Who keeps the profit as nuclear power returns—and can AI demand make new reactors pay?

Nuclear power is a chain of distinct businesses: atomic physics, Cold War institutions, uranium mines, centrifuge halls, specialised factories, construction sites and electric grids. The industry is returning as energy security, climate policy and a new class of customer converge: technology companies building AI data centres that need power around the clock. For now, the clearest profits accrue to owners of reliable, already licensed reactors and to the few suppliers of scarce, qualified fuel services. New reactors can pay, but only when a standard design, secure fuel, a licence, patient capital and a creditworthy buyer align. AI can provide the last of these—a long-term buyer. It cannot, by itself, make a first-of-a-kind reactor cheap or punctual.

The atom escapes the laboratory

Late in 1938, a letter from Berlin reached Lise Meitner in Sweden. An Austrian-born physicist, she had worked alongside the chemist Otto Hahn for three decades before Nazi Germany's racial laws forced her to flee that summer.¹ Hahn wrote that he and his assistant, Fritz Strassmann, had bombarded uranium with neutrons and found barium in the residue. Barium has roughly half uranium's mass. Chemically, the result made no sense.²

Over Christmas, Meitner worked through the puzzle with her nephew Otto Frisch, also a physicist. They pictured the uranium nucleus as a wobbling liquid drop. A captured neutron could make it vibrate so violently that it stretched and broke into two unequal parts. Using Einstein's link between mass and energy, they calculated that each split released about 200m electron volts. Frisch confirmed the energy release in the laboratory in January 1939.² They called the process fission.¹

The liquid-drop image conveys the basic idea: a large, unstable object is pushed past a tipping point and breaks apart. But the analogy has limits. A water droplet is held together by surface tension; a nucleus is bound by the strong nuclear force and governed by quantum rules. That is why the energy release is so large. Burning a coal or gas molecule releases a few electron volts. Splitting one uranium nucleus releases tens of millions of times more. This disparity underlies uranium's value, reactors' small fuel requirements and the weapons that followed.

From one split to a chain

Fission also releases neutrons, each of which can split another nucleus. Picture a floor covered with set mousetraps, each holding a ping-pong ball. Drop in one ball and the room erupts. The engineering challenge was to hold that cascade at a steady rate rather than let it run away.

Enrico Fermi provided the answer. An Italian Nobel laureate who had emigrated to the United States, Fermi led a University of Chicago team that, in December 1942, operated an experimental pile of graphite and uranium to achieve the first controlled, self-sustaining chain reaction.³² In a reactor, control rods absorb neutrons, removing some of the “balls” from the room. A moderator and coolant slow the neutrons and carry away heat. The mousetrap analogy fails in another important respect: once a trap fires, it stays still. A reactor does not. Radioactive by-products continue to generate heat after the chain reaction stops, requiring continuous cooling. The industry's worst accidents exposed the consequences of failing to manage that heat.

A weapon looking for a peaceful job

Fission entered public life through war. Wartime bomb programmes built the first uranium mines, large reactors, specialised metals operations and costly uranium-isotope separation plants. None was an electricity business. Governments funded them in secret for destructive ends. After 1945, political leaders had to give those capabilities a civilian purpose the public would accept and investors could finance.

On 8 December 1953, President Dwight Eisenhower addressed the United Nations General Assembly and proposed “Atoms for Peace”: international cooperation on civilian atomic uses under a new international agency.⁴ In the US, the Atomic Energy Act of 1954 ended the federal government's monopoly over nuclear activity. It allowed private companies to own licensed reactors and established the legal basis for commercial development.⁵ The International Atomic Energy Agency began operating in Vienna in 1957. Its remit contained a lasting tension: promoting peaceful nuclear energy while using safeguards to check that civilian material was not diverted to weapons.⁴

The ambition was global. Homi J. Bhabha, a Cambridge-trained physicist who returned to build India's scientific institutions, presided over the first United Nations conference on peaceful uses of atomic energy in Geneva in 1955.¹ For Bhabha, and for leaders in Paris, Moscow and Ottawa, nuclear power was more than a way to generate electricity. It promised greater national control over energy and signalled industrial standing. That motive endured, helping to explain why governments still shape the industry.

The other thread: the utility

At the same time, a separate story was unfolding. Electricity systems were consolidating around large central stations run by regulated utilities. These monopolies could raise capital over decades because regulators allowed them to recover costs, plus a return, from customers.

Nuclear fit this model. Its fuel is exceptionally energy-dense, allowing a plant to operate for 18 months or more between refuelling outages. Most costs are paid upfront, which suits an owner able to spread them across four decades of assured customers. The same features became nuclear's weakness. Plants are expensive, heavily regulated and slow to build, so higher interest rates or construction delays feed directly into electricity costs.

The fuel becomes an industry

Uranium was the industry's first physical bottleneck, and it still illustrates how the business works. Miners extract or dissolve uranium ore and sell a concentrate known as yellowcake (U₃O₈). Converters turn it into uranium hexafluoride (UF₆), which becomes a gas at modest temperatures. Enrichers raise the share of uranium-235—the isotope that splits readily—from about 0.7% in nature to the few per cent required by most power reactors. Fabricators press the enriched material into pellets and seal them in rods and assemblies designed for a specific reactor type.

The chain resembles aircraft-parts manufacturing: each supplier adds value and must be certified. The difference is that every nuclear step is also governed by international safeguards, export controls and national-security rules. In aerospace, a buyer may switch suppliers after qualification testing. In nuclear fuel, switching can take years and may require government approval.

The record also challenges any heroic account of the industry. Scientific feasibility never guaranteed cheap electricity. The early decades produced many reactor designs that worked technically but never became mass-market power plants.² The tension was present from the start. Nuclear's physical advantage is compact, dependable heat; its economic disadvantage is that safety, licensing and construction turn that heat into a long institutional project. The first commercial race tested that tension on three fronts: Soviet engineers showed that fission could supply a grid, American utilities showed that private capital could own a full-scale plant, and a navy demonstrated an approach to reactor design and construction.

↑ Back to Top

The world chooses its reactor families

In June 1954, a small reactor at the Physics and Power Engineering Institute in Obninsk, south-west of Moscow, began supplying electricity to the Soviet grid. It was the first nuclear power plant to provide civilian electricity.²⁶ At roughly 5 megawatts, it was more demonstration than power station.⁶ But it showed that fission could serve an electric grid.

