Uranium: a market in deficit, the promise of AI, and hidden bottlenecks

Uranium has become, in a few quarters, one of the most talked-about commodities. The story is seductive: structurally growing demand, driven by the nuclear revival and now by the electricity thirst of artificial intelligence, against a supply slow to respond. This story is largely well founded. But a commodity market is never read through demand alone, and uranium hides, behind the ore, a far more binding bottleneck: enrichment. Let us lay out the thesis, then its antithesis, with the figures. For the fundamentals, see our guide on reading the uranium market.

A market already tight before AI

The price signal leaves little doubt about the tension. The spot price jumped about 25% in January 2026 to move back above $100 a pound of uranium oxide (U3O8), a first in two years. Above all, the long-term price, the one that matters for reactor operators’ contracts, reached $93 a pound at the end of March 2026 according to TradeTech’s indicator, its highest level in more than eighteen years. Investor surveys point to a $100 to $120 range, with expectations up to $135.

Behind the price, a fundamental imbalance. World mine production stands around 60,000 tonnes of uranium a year, when reactor needs already approach 69,000 tonnes. The gap is filled by secondary supply, that is, inventories, reprocessing and various non-mining sources. Yet demand is on a steep upward path: the World Nuclear Association projects installed nuclear capacity rising from 398 gigawatts electric mid-2025 to 746 in 2040, and uranium requirements climbing from about 68,900 tonnes in 2025 to more than 150,000 in 2040. At the same time, 78 reactors are under construction worldwide, China in the lead with about 38 units.

AI plugs nuclear back in

It is against this already-tight backdrop that the electricity demand of artificial intelligence arrives. According to the International Energy Agency, world data-centre consumption would rise from 415 terawatt-hours in 2024 to 945 in 2030, a hyperscale data centre dedicated to AI potentially demanding 300 to 500 megawatts, the equivalent of a mid-sized city. This electricity, the tech giants want it dispatchable and decarbonised, and they have turned to nuclear.

The announcements have multiplied. By mid-2026, every major hyperscaler had signed at least one nuclear deal: in total, thirteen announced projects commit more than 9.8 gigawatts of capacity to power AI. Microsoft secured a twenty-year, $16bn power-purchase agreement for the restart of Three Mile Island’s unit 1, some 835 megawatts expected around 2027. Meta targets up to 6.6 gigawatts by 2035, relying on advanced reactors from TerraPower and Oklo and existing plants from Vistra and Constellation. Google signed with Kairos Power and NextEra to restart a plant in Iowa. Amazon is developing small modular reactors (SMRs) with X-energy and Energy Northwest.

A point of method imposes itself here, because it conditions the whole reasoning. These deals cover electricity delivered between 2027 and 2035, via restarts, existing plants or small reactors still to be built. The impact on physical uranium demand is therefore real but deferred, and it concerns the fuel as much as its transformation. AI’s promise supports the long-term story; it does not create an immediate call on uranium.

A concentrated supply, slow to respond

Facing this demand, supply has two flaws: it is concentrated and it responds slowly. The world’s leading producer, Kazakhstan’s Kazatomprom, and Canada’s Cameco together supply more than 40% of production, Kazakhstan alone weighing about 40% of world extraction. This concentration is a risk factor in itself. Kazatomprom has moreover lowered its ambitions: its 2026 production is guided between 27,500 and 29,000 tonnes on a 100% basis, below its nominal capacity, owing notably to a shortage of sulphuric acid, a key input for in-situ leach mining.

Opening or restarting a mine takes years and a lot of capital. In the United States, a few projects are restarting, such as Uranium Energy Corp’s Burke Hollow in Texas or Ur-Energy’s Shirley Basin in Wyoming, but domestic production remains marginal. Over the longer term, supply eventually responds to price, which tempers the most bullish scenarios. In the short term, it does not: the World Nuclear Association notes that rising demand runs into a supply base that is little elastic to current price levels. Notably, the geography of supply is also recomposing, Kazakhstan steering a growing share of its sales toward China, Russia and India, via long-term contracts.

The bottleneck the market underestimates: enrichment

The real point of fragility is not always where one looks. Extracting uranium is not enough: it must be converted then enriched before it becomes fuel. Yet enrichment is an oligopoly where Russia holds a dominant place, with about 44% of world capacity. The United States depends heavily on it: Russia supplied about 35% of its enriched-uranium imports, and US operators bought 4,141 thousand separative work units (SWU) from it in 2023.

Washington has decided to sever this link. The law banning Russian uranium imports, enacted in May 2024, will take full effect in 2028, with waivers possible until then. It unlocks $2.72bn to build a domestic supply chain. The problem is acute for advanced reactors, which require a more enriched fuel, HALEU, of which Russia was until 2024 the only commercial supplier. In the United States, only Centrus currently produces HALEU, at small scale, with more than 900 kilograms delivered to the Department of Energy and a $900m contract signed on 1 July 2026. In other words, the AI-driven nuclear revival could stumble less on the ore than on the capacity, to be rebuilt, to enrich it outside Russia.

The antithesis: why the story could disappoint

Rigour demands taking the objections seriously, because uranium is a market with a cyclical and volatile history, prone to bouts of enthusiasm as much as to disillusion. Several factors invite caution.

