began trial production in late 2025.

Mining without processing is like having crude oil without a refinery. The rock is worthless.

Why Separation Is So Hard

Here is the problem with separation, and why “just build a plant” is a sentence that betrays its speaker’s ignorance. The 17 lanthanide elements have nearly identical chemical properties. They differ by single electrons buried in inner orbital shells. Their ionic radii vary by less than 0.2 angstroms across the entire series. Imagine trying to sort 17 shades of off-white by eye, at industrial scale, to 99.5% purity. That is what solvent extraction does, through hundreds to over 1,000 mixer-settler stages, running 4–15 weeks of continuous processing per batch.

A commercial-scale separation plant costs $200–500 million for the circuit alone, with a fully integrated facility running $1.5–2 billion. Western environmental regulations add 3–5 years of permitting versus Chinese approvals. Production costs outside China remain 5–7x higher. China holds more patents on rare earth separation and magnet technology than the rest of the world combined.

As of early 2026, the non-China separation landscape: Lynas LAMP in Malaysia (~7,000 tonnes/year NdPr equivalent), MP Materials (2,599 MT in 2025, doubled from 2024), Energy Fuels White Mesa (~1,000 tonnes/year), Neo Performance in Estonia (Europe’s only separator). REEtec is commissioning a pilot in Norway. Iluka Eneabba, Lynas Seadrift, Ucore Louisiana are under development but none reach meaningful scale before 2027.

Total non-China NdPr separation capacity: approximately 10,000 tonnes per year. Global NdPr demand: roughly 55,000–60,000 tonnes and growing. That is 15–17% non-China capacity for the magnet-critical elements that the entire EV, wind, defense, and robotics supply chain depends on.

If someone told you that about semiconductor fabrication, or oil refining, or pharmaceutical API production, you would call it a national security emergency. In rare earths, it is Tuesday.

The Balance Problem and the Thorium Problem

Most rare earth analysis ignores two structural barriers that explain why Western supply chains keep failing. These are not market inefficiencies. They are physics and politics, and no amount of capital solves either quickly.

The balance problem. Geology determines the element mix. A typical bastnäsite deposit yields roughly 50% cerium, 25% lanthanum, 17% neodymium, and small percentages of everything else. Cerium and lanthanum are abundant, low-value, and in structural oversupply. But to get the neodymium and dysprosium the market needs, you must mine, process, and find buyers for the cerium and lanthanum too. Projects that look spectacular on a NdPr-only basis often fail when the low-value elements pile up unsold. China manages the balance problem through state coordination. Western miners face it as an open market problem with no coordination mechanism.

Thorium. Rare earth ores, particularly monazite, contain thorium, a naturally occurring radioactive element. This is the single biggest regulatory barrier to Western processing. Lynas has faced sustained political opposition in Malaysia over radioactive waste. Any new Western separation facility must navigate regulations that add years and hundreds of millions in costs. This is a structural cost advantage that cannot be replicated without environmental trade-offs that democratic societies may not tolerate. It is also, bluntly, the reason China dominates this industry. Beijing made trade-offs in the 1990s that no Western democracy would have accepted, and the resulting cost structure is now 30 years ahead. You do not close that gap with subsidies. You close it by accepting some version of the same trade-offs, or you do not close it.

Four Elements That Run the World

Seventeen rare earth elements exist. The investment thesis depends on four.

Neodymium and praseodymium (traded as NdPr in a 75:25 ratio) provide the magnetic energy density that makes NdFeB magnets the strongest permanent magnets commercially available: 40–55 MGOe, roughly ten times ferrite and twice samarium-cobalt. When you need maximum torque per unit weight, NdFeB is not optional. It is required by physics.

Dysprosium dramatically improves coercivity (resistance to demagnetization at high temperatures), essential for motors and generators operating above 150°C. Terbium performs a similar function even more effectively per unit weight but is scarcer and roughly four times more expensive.

Both are heavy rare earth elements concentrated in ionic clay deposits in southern China and Myanmar.

Myanmar’s rare earth supply is in flux. In October 2024, the Kachin Independence Army seized Chipwi and Pangwa, the two towns at the centre of the country’s heavy rare earth extraction. China closed border gates in response, halting shipments. A new arrangement emerged by early 2025 (KIA taxation at 35,000 yuan per metric ton, exports partially resuming), but as of early 2026, Kachin mining has not fully recovered. Production is shifting to Shan State under UWSA control, and both the US and India are exploring direct engagement with the KIA to bypass Chinese processing entirely. The net effect on global supply: disrupted, not eliminated, but structurally less reliable than the 31,000 MT headline suggests, and any further escalation in Kachin threatens ~40% of global heavy rare earth mine output.

Pricing: The Two-Tier World That Is the Thesis

Three price points matter. Most analysis conflates them, and the conflation is not innocent. It obscures the single most important variable in the thesis.

Chinese domestic NdPr oxide (the Baotou benchmark): approximately $90–100/kg as of late January 2026, hitting $97.97/kg on January 15 and $99.94/kg on January 27. Multi-year highs. BMI/Fitch Solutions revised its 2026 average forecast to $90/kg.

Neodymium metal (one processing step downstream of oxide): 1,065,000 CNY/tonne on Chinese domestic markets as of mid-February 2026, up 94.5% year-over-year. Approximately $146/kg. Ex-China pricing data is thinner: off-market transactions suggest premiums of 30–50% over Chinese domestic, implying $190–220/kg.

Ex-China NdPr oxide: approximately $90–110/kg through January 2026. NdPr oxide has gained over 40% year-to-date and these numbers may already be stale.

Dysprosium oxide: ~$930/kg ex-China, up 105% year-to-date. Terbium oxide: reached $3,625/kg in ex-China markets during mid-2025, nearly four times the Chinese domestic price.

On an oxide-to-oxide basis, the ex-China premium over Chinese domestic NdPr has historically run 50–80%. On dysprosium oxide, the premium exceeds 200%. Both are historically unprecedented.

There is no LME contract for NdPr. No COMEX equivalent. Price discovery occurs through Chinese domestic pricing, ex-China assessed pricing (Fastmarkets, Asian Metal, compiled from reported deals, not transaction prices), and increasingly long-term contracts with floor-ceiling mechanisms (the DoW–MP $110/kg being the most prominent). As the market bifurcates, the “NdPr price” that most analysis cites becomes less meaningful. The thesis depends on the ex-China price and the friendshoring spread. I want to be unambiguous about this: the spread is the thesis. Every other number in this piece, the mining tonnages, the demand forecasts, the government subsidies, is context for a single question: how wide is the gap between Chinese domestic pricing and what the rest of the world has to pay? If the spread widens, the equities reprice. If it compresses, the trade dies. Everything else is narrative.

The Cost Curve

Most rare earth analysis skips the marginal cost curve. That is an extraordinary omission for a commodity thesis. I have reconstructed it from public disclosures, industry estimates, and conference presentations.

Tier 1: Chinese integrated producers (China Northern Rare Earth, China Rare Earth Group). Fully loaded NdPr oxide at $25–40/kg. Thirty years of learning-curve effects, state-subsidised infrastructure, domestic cerium/lanthanum demand. Critically: these are instruments of industrial policy, not profit-maximisers. Their floor is set by political objectives, not marginal cost.