Britain's Calder Hall followed in 1956 with a 50-megawatt, gas-cooled Magnox reactor and operated until 2003.² The 60-megawatt Shippingport plant in Pennsylvania began operating in 1957. In 1960, Dresden-1 in Illinois, a 250-megawatt General Electric reactor, became the first full-scale American nuclear station financed privately.²⁷

The admiral who set the template

Much of the Western commercial fleet grew out of naval engineering. Hyman Rickover, born in the Russian Empire and raised in an immigrant family in Chicago, rose through the US Navy as an engineer. As head of its reactor programme, he developed the pressurised-water reactor (PWR) for submarines. The Mark 1 prototype started up in Idaho in March 1953, and the USS Nautilus, the first nuclear-powered submarine, was launched in 1954.² Shippingport was also a pressurised-water demonstration plant.²

Rickover's legacy was as much institutional as technical: exact specifications, individual accountability and little tolerance for deviation. That approach still shapes specialist suppliers to naval reactors. BWX Technologies $BWXT is a listed example, making naval-reactor components and nuclear fuel—businesses in which certification, precision and long customer relationships matter more than volume.⁸ Revenue rose 18.3% in 2025, to $3.2 billion. Operating margin fell to 10.1%, from 16.3% in 2021, showing that faster sales had not yet translated into broader margins.⁸

Why the design choice stuck

A reactor is a licensed system: its fuel, pumps, control room, containment, operator training and emergency plans are designed and approved together. Costs fall only when a country builds the same design often enough for workers, suppliers and regulators to accumulate experience. The choices made in the 1950s and 1960s therefore still shape the industry.

Light-water designs became dominant. Pressurised-water reactors account for about 69% of global nuclear capacity, while boiling-water reactors account for about 20%.² Countries then built institutions around their chosen designs, reinforcing those choices. France concentrated authority in the state, EDF and a national reactor and fuel industry, later organised around Framatome for reactor equipment and fuel and Orano for the fuel cycle. The Soviet programme developed the VVER, its own pressurised-water family with dedicated fuel. It later expanded into Росатом Rosatom, a state corporation that packages uranium, enrichment, fuel, reactors and construction for export. South Korea combined 한국전력공사 Korea Electric Power Corporation (KEPCO) $015760.KS, KEPCO's engineering and maintenance arms, and domestic equipment makers into a single programme. China entered later but built more consistently than other countries, giving 中国核工业集团有限公司 China National Nuclear Corporation (CNNC) and 中国广核电力股份有限公司 CGN Power $1816.HK a home market of repeat orders that Western vendors have lacked for decades.⁸

The United States took a different institutional path. Initially, the Atomic Energy Commission both promoted and regulated nuclear power. The Energy Reorganization Act of 1974 abolished it, creating the Nuclear Regulatory Commission (NRC) for safety and a separate development agency.⁷⁹ The split separated commercial promotion from safety oversight, but left no single US institution responsible for driving down nuclear construction costs.

The installed base becomes the business

Westinghouse's pressurised-water technology became the basis for much of the Western fleet, while General Electric's boiling-water design became the main alternative.¹⁰ Framatome played a comparable role in France. Because reactors operate for decades, their original technology providers can continue to sell fuel, spare parts, inspections and upgrades after construction work ends.

Fuel illustrates the durability of that position. Changing suppliers is less like switching petrol stations than replacing certified parts in an aircraft engine. New fuel assemblies must be shown to perform safely in the reactor core, and regulators must approve them. This helps explain why Westinghouse's 2025 revenue rose to $5.0 billion from $4.3 billion a year earlier, despite few new reactor builds.⁸ Framatome, now part of EDF, reported sales of €5.4 billion.⁸

Was it inevitable?

By the early 1970s, utilities were placing orders, forecasters expected rapid demand growth and governments treated reactors as symbols of national prestige. That outlook depended on stable inflation, cheap capital, predictable construction and public consent. When those conditions weakened, many boom-era orders were cancelled and the US order book stalled for a generation.⁷ Nuclear technology worked; its economics depended on the institutions and financial conditions around it. In the late 1970s and 1980s, high interest rates, changing power markets and two disasters undermined those conditions.

↑ Back to Top

The disasters that changed the price of trust

At about 4 a.m. on 28 March 1979, feedwater pumps failed at Unit 2 of the Three Mile Island plant near Middletown, Pennsylvania. The reactor shut down automatically, as designed. A relief valve then stuck open, but control-room instruments suggested that it had closed. Operators did not realise coolant was escaping. Their response left the core partially uncovered; it overheated and partly melted.¹¹ It remains the most serious accident in the history of US commercial nuclear power.¹¹

Metropolitan Edison owned the plant. For several days, the scale of the emergency was unclear both outside the site and within it. Harold Denton, the NRC's director of reactor regulation, arrived and became the principal public voice of the response. He briefed officials and reporters while the government assessed whether a hydrogen bubble inside the reactor could explode; it could not.¹²

The health and institutional consequences diverged sharply. About two million people lived nearby and received, on average, roughly 1 millirem of additional radiation—about one-sixth of a chest X-ray. Comprehensive studies found no detectable health effects among workers or the public.¹¹ The accident nevertheless exposed shortcomings in design, operator training, instruments, emergency planning and information-sharing. The response reshaped the industry: improved training and instruments, continuous emergency-response capacity, resident inspectors at every plant and the Institute of Nuclear Power Operations, an industry body through which operators reviewed one another.¹¹

The cost of trust enters the bill

These changes improved safety, but they also raised construction costs. Plants already under construction required redesigns, often in the middle of projects. Schedules lengthened; with interest rates high, every month added financing costs that would ultimately be reflected in customers' bills. Three Mile Island physically damaged one reactor. Economically, it increased the cost and uncertainty facing many reactors still being built. Public trust had become a cost of nuclear electricity.

Chernobyl

On 26 April 1986, a test at Unit 4 of the Chernobyl plant in Soviet Ukraine went catastrophically wrong. The RBMK was a graphite-moderated channel-reactor design with no Western equivalent. In certain conditions, its power could surge rather than fall, and the plant lacked a robust containment building. The explosion and fire released radioactive material across Europe. Emergency workers died of acute radiation sickness within months; whole towns were evacuated and never reoccupied; and thousands of thyroid cancers were later diagnosed among people exposed as children.¹³

Technically, Chernobyl and Three Mile Island had little in common. One involved a Soviet design operated within a secretive system; the other was a Western light-water reactor whose containment largely worked. Politically, however, both became part of the same public narrative. Support for new reactors fell sharply across Western Europe and North America and remained weak for years.

For investors, the distinction matters. Safety improvements after each accident were real, but they could not eliminate political risk. A plant's economic life depends on a social licence tested by accidents anywhere in the world, elections, licence hearings and disputes over waste. Nuclear assets therefore carry both a physical operating life and a political one.

The mine that financed its own bust

Uranium markets offered a related lesson. High prices in the late 1970s encouraged new mine development. Reactor orders were then cancelled, but the mines still entered production, helping to keep uranium prices depressed through the 1980s and 1990s—often below the level needed for most mines to cover costs.¹⁴ The end of the Cold War added supply. From 1993, uranium from dismantled Soviet weapons reached Western reactors as diluted fuel, extending the oversupply for years.¹⁵

The pattern recurs throughout the industry. A mine can take a decade to plan, permit and build, leaving it to open into whatever market exists by then. The boom that financed it may already have become the glut it helps deepen.

