This is BMInsider PRO content
You're reading a Deep Dive — our most in-depth research. Unlock full access to all analyses, Smart Money breakdowns & more.
Unlock BMInsider PRO →One-time payment · instant access · no account needed
Wall Street has fixated on the wrong shortage. For months investors have fought over every Nvidia chip and every gigabyte of memory, yet the real bottleneck of the AI era is taking shape somewhere else entirely: at the wall socket. Data centers are swallowing electricity at a pace that dwarfs the consumption of entire countries — and the quietest, most consequential bet on that fact carries a name almost no portfolio is watching: uranium.
This analysis traces the whole chain — from the kilowatt-hour inside the data center, through the reactor, down to the pound of uranium and the enrichment centrifuge — and shows where along that chain money can be made and lost. It is deliberately built to last: the hook is the AI mania of 2026, but the structural shifts at its center will play out over a decade, not a quarter.
The chips aren't scarce — the power is
This spring the International Energy Agency (IEA), in its landmark Energy and AI report, produced a number that reframes the whole debate: global data-center electricity consumption is set to roughly double to about 945 terawatt-hours by 2030 — just under 3 percent of world electricity, more than Japan consumes today. The lion's share of the growth comes from AI: accelerated servers (the GPU racks) grow around 30 percent per year in the base case, and electricity from AI-focused data centers triples.
The geographic clustering is what makes it acute. In the United States alone, data-center consumption rises by roughly 240 terawatt-hours by 2030 — up about 130 percent from 2024. That demand is not spread thinly across the country; it concentrates in a few dozen sites in Virginia, Texas and Ohio, where a single campus can pull as much power as a mid-sized city. You cannot meet a gigawatt of new load with a few extra wind turbines down the road.
An AI data center does not need just any electricity — it needs a very specific kind: around the clock, firm, uninterrupted. A training run that lasts 60 days and ties up hundreds of millions of dollars of hardware cannot tolerate a lull. Solar and wind are cheap but weather-dependent, and batteries bridge hours, not weeks. What is left is baseload: gas, coal, or nuclear. Of those three, nuclear is the only source that is both carbon-free and arrives in large, dispatchable blocks.
There is a further constraint that the public debate largely ignores: the grid itself. In the core U.S. markets, applications for grid interconnection have piled into queues that stretch out for years. A new gas plant or solar farm does a data center little good if there is no wire to connect it. That is precisely what makes the plants already attached to the grid — the operating reactors — so valuable: they deliver not eventually, but now. In 2026 the worth of a reactor is measured less by its running costs than by the simple fact of its existence at a location with a live grid connection.
Why the hyperscalers are going nuclear
The proof is not in forecasts but in signatures. Over roughly a single year, the four big cloud companies — Microsoft, Amazon, Google and Meta — have contracted more than 10 gigawatts of U.S. nuclear capacity. By May 2026 the industry counted 13 announced projects totaling over 9.8 gigawatts. What is striking is that this bet is not being driven by oil majors or utilities, but by the most profitable technology firms on Earth, which have effectively declared electricity their number-one input.
The most dramatic deal is the restart of Three Mile Island. Microsoft signed a 20-year power purchase agreement with Constellation Energy for 835 megawatts from the shuttered Unit 1 — the reactor whose neighbor suffered the most famous accident in U.S. history in 1979. The plant is being revived as the "Crane Clean Energy Center"; after a June 2026 waiver from federal regulators at FERC, the restart was pulled forward to the second half of 2027, a full year ahead of plan.
| Deal | Plant / technology | Capacity | Structure |
|---|---|---|---|
| Microsoft × Constellation | Three Mile Island 1 (restart) | 835 MW | 20-year PPA |
| Amazon × Talen | Susquehanna (2.5 GW) | min. 480 MW | Behind-the-meter campus, $650M purchase |
| Meta × Constellation | Clinton (Illinois) | ~1.1 GW | 20-year PPA, life extension |
| Meta (pipeline) | TerraPower, Oklo, Vistra, etc. | up to 6.6 GW | Mix of existing reactors and SMR commitments |
Amazon took a different route, paying $650 million for a data-center campus sited next to the Susquehanna nuclear plant — power taken "behind the meter," bypassing the public grid entirely. Meta, for its part, used a 20-year contract to extend the life of the Clinton reactor in Illinois, which without that commitment faced closure. The pattern is everywhere the same: the tech giants pay fixed, often premium prices for two decades in exchange for the certainty that the power flows. For the operators, these are the most predictable revenues in a generation.