Tier 2: Lynas. Cash cost ~$45–55/kg. Fully loaded (sustaining capex, LAMP licence, Kalgoorlie amortisation) likely $55–70/kg. Below $55, survival-level margins. At $90+, expansion economics are compelling.

Tier 3: MP Materials. Mountain Pass mining is cheap ($15–20/kg REO equivalent) because the deposit is high-grade and operational. Separation is new, still ramping. Fully loaded NdPr oxide likely $60–80/kg at current volumes, improving toward $50–65 as throughput scales by 2027–2028. The DoW floor on magnet offtake provides margin buffer the oxide business alone does not.

Tier 4: Energy Fuels, Neo Performance, emerging separators. Estimated $80–110/kg at current small scale. Marginally viable at $90/kg ex-China NdPr. Cash-negative at $70 without government subsidies.

Tier 5: Pre-production projects (Iluka, Arafura, others). Projected $50–70/kg at nameplate, but these are feasibility-study estimates. Apply a 30–50% contingency. Iluka’s 50% capex overrun before commissioning is instructive.

The structural implication is stark. If Beijing crashes NdPr to $40/kg (which it can sustain for years on state-subsidised production), every non-China producer except Lynas is below cash cost, and Lynas is at breakeven. This is the same kill mechanism that destroyed the 2011 cycle and wiped $1.5 billion of Molycorp equity to zero. The countervailing factor, absent in 2011: Western governments have demonstrated willingness to backstop producers through the trough. But “willingness” is doing heavy lifting in that sentence. The backstops are partial, politically contingent, and subject to the attention span of democratic legislatures. They are better than nothing. They are not a guarantee.

The cost curve says: the thesis survives at $80+ NdPr oxide. It becomes a survival test between $50 and $80. Below $50, it dies without full government bailouts. This is why the equity valuation later in this piece uses $90/kg as its base case rather than the $110+ the Street appears to assume: $90 is the price at which non-China producers are profitable and expanding, but not the price at which speculative supply floods in. It is the sustainable middle of the cost curve, and the valuation framework that follows reflects that conservatism.

The 2011 Ghost, or: Why Every Structural Thesis Needs a Stress Test

The cost curve above looks compelling. Before we proceed to the equities, we need to stress-test it against the only real-world precedent: the 2010–2014 cycle that killed Molycorp and nearly killed Lynas.

Chinese fishing trawler collides with Japanese Coast Guard near the Senkakus. Undeclared embargo. Quotas slashed 72%. Dysprosium: $91 to $2,377. Neodymium: $25 to $340. Hundreds of exploration companies floated. Molycorp: $14 IPO to $79.16. Then the WTO ruled against China’s export quotas in 2014. China shifted strategy: domestic production controls, environmental enforcement, consolidation. Same outcome, legally distinct. Prices collapsed 80–95%. Molycorp: $1.5 billion spent, $1.7 billion in debt, sold for $20.5 million. Lynas survived only because Japanese strategic investors provided $250 million that private markets would not.

“This time is different” is the most dangerous phrase in investing, and I have watched analysts construct “five reasons this cycle is different” for uranium (2007), lithium (2018), cobalt (2017). The hit rate is poor: 20–30%. I am aware that virtually everyone who has been wrong also thought their cycle was the exception. So rather than argue that this time is different, I want to make the differences testable. Four structural factors distinguish 2025–2026 from 2010–2011, and each can be verified or falsified against the cost curve.

Government floor prices. The DoW $110/kg floor on MP’s magnet offtake directly addresses the kill mechanism that destroyed Molycorp. At $40/kg oxide, every non-China producer except Lynas is below cash cost. In 2011, they died. In 2026, the floor keeps MP’s magnet revenue stream alive through the trough. Critical caveat: the floor covers one revenue stream, not the company. Market-price oxide sales remain fully exposed.

Structural demand. 2011 was speculative hoarding. 2025–2026 is 20.7 million EVs, 34 GW planned offshore wind, robotics at 29% CAGR. These do not disappear when the crisis fades.

Multiple governments simultaneously. US ($7.5 billion in the “One Big Beautiful Bill Act”), Australia (A$1.65 billion to Iluka), Japan (co-funding Lynas and JAMSTEC), EU (RESourceEU, approximately €3 billion), India ($4 billion National Critical Mineral Mission).

Extraterritorial controls. The October 2025 controls demonstrated Beijing seeks to control how Chinese-origin materials are used even after leaving China. That generates sustained political will rather than post-crisis amnesia.

Against this: China took 30+ years to build what it has. The West is attempting 5–10 under far less favourable conditions. And the lesson of Molycorp is not just that the company failed. The market failed to distinguish companies that could actually process rare earths from those that merely claimed they could. The same risk exists today. Even if every Western project executes on schedule (a historically generous assumption), meaningful non-China supply does not arrive until 2028–2030. Iluka is instructive: most government backing, existing brownfield site, most mining-friendly jurisdiction on earth. Original target: 2025. Slipped to 2027. Capex: A$1.2 billion to A$1.7–1.8 billion, a 50% overrun.

The cost curve says the thesis survives at $80+ and dies below $50. The 2011 precedent says the kill mechanism is partially, not fully, addressed. The honest assessment: this time is genuinely different in ways that matter (government floors, structural demand, multi-government coordination), but not so different that a 12–18 month flood-and-destroy campaign cannot inflict 50%+ equity drawdowns before the structural thesis reasserts itself.

That distinction, between “the thesis is right” and “the trade works on your timeline,” is what determines position sizing. We will return to it.

Three Waves in Eight Months

Between April and November 2025, China escalated rare earth export controls across three waves, ultimately covering 12 of 17 rare earth elements, imposing 45-day licensing on all forms (oxides, metals, alloys, magnets, and equipment), and asserting extraterritorial jurisdiction over foreign-made products containing Chinese-origin rare earth content. The November 2025 suspension following the Trump-Xi meeting in Busan froze the broadest October controls until November 10, 2026, but critically, the April regime covering seven heavy rare earth elements was not suspended. Every dysprosium and terbium shipment leaving China still requires an export license today. The legal architecture, enforcement mechanisms, and institutional capacity are fully constructed and held in abeyance on a one-year timer.

The February 2026 targeting of 20 Japanese entities confirms the trajectory: the administrative capacity to administer granular, entity-level controls is growing, not shrinking.

I treat some version of reimposition as the base case. The evidence is directional rather than statistically independent: MOFCOM has imposed or tightened critical mineral export controls across six categories since 2023 (gallium, germanium, antimony, graphite, rare earths twice) and reversed zero. These are correlated policy actions within the same trade-war escalation, not independent coin flips, so “six for six” overstates the statistical confidence. But the direction is unambiguous: the institutional regime is expanding, not contracting. The case for easing rests on a comprehensive trade deal that includes rare earth provisions. The Phase One deal included them. They were never enforced.

Paradoxically, China shipped a record 62,585 MT of rare earths in 2025, up 12.9% year-over-year. Controls target high-value heavy rare earths and magnets, not bulk light rare earths. Volume is up, but the materials that actually matter are constrained. The headline number is meaningless. It is the composition that matters.