First, the mining deficit is not an immediate shortage. It has been filled for years by secondary supply, and operators long preferred to draw on their inherited inventories rather than contract anew. This cushion delays the reckoning. Next, AI demand is largely a story of the 2030s. The commercialisation of small reactors has fallen behind, a flagship project having been abandoned in the United States in 2023, and the first SMRs would not be operational before 2030 or 2031. The tech giants’ deals deliver electricity, not uranium tomorrow.

To this is added an execution risk on the producer side, with licensing delays and forecast revisions on several projects, and a precedent that should breed humility: after Fukushima in 2011, nuclear demand collapsed and uranium spent a decade in lethargy. Finally, AI demand itself rests on spending expectations we have learned to view with caution, from the real productivity of AI to the risk of a bubble in valuations. If AI investment slows, part of the promised electricity demand evaporates with it. It should be noted, lastly, that some of the bullish talk comes from actors selling uranium-exposure products, which invites distinguishing analysis from the sales pitch.

Outlook, and the signals to watch

A balanced reading emerges. The uranium market is structurally tight, with credible rising demand and a concentrated, rigid supply, which argues for durably firm prices. AI demand reinforces this story, but on a 2030 horizon more than the immediate one, and the most binding bottleneck sits at enrichment, not the mine. The risk is not so much that the thesis is wrong as that its direction, probably right, gets confused with its timing, highly uncertain.

For anyone wanting to follow this commodity, a few indicators concentrate the information. TradeTech’s long-term price, more telling than the spot, says operators’ conviction. The pace of utility contracting signals the end of the inventory cushion. Kazatomprom’s and Cameco’s guided production give the pulse of supply. The milestones of restarts and SMRs, along with the ramp-up of enrichment outside Russia, notably HALEU, will say whether the promise materialises. And the trajectory of AI spending will remain the arbiter of the expected electricity demand. It is this whole, more than any single figure, that deserves sustained watching.

Sources

1. World Nuclear Association, uranium markets, supply and demand: installed capacity from 398 GWe mid-2025 to 746 GWe in 2040, requirements rising from about 68,900 to more than 150,000 tonnes of uranium, supply little elastic at current prices: https://world-nuclear.org/information-library/nuclear-fuel-cycle/uranium-resources/uranium-markets
2. TradeTech, uranium price indicators: long-term price at $93.00 a pound of U3O8 at end-March 2026, highest in more than eighteen years: https://www.uranium.info/press_releases.php
3. Sprott, “Uranium Outlook 2026”: spot jump above $100 in January 2026, contracting dynamics, structural tension (source from an issuer of uranium products, to be read as such): https://sprottetfs.com/insights/uranium-outlook-2026/
4. International Energy Agency, data-centre electricity consumption from 415 TWh in 2024 to 945 TWh in 2030, hyperscale data-centre needs: https://www.iea.org/reports/energy-and-ai
5. smrintel, census of tech giants’ nuclear deals: thirteen projects, more than 9.8 GW for AI, Microsoft, Meta, Google, Amazon detail: https://smrintel.com/nuclear-data-center-deals/
6. Forbes, 19 February 2026, Microsoft and Amazon turn to nuclear for AI, $16bn contract for Three Mile Island: https://www.forbes.com/sites/rrapier/2026/02/19/why-microsoft-and-amazon-are-turning-to-nuclear-power-for-ai/
7. World Nuclear News, 2025 production results for Cameco and Kazatomprom, Kazatomprom’s 2026 guidance (27,500 to 29,000 tonnes, sulphuric-acid constraint): https://world-nuclear-news.org/articles/Cameco-Kazatomprom-release-2025-figures
8. World Nuclear Association, uranium and nuclear power in Kazakhstan, the country’s share of world production: https://world-nuclear.org/information-library/country-profiles/countries-g-n/kazakhstan
9. U.S. Department of Energy, ban on Russian uranium imports and development of the domestic supply chain, $2.72bn: https://www.energy.gov/ne/articles/russian-uranium-ban-will-speed-development-us-nuclear-fuel-supply-chain
10. Nuclear Regulatory Commission, context of the ban: Russia at about 44% of world enrichment capacity, about 35% of US imports, 4,141 thousand SWU bought in 2023: https://www.nrc.gov/reading-rm/doc-collections/fact-sheets/uranium-import-ban
11. Utility Dive, law banning Russian uranium, full effect in 2028, $2.7bn unlocked for HALEU and enrichment: https://www.utilitydive.com/news/congress-passes-russian-uranium-import-ban-haleu-nuclear-fuel-advanced-reactors/715256/
12. Power Magazine, Centrus, sole US producer of HALEU, more than 900 kg delivered to the DOE, $900m contract of 1 July 2026: https://www.powermag.com/centrus-completes-900-kg-haleu-delivery-to-doe-in-u-s-nuclear-fuel-enrichment-milestone/
13. Sprott, “Uranium’s Tale of Two Markets”: SMR delays (first reactors operational around 2030-2031), utilities drawing on inventories, execution risks: https://sprott.com/insights/uranium-s-tale-of-two-markets/