The uranium market: a structural deficit
Every one of these reactors needs fuel, and here the supply story begins in earnest. The long-term contract price for uranium (U3O8) stood at about $90 per pound at the end of the first quarter of 2026 — its highest level since 2008. The spot price hovered around $86 to $87. Some individual long-term contracts are already being struck near $150 per pound, a sign that utilities now weigh security of supply above the daily price.
On the demand side, the World Nuclear Association projects reactor uranium demand rising 28 percent by 2030 and more than doubling by 2040. Against that stands a supply base that is shrinking at precisely the wrong moment. Kazatomprom, the world's largest and lowest-cost producer, cut its quota by 8 million pounds and lowered nominal 2026 capacity to about 77 million pounds (from 85), with an option to "downflex" actual output up to 20 percent below that. Cameco, the Western leader, cut its 2025 production guidance to 14–15 million pounds — instead of the hoped-for 18 — because of delays at its McArthur River mine.
| Metric | Value (2026) | Significance |
|---|---|---|
| Long-term contract price | ~$90/lb U3O8 | highest since 2008 |
| Spot price | ~$86–87/lb | new contracts partly near $150 |
| Reactor demand to 2030 | +28% | more than doubling by 2040 (WNA) |
| Kazatomprom 2026 capacity | ~77M lb | cut from 85, plus downflex up to −20% |
| Cameco 2025 output | 14–15M lb | vs. planned 18 (McArthur River delay) |
What makes this market unusual is that it is small, opaque and transacted through long-term contracts that are never publicly posted. There is no liquid futures market as there is for oil. A utility running a reactor for 60 years cannot afford a fuel misstep — uranium is only a small slice of the cost of generating power, but without it the reactor sits idle. That price-insensitivity on the demand side now collides with a tight supply base. Historically, exactly this configuration has produced violent price moves.
The hidden second bottleneck: enrichment and conversion
Anyone watching only the uranium price is missing what may be the more interesting part of the story. Raw uranium is useless until it passes through two processing steps: conversion into uranium hexafluoride (UF6) and enrichment of the fissile U-235 isotope. Both steps have become bottlenecks in their own right — and geopolitical ones at that.
The price of enrichment services, measured in Separative Work Units (SWU), has jumped from around $40 before 2022 to roughly $160 — a fourfold increase. Conversion and enrichment prices combined rose five- to tenfold. The cause is the war in Ukraine: Russia controlled a large share of the world's enrichment capacity. Western conversion capacity runs at about 15,000 tonnes of uranium per year against a global requirement near 75,000 tonnes. The West, in other words, covers only about a fifth of its own needs and depends on Russia and China for the rest.
That dependence now runs into a hard deadline: a full U.S. ban on imported Russian enriched uranium takes effect in 2028. The West must rebuild, in a few short years, an industry it allowed to atrophy for decades. In January 2026 the U.S. Department of Energy selected four contractors to spur enrichment; Centrus Energy alone was awarded $900 million, in part for producing HALEU — the high-assay fuel the next generation of reactors requires and which, until recently, only Russia supplied commercially. For investors this is a distinct and often-overlooked bet: not on the metal, but on refining it.
Small modular reactors: the bet on the next decade
The acronym that appears in every AI-energy pitch deck is SMR — the small modular reactor. The idea: smaller, factory-built reactors of 15 to 300 megawatts that could be sited beside a data center faster and cheaper than a conventional gigawatt plant. Meta, Google and Amazon have all signed letters of intent with SMR developers.
But a sober look is mandatory. As of early 2026, not a single commercial SMR is operating anywhere in the Western world. NuScale holds the only fully U.S.-certified design and is working with the utility TVA toward as much as 6 gigawatts of capacity — yet the first commercial unit is not yet running. GE Hitachi's BWRX-300 (300 MW) is under construction in Canada and slated for projects in the U.K. and Poland. Oklo — prominent for its ties to Sam Altman — broke ground on its 15-to-75-megawatt "Aurora" reactor at Idaho National Laboratory in September 2025, and generates no revenue to date.
Here lies the dividing line between the two investment theses. The incumbent thesis — operating reactors, uranium miners, enrichers — rests on real cash flows and real contracts. The SMR thesis is a venture bet: enormous upside, but years from commercialization, with equities whose valuations consist almost entirely of expectation. Both can be right — but they carry completely different risk profiles.