The Inventory Buffer Nobody Quantifies

Every rare earth bull thesis I have read waves a hand at stockpiles and moves on. In a squeeze thesis, the difference between six months of buffer and two years is the difference between making money and being early, which is the same thing as being wrong.

Japan’s JOGMEC maintains strategic reserves estimated at 60–90 days of national consumption across NdPr, Dy, and Tb: 1,300–2,500 tonnes in government vaults. The US DLA National Defense Stockpile was largely emptied in the post-Cold War drawdown. The 2025 Modernization Act authorised $1 billion in new acquisitions, but try buying rare earth oxides at scale when you are competing for the same constrained supply everyone else needs. South Korea has a 60-day reserve. The EU announced plans. Plans are not stockpiles.

Working backwards from non-China NdPr consumption of ~55,000–60,000 tonnes and assuming 3–5 months of average pipeline inventory, the aggregate non-China buffer is probably 15,000–25,000 tonnes. Non-China NdPr separation is ~10,000–11,000 tonnes/year against demand of 55,000+ tonnes. The annual deficit sourced from China or drawn from inventory: 44,000+ tonnes. Those 15,000–25,000 tonnes cover 4–7 months. Under current partial restrictions, maybe 6–12 months of runway. Under a full embargo: months, not years.

The wild card: China’s State Reserve Bureau holds its own strategic inventories. Nobody outside Beijing knows how much (estimates range from 30,000 to 80,000 tonnes of mixed REO). Beijing has historically dumped state reserves onto the domestic market to cap prices, but those reserves cannot be exported under current controls.

The Pull-Forward Problem: Precautionary Stockpiling and the 2028 Destock Risk

There is a dynamic in rare earth markets right now that benefits the near-term bull case and undermines the medium-term one, and almost nobody is modelling it explicitly.

When MOFCOM imposed export licensing in April 2025, every sophisticated buyer of rare earth materials did exactly what you would expect: they started building safety stock. Nikkei Asia reported in mid-2025 that major Japanese manufacturers had moved from just-in-time procurement to 6–12 months of rare earth inventory. European automakers and wind turbine OEMs have reportedly done the same, though corporate disclosures are less granular.

The maths matter. If the 20 largest non-China NdPr consumers (Toyota, Volkswagen, Siemens Gamesa, General Motors, Stellantis, Lockheed Martin, and their equivalents) each built an additional 3–6 months of safety stock through 2025–2026, that represents roughly 7,000–15,000 tonnes of NdPr pulled forward from future demand into current demand. On a market where the annual non-China supply deficit is already ~44,000 tonnes, an additional 7,000–15,000 tonnes of precautionary buying explains a meaningful portion of the price spike from $55/kg to $100/kg.

This is bullish for 2026. It is a risk for 2028–2029.

Here is why. Precautionary stockpiling is a one-time demand event. Once Toyota has 12 months of NdPr inventory rather than 3 months, it does not need to build another 9 months next year. The restocking demand disappears. If that demand disappearance coincides with (a) new Western separation capacity coming online (Iluka targeting 2027, MP heavy separation mid-2026, Energy Fuels Phase 2), and (b) potential diplomatic easing of export controls, the 2028–2029 pricing environment could be materially softer than the 2026–2027 environment even if the structural deficit thesis remains intact.

The precedent is instructive. In 2011–2012, panic buying inflated rare earth demand by an estimated 15–25% above underlying consumption. When prices peaked and inventories were full, demand collapsed not to trend, but below trend, as destocking compounded the price decline. NdPr fell from $340/kg to under $40/kg in 18 months. The structural deficit was real (it reasserted itself by 2020), but the timing destruction was total.

I am not calling for a repeat of that magnitude, and the structural differences I outlined above are real. The 2011 spike was driven by speculative hoarding by entities with no underlying consumption need, a different dynamic from industrial safety stock. And the current demand base (20+ million EVs annually) is structurally higher than the 2011 equivalent. But the directional risk is the same: today’s prices include a stockpiling premium that will eventually unwind.

The implication for positioning: the bull thesis for 2026–2027 is strengthened by stockpiling (prices may overshoot because hoarding adds to real demand). The bull thesis for 2028–2030 is weakened by it (prices may undershoot because destocking subtracts from real demand). If you are sizing a position for a 2–3 year hold, the destock risk is the most important variable you are not pricing, and it argues for taking some profit if NdPr exceeds $120/kg rather than holding for the structural target.

For MP specifically, the DoW $110/kg floor on magnet offtake provides a partial hedge: even if spot oxide prices correct during a destock phase, the magnet revenue stream remains protected. But oxide sales (which currently represent the majority of MP’s revenue) are fully exposed to spot pricing. A 30–40% oxide price correction during a 2028–2029 destock, while magnet revenue is floor-protected, produces a company that is surviving but not thriving, and an equity that could easily retrace to $35–45 before the structural deficit reasserts itself.

This is the scenario where position sizing matters more than the thesis.

Demand: Consensus but With a Ceiling Nobody Models

I will not walk you through the demand drivers. You know them. EVs, offshore wind, defense, robotics. NdPr demand is growing at 8–11% annually. Even at 5% CAGR, the non-China supply deficit exceeds 40,000 tonnes by 2028. At 11%, it exceeds 53,000 tonnes. The deficit exists under any reasonable demand scenario, and that is why I am spending 80% of this piece on supply.

But there are two things about demand that are not consensus and that matter for sizing.

First, the substitution ceiling is real but narrow. Tesla’s rare-earth-free motor, BMW’s wound-rotor iX3, IDTechEx’s projection of ~30% rare-earth-free EV traction motors by 2034: these target one application. You cannot build a rare-earth-free precision-guided munition, a competitive direct-drive offshore turbine, or high-density robotics actuators without NdFeB. The magnet is being substituted out of the one end market where volumes are largest, while remaining irreplaceable in the four where strategic urgency is highest. That means the demand story gradually shifts from “EV volume play” to “defence and infrastructure premium play,” and the pricing dynamics of those markets are very different. Defence procurement pays whatever it costs. EV OEMs do not.

Second, the elasticity threshold is closer than bulls acknowledge. At $30–60/kg NdPr, demand is genuinely inelastic. Above $150, history shows it becomes highly elastic, rapidly. Toyota redesigned the Prius motor after 2011, demand fell roughly 15%, and manufacturers permanently redesigned thousands of products to use less or no rare earths. At $90–110, we are in the sweet spot. Above $120, you are simultaneously making money on your position and funding the R&D that eventually kills it. The thesis works best at $80–120, and that is the range I model around.

The Supply Variables Nobody Covers

Japan has been managing this risk for 15 years, not 15 months. Proterial invented sintered NdFeB magnets in 1984. JOGMEC co-funded Lynas’s survival with $250 million through Sojitz. Japanese manufacturers produce ~10,000–15,000 tonnes of NdFeB magnets annually, roughly half of non-China output. Japan’s recycling infrastructure is a decade ahead of North America. The February 2026 MOFCOM action targeting 20 Japanese entities underscores that Beijing views Japan’s position as the most immediate competitive threat.