What the accidents did not do

“The accidents killed nuclear power” is a familiar summary, but the record is more uneven. France continued operating and expanding its fleet. Asian programmes—particularly South Korea's and, later, China's—grew. The US built few new reactors for decades, yet existing operators substantially improved fleet reliability after Three Mile Island. The narrower conclusion is more useful: accidents changed where nuclear could be built, how it had to be built and the financial terms on which it could proceed. Countries with strong state support, standard designs and repeat orders continued building. Those reliant on private utilities in liberalised markets largely stopped. The next disaster would divide the industry again, along lines that still shape investors' choices.

↑ Back to Top

Fukushima, cheap gas and the long winter

On 11 March 2011, a magnitude-9 earthquake struck off north-eastern Japan. The reactors at Fukushima Daiichi shut down as designed. The ensuing tsunami overtopped the sea wall, flooded backup diesel generators and cut the plant off from electricity. Without power, the pumps stopped.¹⁶

What followed was the problem nuclear engineers had long emphasised: a reactor does not become cold when fission stops. Radioactive fission products continue to decay, producing heat that initially equals several per cent of full power and declines over days and weeks. At Fukushima, cooling failed; the cores of three reactors melted, and hydrogen explosions damaged reactor buildings.¹⁶ More than 100,000 people were evacuated.¹⁶

A company and a country reset

The plant belonged to 東京電力ホールディングス Tokyo Electric Power Company $9501.T (TEPCO), Japan's largest utility. The accident reshaped its ownership, finances and regulatory setting. Japan created a new independent nuclear regulator, shut its entire fleet and permitted restarts only after safety reviews that took years for each unit.¹⁶ TEPCO's effort to return reactors at Kashiwazaki-Kariwa to service illustrates how an asset can remain economically stranded after its engineering work is complete, pending regulatory approval and public acceptance. In the year to March 2026, TEPCO generated roughly $42 billion in revenue but reported a net loss; its market value was about $5.3 billion.⁸

Utilities that restarted reactors demonstrated the converse. 関西電力 Kansai Electric Power $9503.T generated about 49 terawatt-hours (TWh) from restarted units in its 2024 financial year, and its net margin reached 9.4% in the year to March 2026.⁸ In Japan, a restart can turn an idle asset back into a cash-generating one.

The world splits

Countries drew different conclusions from Fukushima. Germany accelerated its nuclear exit. Belgium's government and ENGIE $ENGI.PA, owner of the country's reactors, debated closures and extensions for years. Spain set a phase-out timetable that Iberdrola $IBE.MC and Endesa $ELE.MC, the main reactor owners, must plan around.⁸ China briefly paused approvals, reviewed safety and then resumed its state-directed building programme. Nordic fleets, including those of Fortum $FORTUM.HE, and central-European operators such as ČEZ continued operating and gradually extended reactor lives.⁸

The American winter was about price

The American downturn had a different cause: economics. The shale boom produced abundant cheap natural gas as electricity demand flattened. In wholesale markets, the last gas plant needed to meet demand usually sets the price for all generators. Cheaper gas therefore reduced the revenue earned by nuclear plants, even though they produced carbon-free electricity around the clock. Markets provided no additional payment for either attribute. Some safe, licensed plants closed because they were unprofitable, including Three Mile Island Unit 1—the undamaged twin of the 1979 reactor—which shut in 2019 for economic reasons.¹⁷

The second uranium bust

Uranium experienced a parallel boom and bust. Between 2003 and 2007, spot prices rose about thirteen-fold as traders and utilities anticipated a “nuclear renaissance”, drawing capital into exploration and mine development.¹⁵ Prices recovered through 2009, then remained weak for much of the following decade as Fukushima and cheap gas reduced expectations for reactor demand.¹⁴ Producers mothballed mines. Paladin Energy's Langer Heinrich mine in Namibia entered care and maintenance and restarted only after the market recovered. In the year to June 2026, Paladin's net margin was still only 1.8%, indicating how slowly a restarted mine can recover its fixed costs.⁸

The period also clarified how uranium is sold. Most supply moves through direct term contracts between producers and utilities, typically lasting three to fifteen years. The spot market, which drives headlines, represented only a small share of supply for most of this period and still accounts for only about a quarter.¹⁴ Spot prices signal sentiment; term contracts determine most industry revenue.

Plants that survived became different businesses

Reactors that survived the downturn became economically distinct assets. Extending a plant's life can resemble refurbishing a paid-off bridge: maintenance is costly, but usually cheaper and quicker than constructing a replacement because the site, permits and grid connections already exist. The analogy is incomplete. A reactor also accumulates spent fuel, requires security and depends on enriched-uranium supplies.

Policy eventually addressed what wholesale markets had not priced. Several states created payments for zero-emission generation. The Inflation Reduction Act, signed on 16 August 2022, then created a federal production tax credit for existing nuclear plants under Section 45U. It pays more when electricity prices are low, helping establish a revenue floor.¹⁸¹⁹ The policy recognised that firm, carbon-free generation may be worth more to a grid than its hourly energy price alone suggests.

Who owns the survivors

Much of the US fleet belongs to diversified or regulated utilities, whose accounts rarely isolate nuclear profitability. Public Service Enterprise Group $PEG owns nuclear plants in New Jersey and reported a 17.3% net margin in 2025. Duke Energy $DUK operates 11 reactors at six sites in the Carolinas, totalling about 10.8 gigawatts. Southern Company $SO, Dominion Energy $D, NextEra Energy $NEE, Entergy $ETR, PG&E $PCG with its Diablo Canyon life extension, Ameren $AEE with Callaway, DTE Energy $DTE with Fermi 2, and Xcel Energy $XEL also own reactors.⁸ Nuclear can represent a substantial share of their generation, but none discloses enough to treat it as a pure-play nuclear investment. Rate cases, wildfires, gas networks and renewable portfolios can matter as much to their valuations.

Testing the recovery

Did subsidies and life extensions establish that nuclear pays? Only in part. Support that preserves a valuable plant can transfer returns to ratepayers or taxpayers rather than shareholders. Regulated utilities exchange some upside for lower risk. An extension or restart generates revenue from a plant whose original construction cost was incurred long ago, offering little evidence that a new reactor bearing its full capital cost would earn an adequate return. The long winter established that many existing reactors were worth retaining. It did not establish that new ones were worth building. In February 2022, the industry's central question shifted again: from whether nuclear was clean to who controlled its fuel.

↑ Back to Top

A fuel system built for peace becomes a strategic choke point

When Russian troops crossed into Ukraine in February 2022, utilities in Europe and North America confronted an awkward dependency: Russia's state nuclear industry supplied much of the world's conversion and enrichment capacity, as well as fuel for VVER reactors in Central and Eastern Europe.²⁰ Contracts that had appeared commercial became strategic liabilities. Governments began treating non-Russian conversion, enrichment and fuel supply as matters of national security.²⁰

Walking the chain

The value chain begins with yellowcake sold under long-term mining contracts. Converters turn it into UF₆ gas, and enrichers raise its uranium-235 content. Enrichment is sometimes compared with concentrating coffee: natural uranium contains only a small share of the useful isotope, while enrichment raises that share. But the process is far more exacting. The isotopes differ in mass by barely 1%, so centrifuges spinning at extraordinary speeds separate molecules through long cascades under international inspection. The industry sells this service in separative work units, or SWU—a measure of effort, not material. Fabricators then turn the enriched uranium into pellets, rods and assemblies certified for a particular reactor.

