The 900-Kilogram Problem
The cylinders are about twelve feet long and four feet across, and from a distance they look like enormous propane tanks. They are stacked two high, in double rows, on concrete saddles, across open yards in western Kentucky and southern Ohio.
There are roughly sixty-five thousand of them.
Together they hold more than seven hundred and thirty thousand metric tons of depleted uranium hexafluoride — the residue of fifty years of uranium enrichment at the Paducah and Portsmouth gaseous diffusion plants, the twin engines of the American atomic programme. The oldest cylinders in the family, the ones at Oak Ridge, date to the Manhattan Project.
For decades nobody knew what to do with them. Now they do.
The material is being converted into a stable oxide, packed into modified cylinders, and loaded six at a time onto rail gondolas. The gondolas run west. The destination is a licensed disposal facility in west Texas.
It is being buried.
I want you to hold that image, because I am going to spend the rest of this article explaining why it is the most useful thing I have seen in the energy sector in two years — and it has nothing to do with whether burying it is right or wrong.
It is right, mostly. Most of that inventory is genuinely spent, assaying below natural uranium, uneconomic to touch at any plausible price. Treating waste as waste is correct.
But here is what is happening on the other side of the same government, in the same year, with the same commodity.
The United States has banned imports of Russian uranium. It is granting emergency waivers to keep the fuel flowing anyway. It has committed $2.7 billion to standing up domestic enrichment from a standing start. It has directed the Department of Energy, by act of Congress, to release weapons-grade material from the national stockpile because the civilian supply chain cannot produce what its reactors need.
One arm of the state is paying to put uranium in the ground. Another is paying to find some.
That is not hypocrisy. It is the fingerprint of an industrial base that was dismantled so completely that its own operators no longer have an accurate mental model of what they possess.
And wherever you find that pattern, you find mispriced assets.
Before we continue, let me introduce you to Altis Terminal — the platform where you can follow all of our investment theses, model portfolios, and positions in real time.
Altis Terminal will be included at no additional cost for all Premium subscribers who join before the official launch on August 14. Subscribe today to secure your spot on the waiting list and get access as soon as the platform goes live.
More than 100 investment theses are available on the platform. We follow a top-down investment approach, where our macroeconomic research serves as the foundation for building and managing our portfolios.
To celebrate the launch, we’re offering you an exclusive discount that lets you save $50 on your first year of Macro Notes Premium.
Instead of paying $300 for your first year, you’ll pay just $249, with your subscription starting on August 14.
Subscribe today to lock in this launch offer and secure your access to Altis Terminal from day one.
Follow the atom
To understand why a country can sit on three quarters of a million tonnes of uranium and still have a fuel emergency, you have to follow a single kilogram through the process. There are four industrial steps between rock and reactor, and I had assumed — as I think most investors assume — that the difficulty was concentrated in one of them.
It isn’t. It’s in three.
Step one: mining. Uranium comes out of the ground in Kazakhstan, Canada, Namibia, Australia. This step is cyclical, geopolitically messy, and endlessly discussed. It is also, by a wide margin, the least constrained link in the chain. Which is precisely why it attracts the most retail money: it is the only step with dozens of listed pure-plays and a spot price on a screen.
Step two: conversion. The oxide has to become uranium hexafluoride — a gas, because you cannot centrifuge a rock. This is unglamorous fluorine chemistry, and the entire Western world depends on essentially three commercial facilities: one in Ontario, one in France, and one in Metropolis, Illinois.
That American plant was switched off between 2017 and 2023. Not sabotaged, not regulated out of existence. Idled, because the market was too weak to justify running it.
Think about what that means for a moment. The single American facility performing a mandatory step in the nuclear fuel cycle was mothballed for six years for commercial reasons, and almost nobody in finance noticed either the closure or the restart.
Step three: enrichment. The gas is spun until the U-235 concentrates. In the West this is Urenco, across four sites in four countries, and Orano in France. In the United States, exactly two entities hold enrichment licences.
Russia holds somewhere between 40 and 44 percent of global enrichment capacity, and supplied around a quarter of American enriched uranium before the ban.
Step four: deconversion and fabrication. The enriched gas has to become an oxide or a metal again, in a physical shape a reactor core can accept.
And for the specific fuel that the next generation of reactors requires, this step has no commercial capacity anywhere in the Western world. Zero. Centrus said as much on its most recent earnings call, disclosing that it is exploring a joint venture with Oklo to provide the service, and describing it plainly as a capability that does not currently exist commercially.
Four steps. One is fine. Two are oligopolies of three. One does not exist.
The fuel that isn’t there
The material at the end of that chain is called high-assay low-enriched uranium — HALEU, uranium enriched to between 5 and 19.75 percent, against the 3 to 5 percent that today’s reactors run on.
It is what TerraPower’s Natrium is designed for. And X-energy’s Xe-100. And Kairos. And Oklo. And Radiant, and Westinghouse’s eVinci. It is the fuel underneath a very large fraction of the private capital that has flowed into nuclear over the past three years.