I covered Myanmar’s supply disruption earlier. The detail that matters here is what it reveals about China: you do not import from politically unstable conflict zones through smuggling networks because your domestic supply is plentiful. China’s increasing dependence on Myanmar feedstock is itself the proof that Jiangxi’s ionic clay deposits are insufficient.

China’s own depletion problem. Jiangxi’s ionic clay deposits have been strip-mined for thirty years, much of it illegal. The Chinese Society of Rare Earths has published assessments suggesting these reserves could be substantially depleted within 15–25 years at current extraction rates. Beijing faces an estimated RMB 38 billion ($5.2 billion) in Jiangxi remediation costs alone. The bear case, that China floods the market, gets weaker the longer you extend the timeline.

Deep-sea mining wildcard. Japan’s JAMSTEC has identified rare earth-rich mud near Minamitorishima containing an estimated 16 million tonnes of REO, with heavy rare earth concentrations 3–5x Chinese ionic clays. Commercial production: 2032–2038. Too late for the current thesis. For the longer-term structural story, the supply variable most likely to blow up the permanent deficit assumption.

Substitution and Recycling (Taken Seriously)

I covered the substitution ceiling in the demand section: narrow, real, confined to EV traction motors. Here I want to focus on the two supply-side responses that bulls dismiss too quickly. Iron nitride is the most watched alternative. Niron Magnetics, $300+ million deployed.

It has achieved only ~10 MGOe versus NdFeB’s 40–55 MGOe. Expert consensus: high-potential complement, not near-term replacement. Grain boundary diffusion has already reduced heavy rare earth usage by 40–70% at industrial scale, partially pricing in dysprosium substitution risk.

Recycling deserves more attention than bulls give it, and this is where I diverge from consensus. Hydrogen decrepitation recovers NdPr alloy at ~$40–70/kg, below most Western mining projects on the cost curve. The constraint is collection, not economics. Global end-of-life NdFeB availability is ~15,000–20,000 tonnes/year but collection rates are below 5%. This number will move faster than most bull models project: the EU End-of-Life Vehicles Regulation mandates rare earth recovery and labelling for new vehicles, and if automakers replicate Apple’s closed-loop model for EV magnets, collection rates could reach 15–25% by 2030, recovering 3,000–5,000 tonnes of NdPr. Not enough to close the gap alone, but combined with substitution, capable of narrowing the deficit faster than bull models assume.

Bottom line: NdFeB irreplaceable for highest-performance applications through at least 2035. The 2026–2030 window is what matters for the trade.

The Widening Gap: Why Catching Up May Mean Falling Further Behind

This is the bear case that keeps me up at night, and it survives even if every other bull catalyst fires.

While the West debates permitting timelines and celebrates first-batch commissioning, China’s magnet manufacturers are compounding a technology lead measured in process generations, not percentage points. Grain boundary diffusion (GBD), the technique that concentrates expensive heavy rare earths at grain surfaces rather than wasting them throughout the magnet body, reduces dysprosium usage by 40–70% while matching or exceeding the coercivity of conventionally produced magnets. Chinese manufacturers have adopted GBD across an estimated 40–60% of high-performance sintered NdFeB production. Japanese manufacturers (Proterial, Shin-Etsu, TDK) sit at roughly 15–25%. Western manufacturers outside Japan are at near-zero commercial adoption.

This is not a static snapshot. It is a trajectory. And it points in the wrong direction.

China’s leading magnet producers, JL MAG, Zhenghai, Ningbo Yunsheng, Zhong Ke San Huan, collectively produce over 130,000 tonnes of NdFeB annually and are scaling capacity further. They operate at the frontier of cerium-substituted formulations (replacing scarce neodymium with abundant cerium while maintaining acceptable performance), hot-deformed magnet processes, and additive manufacturing techniques for complex geometries. The latest academic literature on GBD, overwhelmingly authored by Chinese institutions, is advancing pressure-assisted diffusion, Cu-modified grain boundary engineering, and novel diffusion sources that promise further heavy rare earth reduction. These are not laboratory curiosities. The research-to-production pipeline in Chinese magnetics runs 12–18 months, compared to 3–5 years in the West, because the same state-linked conglomerates fund both the research and the factory lines.

MP Materials’ Fort Worth facility targets 1,000 tonnes of NdFeB magnets per year at initial scale. That is less than 1% of China’s current annual output. The 10X expansion targets 10,000 tonnes by the late 2020s. Even at full buildout, MP would produce roughly 7–8% of China’s projected 2028 output. The question is not whether these magnets will find buyers. The DoW price protection and Apple’s $500 million offtake guarantee that. The question is whether Western magnets will be competitive on merit, or whether the friendshoring premium is a permanent tax rather than a temporary bridge.

Consider the political half-life of that premium. In 2026, with Chinese export controls fresh and geopolitical tensions elevated, the appetite to pay 30–40% more for an American or Australian magnet is robust. By 2031, if Chinese manufacturers have moved another process generation ahead, the cost differential may be wider, the performance gap may be real, and the political coalition supporting price floors will face industrial lobbying from OEMs whose competitors source cheaper Chinese magnets through third countries. The Phase One trade deal included rare earth provisions. They were never enforced. Policy support has a shelf life, and the history of Western industrial policy suggests that shelf life is shorter than the timeline required to reach process parity.

The strongest version of this argument: the friendshoring premium creates a revenue stream sufficient to sustain Western producers at sub-scale volumes, but insufficient to fund the R&D intensity needed to close the technology gap. Western companies become permanent wards of defense procurement rather than commercial competitors. The magnets go into F-35s and Abrams tanks, not into the mass-market EV motors and wind turbines where volume economics determine long-term viability. In that scenario, the equities re-rate to defense contractor multiples (8–12x EBITDA), not growth multiples (20x+), and the premium that investors pay today for the “strategic asset” narrative evaporates.

I do not think this is the base case. Japanese manufacturers have demonstrated that non-China process excellence is achievable, and the Proterial/Shin-Etsu ecosystem provides a template. But I assign this scenario roughly 20–25% probability over five years, and it is the bear case most likely to be underweighted by investors who conflate “strategically necessary” with “commercially competitive.” Those are not the same thing.

The Bear Case, and I Mean It

China floods the market. The kill switch from 2011. If the CPC decides destroying Western investments exceeds the loss of selling at $40, they can sustain it for years. Strategically less likely this time (the 2025 consolidation enables strategic pricing, depletion dynamics constrain heavy rare earth flooding), but “less likely” is not “impossible.” Light rare earths are nowhere near depletion, and a light REE flood alone is enough to crush equities.

Beijing is not a unitary actor. MIIT controls production quotas, MOFCOM controls export licensing, NDRC sets industrial policy. Provincial governments resist centralisation. These institutional seams create scenarios a “Beijing decides” model misses.

The November 2026 cliff. If controls are renewed permanently, urgency compresses. But the friendshoring premium is a structural tax on Western industry, and powerful lobbies are working to eliminate it. Reuters reported in early 2026 that the Trump administration may be reconsidering price-floor support for domestic critical mineral projects. Even committed policy can shift.

Recession. A severe downturn reduces EV sales, slows wind, compresses defense budgets. Equities fall 40–50% even with the thesis intact.