Utilities pay for each stage, but fuel remains a small part of a reactor's operating cost. That gives qualified suppliers leverage: the cost of a fuel disruption far exceeds the cost of paying for dependable supply.

The miners: a crowded field

Uranium mining is the most visible and crowded part of the chain. National Atomic Company Kazatomprom, Kazakhstan's national uranium producer, is the largest. It uses in-situ recovery, pumping solutions through ore bodies in Kazakhstan to bring dissolved uranium to the surface rather than digging pits or tunnels. It produced 35.1 million attributable pounds in 2025, about a fifth of the world's primary uranium. It sells to reactor operators in Asia, Europe and the Americas under long-term, short-term and spot contracts.⁴⁷ Revenue slipped about 3% in 2025, to $3.4 billion. Its net margin was 31.6%, and EBITDA was just over half of revenue. It puts its all-in sustaining cost at about $30 a pound, less than half the benchmark price of $69 in the second quarter of 2026.⁸ Outside investors hold it mainly through depositary receipts listed on the London Stock Exchange.⁴⁸ On 24 September 2026 its market value was about $17.6 billion, roughly 15 times trailing earnings.⁸ Its strengths are low costs and scale; its risks include sulphuric-acid supply, a location between Russia and China, and buyers' geopolitical concerns.⁴⁷

Cameco is the leading listed Western producer, with Canadian mines among the world's highest-grade. Its EBITDA margin rose from 10.0% in 2021 to 33.2% in 2025, while free-cash-flow margin reached 30.9%.⁸ In November 2023, Cameco and Brookfield completed their purchase of Westinghouse, leaving Cameco with a 49% stake.²¹ The transaction broadened its earnings beyond uranium prices to fuel fabrication and reactor services for an installed reactor fleet. Investors have recognised that shift: on 24 September 2026, Cameco traded at roughly 155 times trailing earnings, about ten times Kazatomprom's multiple, with a free-cash-flow yield below 1%.⁸ A strong business may therefore offer less obvious valuation upside.

Other producers show how varied “uranium exposure” can be. BHP's Olympic Dam mine in South Australia produced uranium as a copper by-product—about 8.2 million pounds in the year to June 2025—material to global supply but not to BHP's overall earnings.⁸ Uranium Energy $UEC, Energy Fuels, Ur-Energy, enCore Energy and Boss Energy operate in the United States and Australia on a much smaller scale. Boss's Honeymoon operation produced 1.4 million pounds in its 2026 financial year.⁸ Uranium Energy describes itself as an exploration-stage issuer; in the year to July 2025, its $66.8m of revenue was exceeded by its net loss.⁸ NexGen, Denison and Deep Yellow own deposits rather than operating mines, making them options on future supply rather than comparable producers. Uranium Royalty holds royalties, streams and physical uranium. Its revenue jumped in the year to April 2026 largely because it sold inventory, an activity unlike operating a mine.⁸

The price, and why it misleads

The benchmark uranium price averaged $69 a pound in the second quarter of 2026, 22% above a year earlier.⁸ That did not translate reliably into miners' margins. Empor's analysis tested seven producers against year-on-year uranium-price changes, allowing for lags of up to four quarters, and found no consistent relationship.⁸ The two largest, Kazatomprom and Cameco, showed no link across six years of results. NexGen and Uranium Energy were the only companies whose margins moved with prices, but their small revenues meant that minor sales changes produced large swings. Uranium Royalty moved in the opposite direction, probably because inventory-sale timing dominated its results. The sample covers only a few years, making it evidence rather than proof.

Contracts explain much of the disconnect. Realised prices reflect agreements signed years earlier, some fixed and some indexed to market prices within floors and ceilings. Delivery schedules, mine restarts and cost inflation also matter. A higher spot price may improve the next contract round, but it usually affects current income only at the margin. The important test is whether a miner can deliver contracted pounds at durable costs while retaining some uncommitted output for a stronger market. Kazatomprom has scale and low costs; Cameco offers Western supply and integration. The data does not establish which will capture more of any further price rise.

Demand remains a scenario rather than a forecast. The World Nuclear Association estimates that reactors required 68,920 tonnes of uranium in 2025. Its reference case puts demand above 150,000 tonnes by 2040, while its low case is just above 107,000 tonnes—a gap that depends on reactors actually being built.²² Primary mine supply met 90% of reactor requirements in 2024, up from 78% in 2022, narrowing the shortfall once filled by stockpiles and former weapons material.²²

The tighter bottleneck: conversion and enrichment

The narrower bottlenecks come after mining. Outside Russia and China, conversion and enrichment are dominated by two groups. Orano, controlled by the French state, has roughly 7.5m SWU of enrichment capacity and 14,000 tonnes of conversion capacity. Its 2025 revenue was about €5.1 billion.⁸ Urenco, owned by the British and Dutch governments and two German utilities, reported 2025 revenue of €2.1 billion, EBITDA of €804m and an order book that rose 14% to €21.3 billion, extending into the 2040s.²³ EBITDA equalled nearly two-fifths of revenue, illustrating the returns scarce capacity can support.

Their protection lies in licensed plants, centrifuge know-how, safeguards obligations, years of customer qualification and government backing. None can be acquired or replicated quickly. Both are expanding. Urenco brought three centrifuge cascades into operation at Eunice, New Mexico, in 2025 and is building capacity in Germany and the Netherlands.²³ It reported spot enrichment prices rising from about $193 per SWU at the end of 2024 to about $200 a year later.²³

HALEU and the one American enricher

Centrus Energy $LEU holds an unusual position. For years, it mainly traded enriched uranium, including material purchased from Russia's state supplier. It now operates the only new US-owned enrichment cascade, at Piketon, Ohio.²⁴ That demonstration cascade began producing high-assay low-enriched uranium, or HALEU, in October 2023.²⁵ By June 2025, Centrus had delivered 900 kilograms to the Department of Energy.²⁶ Its 2025 revenue was $449m and operating margin was 11.2%. At about 59 times earnings, the valuation reflects its strategic position more than current profits.⁸

HALEU contains between 5% and 20% uranium-235—above the usual fuel for large reactors but well below weapons grade. Many advanced-reactor designs require it to operate longer between refuelling or in smaller configurations. But it is not essential to every small-reactor design, several of which use ordinary fuel. Nor does producing HALEU complete the fuel chain: it must still be converted into metal or oxide, fabricated into qualified fuel and licensed for a specific reactor. Nine hundred kilograms from a demonstration cascade is a pilot-scale result, not a commercial market.²⁶ Urenco does not plan to begin constructing Europe's first commercial HALEU plant, at Capenhurst in England, until 2028.²³

Laser enrichment could become a challenger. Silex Systems' SILEX technology, being developed through Global Laser Enrichment, and ASP Isotopes' Quantum Enrichment unit propose using lasers rather than centrifuges to separate isotopes.⁸ Neither has demonstrated commercial production, leaving both as technology options.