Here is the entire Western supply.
One cascade of sixteen centrifuges in Piketon, Ohio, operated by Centrus — the first plant using American technology under American ownership to enrich uranium in roughly seventy years. Cumulative deliveries to date: a little over 920 kilograms. Production rate: 900 kilograms per year.
That is not Ohio’s output. That is not one company’s output. That is the commercial production of the United States of America.
Everything else in the system is inheritance.
Congress directed the Department of Energy to release 21 metric tons to reactor developers on a fixed schedule — three tonnes by September 2024, eight more by December 2025, ten more by June 2026. Fifteen companies applied; five were selected first. That material does not come from a factory. It is drawn down from National Nuclear Security Administration stockpiles, and the traditional way of making it is downblending: taking weapons-grade uranium and diluting it. One tonne of highly enriched material yields four to five tonnes of HALEU.
And roughly ten more tonnes are being recovered, by December 2028, from the spent fuel of Experimental Breeder Reactor-II — a research reactor at Idaho National Laboratory that operated from 1964 and shut down in 1994. The fuel goes into a molten salt bath, an electric current pulls the uranium away from the fission products, and the recovered metal is cast into ingots.
Oklo is allocated five tonnes of it.
So: America’s nuclear renaissance is currently being started on dismantled warheads and the leftovers of a reactor that closed before most of its investors had email addresses.
Against that, the demand side. Department of Energy projections put the American requirement above 40 tonnes by 2030, at roughly 50 tonnes per year by 2035, and 500 tonnes per year by 2050. The Nuclear Energy Institute’s figure is on the order of 3,000 tonnes cumulative by 2035.
Fifty tonnes a year, from a manufacturing base of nine hundred kilos.
The steel cans
There is one detail in the Centrus contract history that I keep returning to, because it tells you more about the state of this industry than any capacity forecast.
The 900-kilogram delivery ran late.
Not because the centrifuges underperformed. Not because of a regulatory hold. Under the terms of the contract, the Department of Energy was responsible for supplying the storage cylinders needed to collect the product coming off the cascade — and there were not enough of them. At one point cumulative deliveries stood at around 332 kilograms, with the programme waiting on containers.
The most advanced uranium enrichment effort in the United States was rate-limited by a shortage of steel cans.
This is what a dormant industrial base actually looks like from the inside. Not catastrophic failure. Not conspiracy. Just an unending sequence of small, boring, physical inputs that nobody thought to order eighteen months in advance, each of which costs half a year.
If you followed my work on European artillery propellant, or on grain-oriented electrical steel, you have seen this movie. The bottleneck is never where the press release says it is.
Why HALEU eats the existing fleet
Now the piece of physics that I think is the single most underpriced fact in this sector, and the reason I stopped calling this an SMR trade.
Everything in enrichment is denominated in separative work units — SWU, a measure of effort rather than mass, introduced by Paul Dirac in 1941.
Separative work scales brutally with enrichment level. Producing one kilogram of HALEU at 19.75 percent from natural uranium feed takes on the order of 42 SWU. Do it in two stages instead — starting from feed already enriched to 4.95 percent — and the second stage needs only about 6. Industry analysis suggests a plant with fixed separative capacity can produce roughly seven times more HALEU by mass using that two-stage route.
Either way, it consumes capacity. Capacity that would otherwise have fuelled the 94 reactors already operating in the United States.
The nuclear renaissance competes with the nuclear present.
Every kilogram of advanced-reactor fuel is separative work not sold to a conventional utility — in a market that has just deleted its largest supplier, and in which the existing fleet is being uprated and life-extended precisely because data centres want baseload now rather than in 2032.
Which is why the consensus framing is wrong in a specific, exploitable way. Ask almost any investor about nuclear right now and you get some version of: this is the AI power trade, and if AI capex disappoints, it corrects.
The AI story is what brought the capital in. It is not what causes the shortage.
The shortage is caused by 94 reactors that consume enrichment every single year regardless of what any startup does; by the impossibility of standing up centrifuge capacity in under five years; and by a date written into United States law.
That date is January 1, 2028 — when the waivers permitting continued imports of Russian uranium terminate.
Set it against the arrival schedule of the replacements. Urenco’s New Mexico expansion is not expected to be fully operational until around 2030. Centrus’s fixed-price contract with the Department of Energy requires delivery of one metric ton of HALEU by March 2032. General Matter’s new Paducah facility — a roughly $1.5 billion project — has been publicly described with enrichment operations arriving in the 2030s.
One tonne. By 2032.
The gap between a hard legal stop in 2028 and real capacity in the 2030s is not the bear case for this sector. It is the schedule.
The man who won’t say how
Which brings me to the last thing I found, and the reason I moved from analyst to allocator.
In January 2024, an engineer named Scott Nolan founded a company.
Nolan was an early engineer at SpaceX, where he worked on the Merlin engine systems and the Dragon capsule. He then spent thirteen years as a partner at Founders Fund leading hard-tech investing. That is a career that normally ends with board seats and a house in Woodside.