The competitive gap may be widening. I have given this its own section above because it is the risk I find most intellectually challenging. The short version: Western producers may become permanent wards of defense procurement rather than commercial competitors, and the strategic premium investors pay today may not survive contact with that reality over a five-year horizon.

None of these individually invalidate the thesis within three to five years. But they interact. High prices incentivise substitution and recycling simultaneously. A recession arriving alongside those responses could close the gap earlier. I assign roughly 25–30% to this combined scenario, higher than most bulls acknowledge.

Scenario Architecture

I want to be explicit about how I weight outcomes, and honest about the limits of that exercise. The reimposition case is laid out above. Here is how I translate it into probabilities:

Note these weights sum to approximately 100% with overlap at boundaries, which is honest. Precise point estimates would be false precision. But ranges are better than “heaviest” and “tail risk,” which tell the reader nothing about how you are actually sizing the position.

Probability-weighted, those scenarios produce an expected MP equity value of roughly $48–62/share. At the current mid-$50s, you are paying approximately fair value on a blended basis. The upside skew is real (structural repricing delivers $70–100+), but so is the downside (flood-and-destroy or diplomatic resolution delivers $20–35). This is not a screaming buy at current levels. It is a correctly priced option on structural repricing, with the optionality most valuable if you can add on a drawdown to $35–40.

These scenarios are not mutually exclusive in time. That is the critical point most scenario analysis misses.

A flood-and-destroy phase lasting 12–18 months could precede structural repricing if Western governments backstop producers through the trough. The 2011–2025 sequence is precisely this pattern: prices crashed, Molycorp died, Lynas survived on Japanese capital, and the structural deficit reasserted itself over the following decade. Anyone who bought Lynas at the 2015 trough and held through 2025 made 20x. Anyone who bought at the 2011 peak and sold during the crash lost 90%.

The trade is not the thesis. The thesis says the processing gap does not close on any timeline shorter than three years. The trade says: given that thesis, what position can I hold through the drawdown I know is coming, and still be standing when the repricing arrives? Position sizing is the actual investment decision. Everything else is context.

The Equity: MP Materials (NYSE: MP)

Two companies in the non-China landscape have the scale, processing capability, and government backing to matter: MP Materials and Lynas Rare Earths. Both are effectively extensions of Western industrial policy, which is simultaneously their greatest strength and their most underappreciated risk. I am going deeper on MP because it has more moving parts, more embedded optionality, more ways the market is mispricing the capital structure, and the DoW relationship that creates a valuation floor no other name in the sector has.

What It Is

The only fully integrated US rare earth producer. Mountain Pass mine in California (highest-grade large-scale rare earth deposit in the Western Hemisphere). Independence magnet facility in Fort Worth. First magnets on commercial equipment Q4 2025. FY2025 NdPr oxide: 2,599 MT, up 101% YoY, exiting at an annualised run rate of nearly 4,000 MT.

Critical context most coverage omits: until mid-2025, MP shipped concentrate to Shenghe Resources in China. America’s flagship rare earth independence play was, until 18 months ago, sending its ore across the Pacific to be processed by the country it is supposed to be independent from. Let that sit for a moment. Processing track record: measured in quarters, not decades. This is a mining company that became a processing company and is attempting to become a magnet company. Each step is harder than the last, and each requires capabilities that are qualitatively different from the one before.

The Financial Reality

FY2025 consolidated revenue: $224.4 million, up 10%. Magnetics segment: $66.9 million, from zero. Full-year adjusted EBITDA was $11.4 million, but Q4 alone was $39.2 million, meaning Q1–Q3 combined were still negative $27.8 million. The improvement is real but heavily back-loaded and substantially driven by the DoW Price Protection Agreement (commenced October 1, 2025), which pays MP the difference between $110/kg and the benchmark quarterly average for NdPr products. At current market prices below $110/kg, this is functionally a government subsidy. It is a genuine floor. It is not organic profitability. Q4 EPS $0.09 beat consensus of $0.02 by 350%. Cash: $1.17 billion.

The Balance Sheet Question Nobody Is Asking

Guided 2026 capex: $500–600 million. The 10X facility, heavy separation, and mine expansion collectively require north of $1.5 billion through 2028. Cash covers roughly 18–24 months at current burn. DoW convertible preferred and warrants represent approximately 15% dilution at conversion. If MP raises additional equity (which I assign 40–50% probability before 2028), total dilution could reach 20–25%.

Here is where the reflexivity lives, and where most coverage stops too early. At $58/share, a capital raise is manageable: modest dilution, strong institutional appetite, strategic premium on the equity. At $40/share, triggered by a Chinese price flood or a broader market selloff, the raise becomes more dilutive and simultaneously more likely. The company needs capital to complete the buildout. If the equity price falls, each dollar of capital costs more shares. That reflexivity, the interaction between equity price, capital needs, and dilution, is what killed Molycorp. The DoW floor partially breaks the reflexive loop by guaranteeing a revenue stream that supports debt capacity even at depressed equity prices. Partially. Not fully.

Valuation Framework

The honest version of this exercise starts with what the market already believes, and then asks whether you agree.

What the Street thinks. Fifteen analysts cover MP with target prices. Consensus estimates for FY2028: revenue of $1,052M and EBITDA of $572M, implying a 54% EBITDA margin. At current enterprise value of $9,964M, that prices MP at 17.4x 2028E EV/EBITDA. The ramp is steep: consensus has 2026E EBITDA at $173M (57.6x), 2027E at $439M (22.7x), then $572M in 2028. Revenue is expected to nearly quadruple between 2026 and 2028, from $396M to $1,052M. These consensus numbers embed aggressive assumptions: NdPr prices sustained above $100/kg, the 10X facility commissioned on schedule in 2028, magnet production scaling well beyond initial 1,000 MT/yr targets, and heavy rare earth separation contributing meaningful revenue by late 2026. In the structural bull scenario (controls reimposed, NdPr $100–130, full execution), these numbers are achievable. In any other scenario, they are not.

What my model suggests. At $90/kg NdPr oxide sustained through 2028, MP’s combined revenue from oxide sales (4,000–5,000 MT at realised ex-China pricing), magnet manufacturing (scaling toward 1,000 MT/year with $110/kg DoW floor and Apple offtake), and heavy rare earth separation (targeting Dy and Tb production by late 2026) suggests 2028 revenue of $450–550M with 25–35% EBITDA margins. That implies 2028 EBITDA of $112–192M.

The gap between my estimate and the Street ($112–192M vs $572M) is not a modelling error. It reflects a scenario disagreement. The Street is pricing the structural bull. I am pricing the muddle-through. The difference is driven almost entirely by two assumptions: (1) NdPr pricing ($90/kg in my model vs $110–130 implied by consensus), and (2) 10X ramp timing (partial contribution by 2028 in mine vs full nameplate in consensus). If you believe reimposition is the base case and 10X executes on time, the Street is right and MP is cheap at 17.4x 2028 EBITDA. If you think muddle-through is more likely, the stock is pricing in a scenario that has roughly 35–40% probability.

The comps context. Here is every non-China rare earth and strategic mineral company that matters, from the same Capital IQ terminal, on the same date:

The table tells a story if you read it right.