Fabrication and the VVER problem

At the end of the chain, fuel is made to order. Framatome, BWX Technologies and developers such as Lightbridge, whose advanced fuel remains in development, operate in this part of the market.⁸ Russia's role is clearest here. VVER reactors in Central Europe and Ukraine were designed around Russian fuel. Replacing it requires more than a new supplier: substitute assemblies must be designed, tested and licensed as part of the reactor's safety case. That is why Framatome's and Westinghouse's work on non-Russian VVER fuel has taken years.⁸

The ban and its waiver

Washington acted in May 2024. The Prohibiting Russian Uranium Imports Act, signed on 13 May, banned Russian low-enriched-uranium imports from 11 August 2024. It allowed waivers through 2027 where no alternative was available.²⁷ The waiver acknowledged that Russia's role could not be eliminated immediately without risking fuel shortages at reactors. The Department of Energy paired the ban with federal funding for domestic conversion and enrichment.²⁰²⁸

Is scarcity the same as profit?

Scarce conversion and enrichment capacity can support strategic pricing. Urenco's expanding order book and firmer SWU prices support that case.²³ But much of the business is sold through long contracts at prices agreed years earlier, and some costs pass through to customers. Empor found no reliable link between uranium prices and Centrus's gross margin, consistent with a business driven more by contract terms than commodity prices.⁸ Orano's revenue fell by about an eighth in 2025 even as the market tightened.⁸ Scarcity is real, but its conversion into profit depends on contract renewals that move slowly. The suppliers with the greatest apparent pricing power—Urenco and Orano—are not listed. The fuel chain is being rebuilt; the next question is whether its suppliers can support a renewed push to build reactors.

↑ Back to Top

The factory dream meets Vogtle's bill

On 29 April 2024, Unit 4 at Plant Vogtle near Waynesboro, Georgia, entered commercial operation.²⁹³⁰ Its twin, Unit 3, had begun commercial service the previous July—the first newly built US reactor in more than 30 years.³¹ Together, the units added about 2,234 megawatts of carbon-free capacity.³²

Vogtle showed that a Western nuclear project could still be completed. It also showed the scale of coordination and capital required: construction began in 2009, and the two units took about 15 years to finish.³²

A licensed design is not a repeatable project

Both units used Westinghouse's AP1000, an NRC-licensed design intended to be simpler and more modular than earlier reactors. Georgia Power, a Southern Company subsidiary, led the ownership group, and federal loan guarantees provided substantial support.³¹ The distinction between a licensed design and a repeatable project is central. A licence establishes that a design meets safety requirements; it does not establish that first-time crews, suppliers and managers can deliver it on time.

At Vogtle, they did not. Engineering was incomplete when construction began. Modules arrived late or required rework, while a supply chain that had not built a reactor in a generation had to be reassembled. Westinghouse had accepted construction risk and entered bankruptcy before Brookfield acquired it and later sold a 49% stake to Cameco.²¹

Who carries the risk

The episode shows how reactor construction distributes risk. A vendor designs the technology; an engineering, procurement and construction (EPC) contractor assembles the project; heavy manufacturers forge reactor vessels and steam generators. The utility, its customers or the state generally bears most financing risk. A fixed-price construction contract can appear profitable at signing but become ruinous when engineering, labour or supply assumptions fail—as Westinghouse's did. The implication remains relevant: an order benefits a supplier only when contract terms adequately protect it.

Programme builders

Elsewhere, construction is organised as a programme rather than a one-off project. Rosatom combines design, fuel supply, financing and overseas construction with Russian state backing. In China, CNNC, CGN and the listed contractor 中国核工业建设股份有限公司 China Nuclear Engineering & Construction operate within a domestic programme that continues to place orders. Equipment is supplied by makers including 东方电气 Dongfang Electric $600875.SS and 哈尔滨电气 Harbin Electric. Korea relies on KEPCO Engineering & Construction for design, KEPCO Plant Service & Engineering for maintenance and 두산에너빌리티 Doosan Enerbility $034020.KS for reactor vessels and other heavy components. In Japan, 三菱重工業 Mitsubishi Heavy Industries $7011.T and 日立製作所 Hitachi $6501.T sustain engineering and service capabilities through reactor restarts. In India, Bharat Heavy Electricals $BHEL and Larsen & Toubro $LT supply a separate national programme.⁸

Western participants are more specialised. GE Vernova $GEV, through GE Hitachi, designed the BWRX-300 small boiling-water reactor selected by Ontario Power Generation for Darlington. The IAEA reactor database lists Darlington's first unit among construction starts in 2026.³³⁸ Curtiss-Wright $CW supplies safety-critical pumps, valves and controls. Mirion Technologies $MIR makes radiation-detection and measurement equipment; revenue in its Nuclear & Safety segment rose about 9.5% in 2025, to $615m.⁸ Fluor $FLR is an engineering contractor and NuScale investor. It performed front-end engineering for a proposed six-module NuScale plant in Romania, but is not a reactor maker.³⁴

Several of these groups—GE Vernova, Curtiss-Wright, Hitachi, BHEL and L&T—do not report nuclear results separately. An order announcement alone therefore reveals little about the contribution of nuclear work to earnings.

What the numbers say about execution

Empor's scorecard suggests that construction margins remain thin. Listed design-and-construction companies that disclose revenue generated about $66 billion combined in their latest year, with a revenue-weighted net margin of 4.3%.⁸ Mitsubishi Heavy Industries lifted its group operating margin from 2.6% to 9.0% over four years, indicating that execution can improve returns. Its disclosures do not isolate nuclear earnings, however, and gas turbines and defence also contributed.⁸ China Nuclear Engineering & Construction illustrates the opposite problem: revenue reached about $14 billion in 2025, but net margin was 1.3% and free cash flow was negative in each of the previous five years.⁸ Backlog can provide revenue visibility without generating attractive cash flow.

Share prices have risen faster than the underlying businesses. Doosan Enerbility's market value increased roughly fivefold after 2024, to about $38 billion, even as its operating margin fell to 4.1%; it traded at almost 290 times earnings.⁸ Mitsubishi Heavy's value rose from about $11 billion at its 2022 year-end to about $83 billion.⁸ Those valuations assume a substantial future order cycle and leave less tolerance for delays or weak execution.

Empor also tested whether data-centre spending by the largest power buyers appeared in suppliers' revenue two to four quarters later. It did not. The relationship was negative at Mitsubishi Heavy and China Nuclear Engineering and absent elsewhere.⁸ That result is unsurprising: reactor orders take years to translate into revenue, and Chinese builders serve a domestic programme unrelated to US cloud budgets.