He decided to build a uranium enrichment company instead.
It is called General Matter. Peter Thiel sits on the board. It is backed by Founders Fund — the first institutional investor in SpaceX, Palantir and Anduril — and by In-Q-Tel, the venture arm of the American intelligence community.
In August 2025 the company signed a multi-decade lease with the Department of Energy for federal land at the former Paducah Gaseous Diffusion Plant, the site that enriched uranium for the United States from 1952 until 2013. On January 7, 2026 — twenty-four months after incorporation — the Department awarded it a $900 million ten-year contract to build and operate domestic HALEU capacity.
And here is the part I cannot stop turning over. General Matter has never publicly disclosed how it intends to enrich. Its own public description of the technology is that it is novel, scalable and cost-competitive. That is the whole disclosure.
A Silicon Valley hard-tech founder, funded by Thiel and by the CIA’s venture arm, restarting American uranium enrichment on a Cold War site, using a method he will not describe, on the same acreage where sixty-five thousand cylinders of uranium sit waiting for a train to Texas.
Some of this is in the numbers already. Unevenly, and in the wrong places.
Spot separative work has moved from roughly $193 per SWU at the end of 2024 to about $200 at the end of 2025, after a rise of more than 160 percent from pre-invasion levels. Urenco closed 2025 with a record order book of €21.3 billion — up 14 percent, running into the 2040s — on annual revenue just under €2.1 billion. An order book worth roughly a decade of sales.
Centrus carries a contracted backlog of about $3.9 billion through 2040, against 2026 revenue guidance of $450 to $500 million.
And in February 2026, when the owner of the only uranium conversion plant in the United States announced it would lift output by roughly 20 percent and disclosed a backlog above $2 billion, its shares rose almost 17 percent in a single session.
Seventeen percent, on a conversion announcement, from a company most screens still file under specialty chemicals.
That is what a market looks like when it has read the first line of the story and not the rest.
One more thing about that site in Ohio
The Portsmouth reservation outside Piketon covers about 3,700 acres. At its peak it had a design capacity of 7.4 million separative work units a year, its process buildings covered more than ninety acres under roof, one of them was half a mile long, and according to the old Atomic Energy Commission the plant drew as much electricity as New York City.
Today the same land holds sixteen centrifuges making 900 kilograms a year.
And on March 20, 2026, the Secretary of Energy and the Secretary of Commerce stood on that reservation with shovels, alongside executives from SoftBank, and broke ground on what SB Energy describes as the world’s largest artificial intelligence data centre — with a commitment to build ten gigawatts of new power generation alongside it.
Ten gigawatts of demand and nine hundred kilograms of fuel, on the same square mile of southern Ohio.
If you want a single image for the decade ahead, I don’t have a better one.
Macro Notes Premium
Eight weeks: NRC licensing dockets, the DOE allocation filings, the Portsmouth and Paducah project office briefings, the depleted uranium disposal schedules, Urenco’s audited accounts, and Centrus’s contract history back to the original 2019 award. I mapped every announced Western enrichment, conversion and deconversion project against its realistic commissioning date rather than its press-release date.
What came out of it is in the premium section below.
The full book, with entries and sizing. Every position I hold across the four steps of the chain, on three continents, with ticker, entry price, 24-month target, position size and the one-line thesis. You will recognise one or two. You will not recognise my largest.
The step nobody screens for. My biggest position performs a single step of the fuel cycle at a single American facility, holds an operating licence running to 2060, and carries a backlog above $2 billion. It listed recently enough that the analysts covering it are still modelling it as a different industry entirely. I explain the entry, the target, and the specific contract-repricing catalyst that begins when the Russian waivers lapse.
Which announced projects get built — and the one I’m short. A four-part test: licence status, feedstock security, technology disclosure, contracted offtake. Applied to every announced Western enrichment project. Several fail. One fails badly enough that I am positioned against it.
The reactor triage. I take the listed advanced nuclear names one at a time and answer a single question for each: where does the first core physically come from, and on what date. Several of the most widely held names in the sector have no credible answer. One has an unusually good one, and almost nobody has connected it to the allocation that guarantees it.
The cylinder yard trade. Under a 2016 agreement signed at the bottom of the uranium bear market, one company holds long-term rights to purchase and re-enrich a substantial slice of the higher-assay tails sitting at Paducah. No drilling, no exploration risk, no permitting, no sovereign risk. I explain how to own it, and why the timing has changed.
The calendar to 2028. Waiver expiry, NRC decisions on the two new American facilities, DOE allocation rounds, EBR-II recovery milestones, and the contract negotiations that convert headline awards into recognised revenue — with the 60-to-90-day positioning window ahead of each.
And what breaks it. Three scenarios, including the one nobody discusses: a design consensus drifting back toward sub-5 percent fuel, which would strand HALEU capacity outright. For each, the trigger I watch and the two hedges already in the book.
Everything above this line is public. It is simply scattered across a dozen government document systems, and nobody had put it in one place. Everything below it is what I’m doing about it.