On FY2027E multiples, MP (22.7x) and Lynas (19.1x) are converging. The 3–4 turn premium MP commands reflects US-jurisdiction strategic value, the DoW relationship, and magnet optionality. That premium is not unreasonable. But Lynas’s consensus price target of A$11.68 sits below the current A$12.86, meaning the analysts who cover the closest comp think the sector has already overshot. That is a yellow flag for the entire rare earth group.

Cameco at 26.3x FY2027E is the best analogue for what MP aspires to become: a strategic commodity producer with government backstop, structural demand tailwind, and geopolitical scarcity premium. Cameco has sustained this premium for years and its 17 analysts rate it 1.68 (”strong Outperform”). If MP de-risks execution through 2027, a Cameco-level multiple is the upside case. MP at 22.7x already sits within range.

Albemarle at 10.1x is the gravity well. This is what a de-risked, producing strategic mineral company trades at when the scarcity premium fades. The difference between Albemarle’s multiple and Cameco’s (10x vs 26x) on $439M 2027E EBITDA translates to roughly $25/share vs $65/share for MP. That range maps almost exactly onto the 52-week low and the current price.

Iluka is the cautionary tale. A$2.7B TEV, loss-making through FY2027, A$1.65B government loan, 50% capex overrun, and analysts barely targeting above the current price. This is what happens when “strategic asset with government backing” meets execution reality. Every capex timeline in the piece, including MP’s 10X facility, should be read through the Iluka lens.

Energy Fuels at C$4.4B TEV with negative EBITDA through 2026 and barely breakeven in 2027 is trading on pure optionality, priced entirely on management’s $3.7B NPV claims for projects that have not reached final investment decision. Only four analysts cover it. This is the speculative end of the sector.

Short interest. MP’s short interest stands at 24.8 million shares, roughly 14% of float. That is high for a name of this size and tells you sophisticated money disagrees with the bull case. It also creates squeeze dynamics in both directions: a positive catalyst (reimposition, 10X acceleration) compresses shorts violently upward, while a negative catalyst (trade deal, execution miss) can cascade downward as longs capitulate into existing short pressure.

The dilution question. Guided 2026 capex is $500–600M (consensus: $551M), against guided free cash flow of negative $407M. Cash of $1.17B covers roughly 18–24 months at current burn. Average diluted shares in consensus estimates ramp from 195.5M in Q1 2026 to 206M by Q4 2026, reflecting the DoW convertible preferred and warrants at approximately 16% dilution versus the 177.7M basic shares outstanding. I assign 40–50% probability of an additional equity raise before 2028. At $58/share, that is manageable. At $40, the reflexivity loop I described above kicks in. The maths do not need repeating.

The entry question. At today’s mid-$50s, you are paying 22.7x FY2027E EV/EBITDA, a 3–4 turn premium over Lynas and slightly below Cameco. It requires conviction that the consensus ramp materialises. My own model’s low case ($112M EBITDA in 2028) produces 89x, which does not support the current stock price under any reasonable multiple framework.

At $35–40 (a level the stock visited as recently as early 2025), the risk/reward is asymmetric in your favour. That is the entry zone. At mid-$50s, you are paying roughly probability-weighted fair value: not expensive, not cheap, and vulnerable to any single negative catalyst. At $100 (the October 2025 peak), you are paying for the structural bull plus execution perfection plus no recession. Express that view through options, not equity.

52-week range: $18.64–$100.25. If you cannot stomach a 40–50% drawdown while holding, you should not own this. And if you are reading this thinking “but the thesis is so compelling that I will size up,” go re-read the Molycorp section.

Key Catalysts

MP reports Q1 in late April/May (10X construction update, heavy separation commissioning). Heavy rare earth separation expected mid-2026, with separated Dy and Tb by late 2026. 10X facility targeting 2028 commissioning. November 2026 MOFCOM expiry is the binary macro event.

For completeness, the one-line thesis on each name beyond MP. Lynas is the safer play with less upside: it already does the hardest part, Mt Weld is the world’s lowest-cost NdPr producer, and the mine life exceeds 20 years, but Kalgoorlie ramp execution, the LAMP licence in Malaysia, and Seadrift stalling all weigh on the near-term. At A$11-14, repriced from perfection to reality. Energy Fuels is genuinely differentiated by its dual uranium/rare earth revenue stream and was up roughly 243% in 2025, but Phase 2 scale-up execution is the question and management’s NPV estimates ($3.7B combined) are among the most optimistic figures in mining finance. Iluka is the cautionary tale for all capex timelines in this sector: A$1.65B government loan, brownfield site, most mining-friendly jurisdiction on earth, and still a 50% capex overrun with the timeline slipping from 2025 to 2027. Vulcan Elements is pre-revenue, pre-production, and trading on policy proximity as the second DoW-backed magnet manufacturer; high-risk, high-reward if the Department of Defense relationship matures.

The magnet manufacturing race is where equity creation happens. Miners without integration are selling rock to someone else’s chain, the same value trap 2011 created.

Five Numbers That Determine How This Resolves

1. MOFCOM export license approval rates. The real-time indicator. The February 2026 action against 20 Japanese entities is the latest signal.

2. MP 10X commissioning timeline. Targeting 2028. Heavy separation expected mid-2026, with separated Dy and Tb by late 2026.

3. Ex-China NdPr oxide spot. Above $100 sustained: substitution R&D explodes. Below $80: next wave of Western projects becomes uneconomic and the gap widens. As of early 2026: $90–100/kg and rising.

4. November 2026 expiry. Binary event. Size accordingly. The market starts pricing it by mid-2026.

5. Global EV trajectory. Deceleration weakens demand side. Acceleration widens the gap. BMI’s 22.9 million units for 2026 is the near-term benchmark.

Near-term catalyst calendar. MP reports Q1 in late April/May (10X construction, heavy separation commissioning). Lynas reports FY2026 in August (Kalgoorlie ramp). Energy Fuels quarterly (Phase 1 revenue, Phase 2 decisions). MOFCOM monthly export data (~6 week lag). Myanmar enforcement: any concrete action spikes heavy rare earth prices within days.

Where This Leaves Us

Let me be direct about what I believe.

The rare earth processing gap is the most underpriced structural risk in global commodity markets. China will, on some timeline between now and November 2026, reimpose or tighten export controls, because the institutional architecture is built, the geopolitical incentives are aligned, and the track record is six escalations and zero reversals. MP Materials is the highest-optionality vehicle in the Western rare earth supply chain, and the DoW floor creates a survivability buffer that did not exist in 2011.

I also believe the current equity price does not adequately compensate for execution risk, the 2028 destock scenario, or the possibility that Western producers become permanent wards of defense procurement rather than commercial competitors. Consensus EBITDA estimates embed assumptions I weight at 35–40% probability. And the timeline is less forgiving than most bulls assume: the pull-forward dynamic that is bullish for 2026–2027 is bearish for 2028–2029, and anyone holding for the “long-term structural thesis” needs to define what happens to their position during the intervening valley.

Two numbers above all others. Ex-China NdPr oxide sustained above $100/kg through Q3 2026: the thesis is working. The November 2026 MOFCOM expiry: the binary event around which everything else orbits.