Making the chain investable
The mistake many newcomers make is to buy "uranium" as if it were a single asset. It is in fact a value chain of four very different links, each responding to the same trend in its own way.
| Link | What it is | Examples | Character |
|---|---|---|---|
| Mining | Extraction of U3O8 | Cameco, Kazatomprom, Uranium Energy, Denison | Leverage to the uranium price, operational risk |
| Physical uranium | Trusts holding stored uranium | Sprott Physical Uranium Trust | Pure price bet, no operating risk |
| Enrichment | Refining into fuel | Centrus Energy | Its own bottleneck, geopolitically driven |
| Utilities / reactors | Power generation | Constellation, Vistra | Cash flow today, direct PPA beneficiary |
An investor who wants to play the uranium price itself reaches most directly for the physical trust — it stores real uranium and tracks the price almost one-for-one. Those seeking operational leverage choose the miners, but must accept extraction risk, country risk and capital costs. Utilities are the most conservative option: they already earn money and benefit from the long-term tech contracts, but they are less of a pure uranium play. And the enrichers are the niche bet on the geopolitical chokepoint.
One profile deserves special attention: Cameco is no longer merely a miner. Together with the infrastructure investor Brookfield, it acquired the reactor builder Westinghouse in 2023 and holds roughly 49 percent of it. A single security therefore straddles two links of the chain at once — the raw material and the construction of the reactors that consume it. Buy Cameco and you are buying not just a pound of uranium but a slice of the entire Western nuclear supply chain. Such integrated profiles are rare, and they explain why the stock often trades at a premium to pure-play miners.
Valuation and scenarios
How much of this story is already priced in? After a tripling of the uranium price since 2020 and the surge in stocks like Constellation, the sector is no longer cheap. Rather than a point forecast, it pays to think in scenarios.
| Scenario | Assumption | Uranium price | Implication |
|---|---|---|---|
| Bull | AI power hunger + reactor newbuild + Russia ban | >$120/lb, sustained | Re-rating of the whole chain; enrichers as standouts |
| Base | Deficit real but gradual; PPAs support utilities | $85–110/lb | Utilities and physical uranium solid; miners volatile |
| Bear | AI capex disappoints; Kazakhstan floods market | <$70/lb | SMR stocks collapse; miners under pressure |
The bear case — taken seriously
No deep dive without a bear case, and the uranium market has a credible one. First, this sector is notorious for its sentiment cycles: back in 2007 a strikingly similar narrative drove uranium above $130 a pound — before it slid into irrelevance for years. Anyone who bought at that peak waited more than a decade for a recovery.
Second, AI power demand is a forecast, not a fact. Should the hyperscalers' capex boom prove a bubble — a risk hinted at by the recent whipsaw in chip stocks — a chunk of the demand narrative would evaporate. Third, Kazakhstan is the wild card: Kazatomprom can, in theory, ramp production quickly and flood the market once prices are attractive enough. And fourth, SMR timelines are notoriously optimistic; delays and cost overruns are the rule rather than the exception in the nuclear industry. A bet on reactors that will not deliver power until the end of the decade at the earliest is a bet on patience.
Access and tax for U.S. and international investors
For U.S. investors the menu is broad but has traps. You can buy the miners and utilities directly — Cameco (CCJ), Constellation (CEG), Vistra (VST) — or take diversified exposure through funds such as the Global X Uranium ETF (URA) or the Sprott Uranium Miners ETF (URNM). The most direct price play, the Sprott Physical Uranium Trust, holds physical U3O8 and trades at a premium or discount to its net asset value; crucially, for a U.S. taxpayer it is a passive foreign investment company (PFIC), which generally means filing a QEF election to avoid punitive PFIC tax treatment — a wrinkle worth clearing with a tax adviser before you buy.
On the equity side, qualified dividends from the utilities are taxed at favorable long-term rates, while gains on the miners are ordinary capital gains subject to the usual short- versus long-term holding rules. For tax-advantaged compounding, some investors hold the more tax-inefficient pieces — the physical trust, the enrichers with no dividend — inside an IRA. International investors face a different friction: European UCITS wrappers such as the VanEck Uranium and Nuclear Technologies UCITS ETF exist precisely because U.S.- and Canada-listed vehicles like the physical trust are often blocked for retail buyers in the EU.
One policy backdrop is worth flagging, because it shapes the flow of capital: since 2022 the European Union has classified nuclear energy, under conditions, as "sustainable" under its taxonomy. Austria challenged that label and lost — the EU General Court dismissed its case on September 10, 2025, and Vienna announced an appeal that November. The green stamp matters: it channels ESG-constrained European capital toward exactly the utilities and reactor operators U.S. investors are already buying, adding a structural, policy-driven bid beneath the sector.
Bottom line
The AI revolution will not be decided by the number of chips alone, but by whether enough electricity flows to run them. Nuclear is the only carbon-free baseload available in the blocks required — and the hyperscalers have already answered that question with billion-dollar contracts. For investors this is not a short-term trade but a structural thesis spanning a decade, with four very different entry points and a real bear case. Anyone who plays it should know which link of the chain they are buying — and that this market's cycle is as unpredictable as its raw material is rare.