Do small reactors solve construction?

This leads to the modern proposition. If large reactors are difficult to build, can smaller ones be made in factories? Small reactors themselves are not new: Obninsk and Shippingport were small. What is new is the claim that serial production can turn factory-built modules, shipped and installed as equipment, into lower-cost power plants. That remains a hypothesis about repetition. No commercial-scale factory has yet established a cost curve that proves it, although investors have already priced in the prospect of factory economics before customers have received factory-built power.

↑ Back to Top

The small-reactor promise and its first cancellation

In November 2023, Utah Associated Municipal Power Systems, a group of small public utilities in the American West, cancelled the Carbon Free Power Project. It had planned to build NuScale Power reactors in Idaho, the first of their kind in the United States. Costs had risen, too few members committed to buy the electricity and the project ended.³² The first planned commercial SMR project in America was cancelled before construction began. Subsequent small-reactor proposals must address that result.

The factory proposition

The case for small modular reactors (SMRs) is that they could be built more like railcars than cathedrals: smaller units, a standard design and more factory work, with costs declining as production scales. The comparison breaks down at the site. Each plant still requires a licence, civil works, grid connection, fuel supply, trained operators, security and financing. Those costs do not fall in proportion to reactor size.

The approval that did not create a market

NuScale reached regulatory milestones first. On 11 September 2020, the NRC approved its original design, the first small modular reactor design to reach that point in the United States.³⁵ In May 2025, it approved a larger, 77-megawatt module.³⁴ Those were regulatory achievements, not commercial proof. NuScale generated $31.5 million in revenue in 2025 and recorded a net loss more than eleven times that amount. On 24 September 2026, its market value was about $2.5 billion, down from roughly $5.2 billion at the end of 2025, and its shares traded at around 166 times sales.⁸ At that stage, the sales multiple reflected investors' willingness to fund future development rather than a comparison with an operating business.

A field of different bets

Companies grouped as “SMRs” are pursuing different business models. Oklo $OKLO plans to own small fast reactors and sell electricity rather than reactors, taking operating risk alongside any operating margin. It generated no revenue in 2025. At the end of June 2026, it held about $1.6 billion in cash and another $0.8 billion in marketable securities, while its market value was about $6.6 billion.³⁶⁸

TerraPower, the private company backed by Bill Gates, is building the Natrium sodium-cooled fast reactor at Kemmerer, Wyoming, beside a coal plant. The reactor is rated at 345 megawatts, with molten-salt storage intended to lift output to 500 megawatts. On 9 March 2026, the NRC issued the first construction permit for a commercial reactor not cooled by light water. Completion is expected in 2030, with Department of Energy cost-sharing.³⁷ Natrium requires HALEU fuel, tying its schedule to the supply chain described in the previous chapter.²⁵

X-energy pairs its Xe-100 gas-cooled reactor with its own TRISO fuel—uranium kernels sealed in ceramic layers—and has Amazon's backing.³⁸ Holtec combines SMR plans with an existing spent-fuel storage and plant-decommissioning business. Its SMR-300 completed the second step of Britain's Generic Design Assessment in March 2026. Rolls-Royce SMR, a subsidiary of the aero-engine maker, completed the first two steps and received the UK's formal regulatory-justification decision.⁸ Terrestrial Energy's molten-salt reactor and NANO Nuclear Energy's KRONOS microreactor remained development projects. The NRC accepted a construction-permit application for a KRONOS unit at the University of Illinois.⁸

The race hidden under the reactor race

The decisive constraint may be fuel. Several advanced designs require HALEU; others need specialised fuel forms, coolants or fabrication lines that do not yet exist at scale. A licensed reactor without qualified, deliverable fuel resembles a certified aircraft without an approved engine supplier: permitted in principle but unable to operate.

Dead, alive, or unproven?

The evidence cuts both ways. NuScale's cancellation showed that regulatory approval does not create a market. Construction of Darlington's BWRX-300, TerraPower's permit and site work in Wyoming, and government support in Canada, Britain and the United States show that development continues.³⁷³³ No company has yet established itself as the commercial leader in small reactors. NuScale leads in US design approvals, TerraPower in permitted construction of an advanced design, and GE Vernova in a Western SMR project under construction for a utility customer. None has delivered power at a known cost, much less demonstrated a second or tenth unit.

What AI can and cannot buy

AI changes the financing discussion. A technology company can sign a long-term power contract, provide development funding or offer a project a creditworthy customer—the element Utah's municipal utilities lacked. Amazon's backing of X-energy and Google's contract with Kairos Power illustrate that role.³⁸³⁹ A buyer cannot turn a memorandum into a licence, a first unit into a factory production line or pilot-scale HALEU into a dependable fuel market. Capital can shorten some queues, but it cannot eliminate the underlying steps. For AI buyers, the nearer opportunity lies in reactors already licensed, connected to the grid and able to supply clean, round-the-clock power.

↑ Back to Top

The power plant becomes an AI asset

On 20 September 2024, Constellation Energy $CEG announced plans to restart Three Mile Island Unit 1—the reactor that operated for 40 years after its twin's accident before closing for economic reasons in 2019. It would be renamed the Crane Clean Energy Center. The restart depended on a 20-year agreement with Microsoft $MSFT to buy its output. Constellation expected the plant to return in 2028, subject to approvals.¹⁷ Microsoft describes its capacity as 835 megawatts.⁴⁰ The site of America’s most consequential nuclear accident had become a test of whether a corporate buyer could restore value to a retired reactor.

What a contract changes

A power-purchase agreement (PPA) is a long-term commitment to buy a plant’s output on agreed terms. It resembles a tenant signing a 20-year lease before a building is renovated: the commitment can support borrowing and investment, but it does not secure planning permission, repair the building or connect utilities. The Crane agreement likewise leaves NRC approvals, restart engineering and grid connection as risks for Constellation to manage.⁴¹

The chain today, from mine to server

The chain now extends from mine to data centre. Miners in Kazakhstan, Canada, Australia, Namibia and the United States sell yellowcake under term contracts. A small group of converters and enrichers—Orano, Urenco, Rosatom, China’s state suppliers and, at pilot scale, Centrus—turn it into enriched material. Fabricators including Framatome and Westinghouse produce certified fuel assemblies. Reactor owners then sell electricity, capacity and clean-energy attributes to grid operators, regulated customers or, increasingly, individual corporate buyers.

Owners fall into three broad groups. State champions include EDF in France, with about 69 gigawatts of nuclear capacity and €113 billion of 2025 sales; KEPCO, whose Korean fleet sits within a regulated national utility that earned $69 billion of revenue in 2025 and traded at 2.5 times earnings; and CGN and 中国核能电力股份有限公司 China National Nuclear Power $601985.SS in China.⁸ Merchant generators, mainly in the United States, sell into wholesale markets. Regulated utilities recover costs through customer rates.