The people who profit from structural commodity theses are never the ones who identified them first. They are the ones who sized positions to survive the drawdowns, understood that being right about the thesis guarantees nothing about the next three months, and stayed positioned through the volatility that shakes everyone else out.

The data is public. The interpretation is mine. What you do with it is your problem.

If you found this useful, the single most valuable thing you can do is send it to someone who allocates capital. This kind of analysis compounds when it circulates.

Share

Share TSCS

This publication is for informational and educational purposes only. It is not investment advice, and nothing in this piece constitutes a recommendation to buy, sell, or hold any security. The author may hold positions in securities discussed. Do your own work. You are responsible for your own decisions.

That was not a speculative event. It was a physical one. The distinction matters more than anything else in this piece.

The silver market has been in structural deficit for five consecutive years. The cumulative shortfall, roughly 800 million ounces, has been drawn from a thin layer of deliverable inventory that multiple independent analysts now describe as critically low. COMEX registered stocks breached 100 million ounces in February for the first time in the modern era. Shanghai inventories have collapsed 86% from peak. And the mechanism that is supposed to fix this, higher prices incentivising more supply, is structurally broken, because 70% of the world’s mined silver is produced as a byproduct of other metals whose miners do not care what silver is worth.

I have spent the last several months pulling apart every layer of this market: the supply-demand arithmetic, the physical inventory data, the thrifting rates in solar manufacturing, the analog markets that tell you how commodity squeezes play out, and the policy shifts in Washington and Beijing that have reclassified silver from a commodity to a strategic asset. What I found is a market that is more fundamentally broken than the price action, as dramatic as it has been, suggests. If you read my January analysis, which modelled the solar thrifting question with explicit probability weightings across four scenarios, this builds on it. If you did not, this is self-contained.

I need to say something about intellectual honesty, because most silver analysis has none. The bulls dismiss thrifting and substitution. The bears dismiss deployment scale and the byproduct trap. Both are making the same error: treating a dynamic, nonlinear system as though one variable will dominate. I have tried, throughout what follows, to present the strongest version of every argument, including the ones that make me uncomfortable. The bear case in this piece is not a disclaimer. It is the section I would read twice.

This is Part I of a two-part series. Part II, a collaboration with , unpacks the broader architecture: the commodity supercycle, the monetary regime shift, gold’s structural repricing, and how to express the thesis across precious metals equities and commodities. If this piece is the micro, that one is the macro.

If this is the kind of analysis you want in your inbox, subscribe to TSCS. Part II is coming.

Subscribe now

The Arithmetic: Five Years of Deficits Nobody Talks About

The silver market recorded a structural supply deficit of 148.9 million ounces (Moz) in 2024. That was the fourth consecutive year of deficits, following shortfalls in 2021, 2022, and 2023. The 2025 data, now confirmed by the Silver Institute and Metals Focus at their November industry dinner, marks a fifth consecutive deficit year, bringing the cumulative shortfall for 2021 through 2025 to approximately 775 to 820 million ounces, depending on which 2025 estimate is used (the Silver Institute’s November 2025 revision put the year’s deficit at 95 Moz, down from an earlier interim estimate of 118 Moz).

The Silver Institute estimates total above-ground silver at roughly 19.3 billion ounces, but the vast majority is locked in jewelry, religious artifacts, industrial products, and forms that will never return to the market at any price. The Institute’s own February 2025 report concluded that no correlation exists between overall above-ground stocks and the silver price. Only bullion stocks matter.

After removing ETF-encumbered metal (over 1 billion ounces in ETFs globally as of mid-2025, with SLV alone holding approximately 495 Moz at peak and global silver ETP holdings reaching approximately 1.13 billion ounces), exchange inventories held for specific clients, and metal in transit or in industrial pipelines, the true “free float,” the silver actually available for purchase and delivery, is difficult to pin down precisely but likely falls in the range of 50 to 150 Moz, depending on assumptions about encumbrance and pipeline commitments. Against annual demand exceeding 1.1 billion ounces. The 800 Moz cumulative deficit came out of this thin layer of liquid, deliverable metal, not from the 19 billion ounce headline. Section IV details exactly how thin it has become.

Here is the supply-demand snapshot:

Source: Silver Institute, World Silver Survey 2025, Metals Focus

A necessary caveat on the deficit figures: “total demand” as reported by the Silver Institute includes jewelry (approximately 200 Moz annually) and silverware (approximately 50 Moz), alongside industrial fabrication, bar and coin investment, and other categories. These are not economically equivalent. Industrial demand for semiconductor packaging or solar cells is highly price-inelastic; a chip fabricator in Hsinchu does not stop buying silver paste because the price doubled. Jewelry demand, by contrast, is significantly price-elastic, particularly in Western markets where silver jewelry is a discretionary consumer purchase. Indian jewelry demand is stickier (silver carries cultural and quasi-monetary significance), but even there, demand responds to price at the margin.

This matters for interpreting the deficit. If the reported 148.9 Moz deficit in 2024 were decomposed, some portion reflects jewelry and silverware demand that would contract at sustained high prices, partially self-correcting the imbalance. A conservative estimate: at $90+ silver sustained for two or more years, jewelry demand could decline 15% to 25% (30 to 50 Moz), which would narrow the deficit meaningfully but not eliminate it. The structural core of the deficit, the gap between industrial demand and total supply, would persist even if every discretionary ounce of jewelry demand disappeared. That is the number to focus on, and it remains firmly negative.

Now, 2025’s deficit narrowed to roughly 95 Moz according to the Silver Institute’s November 2025 revision (down from the April 2025 interim estimate of 118 Moz), and the silver permabulls will not tell you this because it complicates their narrative. Total demand pulled back approximately 4% year-on-year. Industrial demand dipped about 2% to 665 Moz (still the second-highest on record), and bar and coin demand fell to a seven-year low of 182 Moz.

The US market, in a display of almost poetic timing, saw substantial retail liquidation throughout much of 2025, even as the price was ripping higher. American retail investors collectively decided the best time to sell was during a structural repricing event.

But the deficit persisted. Even with demand softening, supply could not close the gap. The silver market is not in deficit because of a temporary demand surge. It is in deficit because supply is structurally incapable of meeting even moderately elevated demand.

The Byproduct Trap: Why Silver Cannot Fix Itself

This is the section that separates people who understand silver from people who merely own it. It is also, I believe, the single most important structural argument in commodity markets today, because it represents a first-principles violation of how commodity self-correction is supposed to work.

Approximately 70% to 75% of the world’s mined silver is produced as a byproduct of mining other metals, primarily copper, lead, zinc, and gold. Only about 25% to 28% comes from primary silver mines. In 2024, silver from lead/zinc operations remained the single largest source category at 29.4% of total production, followed by copper byproduct and gold byproduct.

Think about what this means operationally. When the silver price doubles, the rational economic response would be for miners to increase silver production. That is how commodity markets are supposed to self-correct. High prices incentivise supply. Supply meets demand. Equilibrium is restored.

Except in silver, this mechanism is broken.

If you need more silver, you need to convince copper miners, lead miners, zinc miners, and gold miners to increase their production, because that is where the silver comes from. And those miners make their capital allocation decisions based on the economics of their primary metal, not silver. A copper mine does not drill more holes because the silver price went from $25 to $90. It drills more holes because the copper price justifies the capex. Silver is a rounding error on their income statement.

