Tearing apart the uranium fuel cycle

Uranium.

The long-term contract price just closed August at $96.50 a pound — above the 2007 peak, a 19-year high. Financial media is treating this like a momentum story. "AI nuclear hype," says the Bloomberg chyron. "Big Tech drives uranium rally," says Reuters. The implication is clear: this is a trade, driven by sentiment, and probably overextended.

They're watching the dashboard. The engine is somewhere else entirely.

The uranium fuel cycle is not a commodity market. It's a four-stage industrial chain — mine, convert, enrich, fabricate — and every stage is controlled by a different set of hands in a different country. Right now, the narrative is fixated on mines and reactors. New demand from AI data centers! New reactors in China! That part is true. But the bottleneck that will actually break this market sits between the mine and the reactor, in a part of the chain most equity analysts couldn't diagram on a whiteboard.

The enrichment layer. That's today's machine.

Story off. Numbers on.

UxC estimates secondary supply at 27 million pounds for 2025, falling to 17 million by 2030. The cushion is thinning. And the two largest primary producers just made it worse.

Kazatomprom — the world's largest uranium miner, responsible for roughly 20% of global output — cut 2026 nominal production 10%, from 32,777 tonnes to 29,697 tonnes. That's 8 million pounds removed from a market already running short. The stated reason: current supply-demand dynamics "do not warrant lifting production to the 100 percent level." Translation: they'd rather let the deficit widen than sell into a price they believe is still too low.

Cameco, the West's flagship producer, cut McArthur River/Key Lake guidance 22% in 2025, then saw Key Lake temporarily halt production and McArthur River reduce activity in May 2026 when Saskatchewan flooding partially collapsed the Smoothstone River Bridge — the primary supply road in. Operations resumed, but 2026 McArthur River/Key Lake guidance sits at 14.0–16.5 million pounds, and the bridge incident proved how fragile the logistics are. A single damaged road in northern Canada can choke 15% of Western uranium supply.

Meanwhile, the demand side is accelerating into a wall. Seventy-four reactors are under construction globally — 37 of them in China alone. The IEA says 78 GW of nuclear capacity is being built across 15 countries, with total Chinese capacity expected to reach 100 GW by 2030. And now the tech companies have arrived.

Microsoft, Amazon, Google, and Meta have collectively committed to 9.8 GW of nuclear power across 13 deals. Google just signed a 22-year PPA with Fortum for 50% of Finland's Loviisa nuclear plant through 2050 — backed by a €13 billion data center buildout. These aren't speculative. They're contractual commitments to consume reactor fuel for decades.

The mine-to-reactor math doesn't balance. It hasn't balanced for years. But here's the part the mine-focused analysis misses entirely: even if you could dig up enough yellowcake tomorrow, you still can't turn it into reactor fuel. Because the enrichment plant is the bottleneck now. And that's where the real fault line runs.

Rosatom controls 44% of global uranium enrichment capacity. Twenty-seven million separative work units out of roughly sixty-two million worldwide. Urenco has 17 million. Orano has 7.5 million. China's CNNC has 10 million SWU and exports enrichment services abroad. That's the map.

Now erase Russia from it.

The U.S. banned Russian enriched uranium imports in 2024. Russia counter-banned exports to the U.S. on a shipment-by-shipment basis — maximum uncertainty by design. The result: Western utilities that relied on Rosatom for roughly a quarter of their enrichment services are now scrambling for capacity that doesn't exist yet.

Analysts at the Al Habtoor Research Centre call it plainly: 2026–2028 is the "acute structural shortfall" — Western demand for enrichment will exceed total non-Russian supply for at least three years. Urenco started the fifth of eight new cascades at its Eunice, New Mexico facility, adding 700,000 SWU of capacity. That helps. But the larger expansion — 2.1 million SWU across 24 cascades — broke ground in August 2026, but full construction doesn't begin until 2029, with first production in 2032. The DOE awarded Orano $900 million in January to build a new centrifuge enrichment facility. That's a check written against a timeline measured in years, not quarters.

And here's the mechanical detail that the mine-focused analysts keep missing. When enrichment was cheap, enrichers would "underfeed" — overwork the centrifuges to squeeze extra uranium product from less natural feed. This functioned as a hidden secondary mine, quietly adding supply. Now that SWU prices have tripled, that dynamic has reversed. Enrichers are tails-stripping less. The hidden mine just shut down. Net effect: effective demand for mined uranium just increased, on top of everything else.

Then there's the financial buyer problem. Sprott Physical Uranium Trust holds roughly 81.7 million pounds of U₃O₈ — physical yellowcake, sitting in licensed storage, removed from the market. They bought 5.8 million pounds in Q1 2026 alone. In the last uranium bull market, financial buyers like this didn't exist at scale. Now they're competing directly with utilities for physical supply. The U.S. consumed 40 million pounds last year. Domestic production: around 2.1 million pounds. Roughly 95 percent import dependent. I've seen tighter supply chains, but not many, and none where the buyer was also building the reactor.

The enrichment squeeze doesn't show up in a spot price chart. It shows up in delivery timelines, in contract terms that now stretch past 2035, in the number of competitive bids per utility tender — which has quietly collapsed. That's the gap between the dashboard and the engine room.

The sequence from here is mechanical. I've watched versions of it play out across commodities — palladium in 2020, nickel in 2022, cobalt this year. The trigger varies. The choreography doesn't.

Citi put it on the record last week: $100-plus per pound within three months, with a bull case averaging $99 for full-year 2026 and reaching $140 by late 2027. Jefferies raised its long-term forecast 36% to $95 — and that was the conservative case. The tell is in how Citi flagged the second-order risk: junior miners that sold forward production they haven't yet produced will be forced to buy physical uranium in the spot market to fill contracts. That's the same mechanic that blew up nickel in 2022. Forced covering by producers into a thin physical market.

The U.S. last produced meaningful domestic uranium in the 1990s. Forty million pounds of annual consumption, around two million pounds produced at home. The DOE has allocated $2.72 billion to rebuild domestic uranium enrichment capacity. Good. But the money is measured in billions and the timeline is measured in years, and the enrichment shortfall is measured in right now.

Where does the capital go? Not into the broad nuclear-energy ETFs. Those are diluted with engineering firms, utility operators, and reactor component manufacturers — companies whose revenue isn't tied to the pound price of yellowcake. The edge, if there is one, is upstream: Cameco and Kazatomprom as the two producers who actually control swing supply; NexGen and Denison for permitted Athabasca Basin reserves that are already highly economic at current prices; and Sprott Physical Uranium Trust as the de facto inventory proxy that tightens the market every time it buys.

The enrichment pinch is harder to play directly — Urenco is private, Orano is partially state-owned, and Rosatom is sanctioned. But the enrichment squeeze transmits upward. When the centrifuge can't deliver, the utility hoards feed. When the utility hoards feed, the mine gets a call.

The dashboard says momentum trade. The engine says structural deficit at every stage of the fuel cycle, with a 2026–2028 enrichment gap that no amount of mine output can solve. In my experience, when the physical layer and the narrative disagree this sharply, the physical layer wins. It just takes longer than anyone with a quarterly P&L wants it to.