Cash flows in the other direction. Ratepayers, wholesale buyers and technology companies pay reactor owners, which then pay for fuel, operations, maintenance and debt service. Fuel is a small share of total costs. Bargaining power therefore rests with owners of scarce, licensed reactors near constrained grids and with qualified fuel-service providers. Mines, construction contractors and prospective reactor sellers generally have less leverage.

The Chinese operators show why “nuclear” is not a single investment category. China National Nuclear Power’s free cash flow was deeply negative in 2025 as it invested in new units, while CGN’s revenue fell almost 13%.⁸ Both serve a state-directed domestic building programme, helping explain why Empor found their revenue moving against US data-centre spending rather than with it.⁸

The American contest

The contest is clearest in the United States. Constellation owns the country’s largest nuclear fleet. It generated about 182TWh in 2025, and its reactors achieved a 93.0% capacity factor in the second quarter of 2026—producing 93% of the electricity they could have generated at continuous full output.⁸ It has used that fleet to secure long-term contracts. Beyond Microsoft, Meta signed a 20-year agreement in June 2025 for 1,121 megawatts from the Clinton plant in Illinois, beginning in 2027 and supporting the plant’s long-term operation.⁴² Constellation’s market value rose from about $14 billion at the end of 2021 to about $94 billion on 24 September 2026.⁸

Talen Energy $TLN owns most of the Susquehanna plant in Pennsylvania and expanded an agreement to supply Amazon $AMZN with 1,920 megawatts of nuclear power through 2042.⁴³ Vistra $VST, which acquired Energy Harbor’s nuclear fleet in 2024, signed a 20-year agreement with Amazon Web Services for up to 1,200 megawatts from Comanche Peak in Texas. It also signed 20-year agreements with Meta covering more than 2,600 megawatts of energy, capacity and uprates from plants in the PJM region, supporting efforts to extend those reactors’ licences for another 20 years.⁴⁴ Regulated owners including Dominion, Duke and Southern retain their plants for different reasons, earning returns through rate cases rather than market prices.

Contracts, graded by weight

These agreements carry different degrees of commitment. Microsoft–Constellation, Meta–Constellation, Amazon–Talen and Vistra’s agreements are dated, long-term contracts for operating or restarting plants. Amazon’s arrangement with Energy Northwest in Washington State funds development of an initial 320-megawatt X-energy project and gives Amazon rights to buy its output: a meaningful commitment to a plant that does not yet exist.³⁸ Alphabet $GOOGL, through Google, agreed with Kairos Power to enable up to 500 megawatts of advanced nuclear capacity by 2035. That is meaningful, but it remains a future project.³⁹ Google is also helping Elementl Power prepare three US sites for advanced reactors of at least 600 megawatts each; that is site development, not a reactor order.⁴⁵ Amazon’s arrangement with Dominion is a memorandum of understanding to explore SMRs in Virginia.⁴⁶ Meta Platforms $META has issued a request for proposals for one to four gigawatts of new nuclear capacity. It signals demand, not contracted generation.⁴²

What the data does and does not show

The contracts coincide with a sharp rise in data-centre investment. Combined capital expenditure at Microsoft and Meta reached $65.9 billion in the second quarter of 2026, up 96% from a year earlier. Both companies were spending about 35% of revenue on capital projects, roughly three times Microsoft’s 2022 rate.⁸ Empor tested whether that spending appeared in reactor owners’ revenue. The relationship was clear and positive for Talen, Dominion and Public Service Enterprise Group; it was loose for Constellation. No relationship appeared for Duke, Southern, NextEra or TEPCO, while it ran in the opposite direction for the Chinese operators and KEPCO.⁸ Tests against Henry Hub gas prices, which might be expected to raise merchant nuclear margins, found no consistent relationship.⁸

The analysis covers only a few years of consolidated results, with many competing influences. It is evidence that the effect may be emerging, not proof that AI spending broadly raises nuclear earnings. Where a link appears, it is in merchant generators with direct contracts—the businesses where the effect should emerge first.

Testing the central claim

AI demand first increases the value of existing reactors because they are the clean, firm power source that can be delivered within the few years required to build a data centre. For new reactors, the proposition holds only under stricter conditions. A buyer’s commitment must be binding, dated and backed by strong credit; it must also match a site, licence, fuel plan and capital structure. The Department of Energy describes siting nuclear-powered data centres as slow and complex.⁴¹

Shareholders face a further constraint. Constellation, Talen and Vistra are valued partly on contracts already signed. The nuclear theme may remain intact while their shares disappoint if future contracts arrive more slowly or at lower prices than investors expect.

So who keeps the profit as the chain expands, and which signals would show the answer before the income statements do?

↑ Back to Top

The winners are decided before the reactor starts

Three Mile Island captures the industry's full bargain. In 1979, Unit 2 exposed failures in instruments, training and trust. Its neighbouring Unit 1 is now being prepared for restart under a contract with a software company. Between those events lie the conditions required for a reactor to return: public consent, engineering, regulation, contracted demand and capital.

The answer

Nuclear's return does not create a single pool of profit. The early beneficiaries are likely to be owners of existing reactors that operate reliably and qualified suppliers of scarce conversion, enrichment and fuel services. Miners benefit only where costs, contracts and supply discipline allow them to retain higher prices. Construction companies benefit only when contracts shield them from the risks illustrated by Vogtle. Most small-reactor developers remain bets on a learning curve that has yet to be demonstrated. AI can help new reactors pay only as part of a complete package: a standard design, secure fuel, a licence and patient capital.

In the optimistic sequence, binding corporate contracts support restarts and uprates; repeat orders justify investment in factories and fuel plants; and a small number of designs become repeatable. Costs then fall through experience rather than forecasts. In the less favourable sequence, buyers meet near-term demand with gas, renewables, storage, transmission and flexible demand. Fuel and licensing remain constrained, first-of-a-kind projects slip, governments continue to support them, and shareholders fund development while customers wait.

Four signals that move before revenue

The first is output from the existing fleet. Constellation reports it quarterly, making it an early measure of outages, uprates and the dependability of operating reactors. It generated 44TWh in the second quarter of 2026, with a 93.0% capacity factor.⁸ Sustained availability alongside new contracts would support the case for existing fleets; repeated unplanned outages would weaken it.

The second is binding, dated nuclear-power contracts with creditworthy buyers. These agreements precede generation by years and distinguish committed demand from announcements. Microsoft's 835-megawatt Crane agreement, signed in September 2024, is the benchmark.⁴⁰ Further agreements specifying plants, operating dates and financing would indicate that AI demand can support investment. A pipeline dominated by memoranda and requests for proposals would suggest otherwise.

The third is commercial HALEU delivery. It will determine whether advanced reactors have fuel when they are ready. Centrus had delivered 900 kilograms by June 2025, according to the company and the Department of Energy.²⁶ Regular deliveries beyond pilot quantities would ease a central constraint. Output that remains at pilot scale as reactors enter construction would delay the small-reactor timetable.