This is the byproduct trap. Silver supply is, for practical purposes, price inelastic. The metal literally cannot respond to its own market signals.

And even the primary silver mines face constraints. Global silver mine production peaked at approximately 900 Moz in 2016. By 2024, it had fallen to 820 Moz, a 9% decline over eight years despite the price more than tripling during that period. Production in 2025 was essentially flat at 813 to 835 Moz depending on the source, with gains in Mexico and Russia offset by declines in Peru and Indonesia.

The project pipeline for new primary silver mines is thin. Development timelines from discovery to production run 7 to 15 years. Capital intensity has risen dramatically. Ore grades at existing operations are declining, requiring more rock to be processed for equivalent output. Exploration spending during silver’s extended bear market from 2011 to 2020 was negligible, and the projects that might have been discovered during that period simply do not exist.

New primary supply is unlikely to arrive before 2030 at the earliest, and even that assumes exploration spending that has not yet occurred.

The Demand Revolution: Solar, AI, and the Industrial Floor

Silver’s industrial demand hit a record 680.5 Moz in 2024, the fourth consecutive record year. Industrial fabrication now accounts for roughly 59% of total global silver consumption, making silver fundamentally different from gold in its demand profile. Gold is a monetary asset held in vaults. Silver is a monetary asset that gets soldered into circuit boards and laminated into solar cells.

The electrical and electronics sector has been the primary driver, increasing 51% since 2016. This should surprise nobody who understands silver’s physical properties: it is the most electrically conductive metal on earth. When you need electrons to flow with minimal resistance, silver is not optional. It is required.

Solar photovoltaics represent the fastest-growing demand segment. PV-specific silver consumption accounted for 17% of total demand in 2024, up from 5.6% in 2015, growing at an annualised rate of 12.6%. The transition from PERC to TOPCon and heterojunction (HJT) cell architectures initially increased per-cell silver loadings, though 2024 and 2025 saw aggressive “thrifting” (reduction of silver content per module) as manufacturers adapted to elevated prices. Global solar installations hit records in 2025, led by China’s 45% capacity expansion in 2024.

Here is where I need to be blunt, because most silver analysis is dishonest in one of two directions, and I have no interest in joining either camp.

The bulls dismiss thrifting. The bears dismiss deployment scale. Both are making the same intellectual error: treating a dynamic, nonlinear system as though one variable will dominate. The honest answer is that the outcome depends on the relative rates of change of two competing forces, and pretending you can forecast that interaction with precision is arrogance dressed up as analysis.

Let me lay out the data as honestly as I can.

The industry has achieved a roughly 6x reduction in silver per cell since 2010, from approximately 60 mg/W to roughly 10 mg/W. That is extraordinary engineering progress. The transition from older PERC cells to newer TOPCon architecture initially reversed some of those gains (TOPCon requires silver on both sides, consuming roughly 50% more per watt than PERC), but thrifting on the new architectures is proceeding aggressively. HJT cells, which peaked at 22 mg/W, have been driven down to 6.39 mg/W through copper substitution. Germany’s Fraunhofer ISE has demonstrated a lab record of just 1.4 mg/W.

And here is the number that silver bulls do not want to hear: PV silver demand actually declined approximately 5% year-on-year in 2025, despite record solar installations globally. That was the first time thrifting gains outpaced volume growth. Longi Green Energy has announced copper replacement for back-contact cells (mass production targeted for Q2 2026). Jinko Solar has signaled large-scale copper-based production. Shanghai Aiko Solar has launched what it describes as silver-free cells.

The timeline here deserves emphasis: Cu@Ag paste (silver-coated copper) adoption is projected by several industry analysts to reach 30% to 40% of global cell production by 2027, potentially reducing silver content by 50% to 80% per cell at scale. This is not a 2030 risk. It is a 2027 risk, and the R&D acceleration at $90+ silver is non-linear.

The 2011 precedent deserves a direct mention: after peaking at $49.51, silver crashed approximately 60% and did not recover to $30 for over a decade, driven by cascading margin increases. The CME raised margins five times during that run. Similar adjustments should be expected in 2026 if volatility persists.

If you are long silver purely on the solar thesis, this data should make you uncomfortable. It makes me uncomfortable.

But the picture is more complex than “thrifting will kill silver demand.” At current prices exceeding $90 per ounce, silver has become the single largest cost component in PV manufacturing. BNEF estimated silver at roughly 14% of total PV production cost at $42 to $46 per ounce in September 2025, rising to 17% at approximately $50. At $90 or above, silver’s cost share likely exceeds 25% of total panel cost. That is precisely the threshold at which substitution R&D budgets explode. But lab records and mass production are separated by years and billions of dollars of retooling. Copper-based pastes degrade faster, are less conductive, and require manufacturing process changes that cannot be implemented mid-production cycle. The 1.4 mg/W Fraunhofer result was achieved under laboratory conditions that bear no resemblance to a production line running 24 hours a day in Jiangsu province.

A Ghent University/Engie Laborelec study, which models this race between thrifting and deployment more rigorously than anything I have seen from the sell side, projects that PV silver demand could still reach 322 to 450 Moz by 2030, potentially representing 29% to 41% of projected global supply. Their most alarming finding: by 2030, total silver demand could reach 48,000 to 52,000 tonnes annually, but supply may cover only 62% to 70% of that need.

For readers who want the explicit scenario modeling on installation growth versus thrifting rates, I published a probability-weighted sensitivity table in my January analysis that runs four scenarios from bull (214 Moz by 2027) to severe bear (117 Moz).

The probability-weighted outcome supports continued deficits at narrower levels than 2022-2024.

The honest answer on thrifting is this: it is real, it is accelerating, and it will eventually cap the growth rate of solar silver demand. But “cap the growth rate” is not the same as “eliminate demand.” Even aggressive thrifting scenarios still project PV demand above 250 Moz by 2030. And the critical point is that silver’s industrial demand story extends well beyond solar.

Beyond solar, the demand vectors are multiplying:

AI infrastructure. Data centres are silver-intensive, with AI-specific facilities requiring 2 to 3 times more silver per server rack than traditional data centres. Every semiconductor package and high-frequency interconnect requires silver for its conductivity. The AI buildout is a decade-long capital expenditure cycle, and silver is embedded in its physical infrastructure.

Electric vehicles. Each EV contains approximately 25 to 50 grams of silver in its electrical systems, sensors, and wiring. As global EV production scales, this is a steady incremental demand source.

Grid infrastructure. Electrification requires grid expansion and modernisation, both of which are silver-intensive. This is a global phenomenon that will persist regardless of which political party is in power in any given country.

5G and telecommunications. Network densification and millimetre-wave infrastructure deployments require silver-containing components.

The critical point is this: industrial silver demand is structurally higher than it was a decade ago, and the applications driving it (solar, AI, EVs, grid) are all on secular growth trajectories. This is not cyclical demand that will evaporate in a recession. A recession might slow the growth rate, but it will not reverse the structural shift. The floor under silver demand is higher than it has ever been, and it is rising.

Silver Intensity: The Metric Nobody Is Tracking