The fourth is contracted and funded fuel-cycle capacity. Enrichers generally expand only after securing contracts, so their order books signal future supply before reactors consume it. Urenco's order book reached €21.3 billion at the end of 2025.²³ Firm, financed conversion, enrichment and fabrication capacity would support the case that new reactor commitments can be fuelled. Reactor announcements that outpace such capacity would expose the bottleneck.

The lesson of the whole history

The atom has never lacked physical power. What has repeatedly been missing is a system that can turn it into trusted, repeatable and financeable electricity. AI has introduced a large potential customer. The profit is most likely to accrue to companies that control a genuine constraint and can execute across the rest of the chain.

↑ Back to Top

Glossary

• Fission: The splitting of a heavy atomic nucleus, releasing heat and neutrons.
• Chain reaction: A controlled sequence in which neutrons released by one fission cause further fissions.
• U₃O₈ / yellowcake: Uranium concentrate, the product sold by miners.
• Conversion: Turning uranium concentrate into UF₆, the gas used in enrichment plants.
• Enrichment: Raising uranium’s share of uranium-235. The service is measured in SWU.
• SWU: Separative work unit, a measure of the work centrifuges perform to separate isotopes.
• LEU: Low-enriched uranium, the fuel used by most existing power reactors.
• HALEU: High-assay low-enriched uranium, containing between 5% and 20% uranium-235 and required by many advanced-reactor designs.
• Fuel fabrication: Making certified pellets, rods and fuel assemblies for a particular reactor design.
• Capacity factor: A plant’s actual electricity generation as a share of the output it could have produced at continuous full power.
• Life extension: Licensing and refurbishment work that allows a reactor to operate beyond its original design life.
• PPA: Power-purchase agreement, a contract to buy a plant’s electricity over a specified period.
• FOAK / NOAK: First-of-a-kind and nth-of-a-kind. The first unit tests a design; later units are intended to demonstrate repeatable delivery.
• SMR: Small modular reactor. The term describes a smaller reactor intended for modular or repeat production; it does not establish lower costs.
• VVER: A family of Russian-designed pressurised-water reactors requiring specialised fuel.

References

1. History of the IAEA — International Atomic Energy Agency ↩↩↩

2. Outline History of Nuclear Energy — World Nuclear Association ↩↩↩↩↩↩↩↩↩↩

3. The First Pile — IAEA Bulletin ↩

4. Statement to the sixty-eighth regular session of the UN General Assembly (Atoms for Peace history) — International Atomic Energy Agency ↩↩

5. Atomic Energy Act and related legislation — US Department of Energy ↩

6. The world's first civilian nuclear power plant — IAEA Bulletin ↩↩

7. Country Nuclear Power Profiles: United States of America — International Atomic Energy Agency ↩↩↩

8. Nuclear Power research dossier and computed tables: scorecard, trends, links, numbers to watch and results — Empor, 24 September 2026 ↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩

9. NRC facts — US Nuclear Regulatory Commission ↩

10. Nuclear Power Reactors — World Nuclear Association ↩

11. Backgrounder on the Three Mile Island Accident — US Nuclear Regulatory Commission ↩↩↩↩

12. Three Mile Island Accident of 1979 Knowledge Management Digest (NUREG/KM-0001) — US Nuclear Regulatory Commission ↩

13. Twenty Years After the Chernobyl Accident — International Atomic Energy Agency ↩

14. Uranium Markets — World Nuclear Association ↩↩↩

15. Supply of Uranium — World Nuclear Association ↩↩

16. Fukushima Nuclear Accident Update Log — International Atomic Energy Agency ↩↩↩↩

17. Constellation to Launch Crane Clean Energy Center, Restoring Jobs and Carbon-Free Power to the Grid — Constellation Energy, 20 September 2024 ↩↩

18. Congressional Record, Inflation Reduction Act text including Section 45U — US Congress, 3 August 2022 ↩

19. Energy tax provisions (R46865) — Congressional Research Service ↩

20. Russian uranium ban will speed development of US nuclear fuel supply chain — US Department of Energy ↩↩↩

21. Cameco 2025 Annual Information Form — Cameco / SEC ↩↩

22. World Nuclear Fuel Report 2025: investment in nuclear fuel cycle needed as demand for nuclear power grows — World Nuclear Association ↩↩

23. Urenco releases 2025 annual results — Urenco, 2026 ↩↩↩↩↩↩

24. Centrus Energy 2025 Form 10-K — Centrus Energy / SEC ↩

25. Small Nuclear Power Reactors — World Nuclear Association ↩↩

26. Centrus reaches 900-kilogram mark for HALEU production — US Department of Energy, 2025 ↩↩↩

27. Russian Uranium Ban Waiver Guidance — US Department of Energy ↩

28. Administration takes action to secure nuclear fuel supply chain — US Department of Energy ↩

29. Vogtle Unit 4 — US Nuclear Regulatory Commission ↩

30. Remarks delivered by Secretary Jennifer M. Granholm at the startup of Vogtle Unit 4 — US Department of Energy, 2024 ↩

31. Plant Vogtle Unit 3 enters commercial operations — US Department of Energy, 2023 ↩↩

32. Motion to intervene and request for rehearing and stay of public interest organizations (advanced-reactor and Vogtle record) — US Department of Energy ↩↩↩

33. PRIS Analytics: Power Reactor Information System — International Atomic Energy Agency ↩↩

34. NuScale Power reports second quarter 2025 results — NuScale Power / SEC, August 2025 ↩↩

35. NuScale US600 design certification and approval history — US Nuclear Regulatory Commission ↩

36. Oklo Form 10-Q for the quarter ended 30 June 2026 — Oklo / SEC ↩

37. NRC issues construction permit for TerraPower's Natrium advanced reactor — US Department of Energy, March 2026 ↩↩

38. Amazon, Energy Northwest announce plans to develop advanced nuclear technology in Washington — Energy Northwest, 16 October 2024 ↩↩↩

39. Google signs advanced nuclear clean energy agreement with Kairos Power — Google, 14 October 2024 ↩↩

40. A milestone achievement in our journey to carbon negative — Microsoft, 18 February 2026 ↩↩

41. Advantages and challenges of nuclear-powered data centers — US Department of Energy ↩↩

42. Meta and Constellation partner on clean energy project — Meta, 3 June 2025 ↩↩

43. Talen Energy expands nuclear energy relationship with Amazon — Talen Energy ↩

44. Vistra reports fourth quarter and full-year 2025 results — Vistra, 26 February 2026 ↩

45. Oak Ridge National Laboratory and Elementl Power partner to accelerate siting of nuclear — US Department of Energy ↩

46. Dominion Energy and Amazon to explore advancement of small modular reactor nuclear development in Virginia — Dominion Energy, 16 October 2024 ↩

47. Kazatomprom announces 2025 full-year financial results — National Atomic Company Kazatomprom, 2026 ↩↩

48. Joint Stock Company National Atomic Company Kazatomprom (KAP) company page — London Stock Exchange ↩