Big Tech Puts Its Money Behind Nuclear: How AI's Power Hunger Is Reviving an Industry

As recently as 2021, energy analysts were writing obituaries for nuclear power. Aging plants were shutting down, construction costs were ballooning, and cheap natural gas was winning on economics. Then came the large language model. Today, four of the world’s largest technology companies have collectively committed to sourcing tens of gigawatts of nuclear capacity — enough to power tens of millions of homes — to run their AI infrastructure. The obituary has been shelved.

The tipping point arrived in earnest over the first months of 2026, as a wave of deals between hyperscalers and nuclear energy companies closed and became public. On April 10, a new report confirmed that big tech is now going beyond signing power-purchase agreements to becoming active financiers of physical energy infrastructure — providing capital to help build, extend, and certify nuclear plants that might otherwise never break ground.

The Deals Reshaping the Nuclear Landscape

Microsoft signed what was described as the largest corporate nuclear agreement in history with Constellation Energy: a commitment to purchase roughly 2 gigawatts of nuclear power through 2040. The cornerstone of that arrangement is the restart of Three Mile Island Unit 1 in Pennsylvania — the plant famously adjacent to the 1979 accident site — which Constellation expects to bring back online by 2027, delivering 835 megawatts directly to Microsoft’s data center grid. The deal gave Constellation the commercial certainty it needed to invest hundreds of millions of dollars in safety upgrades for the dormant reactor.

Meta went further, executing a trifecta of nuclear commitments that amounts to 6.6 gigawatts of clean power by 2035 — enough electricity to serve roughly five million American homes. The deals span Vistra’s Ohio and Pennsylvania nuclear plants (life extension and uprate funding), TerraPower’s Natrium advanced reactor projects (financing two initial units with options on six more), and Oklo’s microreactor campus in Ohio (1.2 gigawatts of future capacity). Combined with an earlier 20-year agreement with Constellation’s Clinton Clean Energy Center in Illinois, Meta has become one of the most consequential corporate purchasers of nuclear energy in American history.

Amazon secured commitments with X-energy, a developer of high-temperature gas-cooled reactors, targeting more than 5 gigawatts of nuclear capacity by 2039. The company also separately locked in 1.5 gigawatts of dedicated solar capacity in Texas, making clear that its grid strategy involves a diversified clean energy mix, not a single bet.

Alphabet (Google) signed an agreement with Kairos Power to bring a fluoride salt-cooled small modular reactor online by 2030, with options for follow-on units. Like its peers, Google framed the deal not simply as energy procurement but as a catalyst for commercializing nuclear technology that needs an anchor customer to clear regulatory and financing hurdles.

Why Nuclear, and Why Now

The physics are straightforward. Running a cluster of tens of thousands of AI training chips at full utilization, 24 hours a day, requires a reliable, dispatchable power source. Solar and wind generate electricity only when the sun shines or wind blows; battery storage at scale remains expensive and short-duration. Natural gas can meet the dispatch requirement, but it carries carbon liability and price volatility. Nuclear is the only large-scale clean energy source that provides always-on, carbon-free baseload power.

The scale of the demand shock is hard to overstate. Hyperscaler capital expenditure is forecast to exceed $600 billion in 2026, with roughly 75% — approximately $450 billion — directly tied to AI infrastructure. The Uptime Institute estimates that the total global data center power load specifically attributable to AI will hit 10 gigawatts by the end of 2026, constrained primarily by insufficient grid and generation capacity, not by hardware supply. In the United States, grid interconnection queues now stretch for years, and utilities have told some hyperscalers they cannot guarantee delivery within the timeframes the companies need.

That gridlock has pushed tech companies to go upstream. Rather than wait for utilities to build capacity speculatively, Microsoft, Meta, and their peers are providing contracts, capital, and in some cases direct project funding to unlock construction that the regulated utility model could not finance on its own.

The Small Modular Reactor Bet

The most forward-looking aspect of these deals is the commitment to small modular reactors — nuclear plants that are factory-built in standardized modules, then assembled on-site. SMRs are not yet commercially proven at scale. TerraPower’s Natrium design, backed by Bill Gates, and Oklo’s compact microreactor, backed by Sam Altman, are both still working through the Nuclear Regulatory Commission’s approval process. X-energy and Kairos Power are in similar positions.

What hyperscaler contracts provide is something the SMR industry has never had: a creditworthy customer willing to sign decade-long offtake agreements in advance. Without that, private capital could not confidently underwrite construction. With it, the financing model for first-of-a-kind nuclear projects shifts from the traditional utility rate base — where costs are spread across millions of ratepayers and approved by state regulators — to direct corporate commitments. It is analogous to what Amazon’s early AWS commitments did for cloud data center construction in the 2000s: providing the demand certainty that unlocked an entirely new category of infrastructure investment.

“We are moving from being customers to being partners in building out the energy infrastructure the country needs,” a senior Microsoft energy executive told Fortune in February. “The only way these projects get financed is if we step up.”

Policy and Grid Implications

The nuclear revival is not happening in a vacuum. In late 2025 and early 2026, the U.S. Congress passed bipartisan legislation streamlining the NRC’s licensing process for advanced reactor designs, cutting the typical review timeline from over a decade to as few as four years for standardized SMR designs. The ADVANCE Act, signed by the president, explicitly cited AI infrastructure demand as a driver. Several states, including Georgia, Wyoming, and Virginia, have enacted additional incentives for nuclear development in data center corridors.

The grid implications are still being worked out. Utilities and independent system operators are grappling with how to handle hyperscalers that want direct connections to nuclear plants, bypassing standard interconnection queues. PJM, the grid operator covering the Mid-Atlantic and Midwest, has opened a dedicated proceeding on the topic. Critics argue that allowing tech companies to build effectively private generation capacity erodes the cross-subsidization model that keeps electricity prices affordable for residential customers; proponents counter that the alternative is continued grid congestion and higher prices anyway.

The Energy Industry’s New Anchor Customers

For the nuclear sector, the arrival of hyperscalers as customers represents an unexpected second act. A generation of nuclear entrepreneurs who spent years struggling to convince utilities, regulators, and investors that their technology was viable now find themselves fielding inbound calls from the world’s most cash-rich companies. Oklo, TerraPower, and Kairos Power have all seen their private valuations rise sharply since 2024; the pipeline of announced SMR projects has more than doubled in two years.

The risk, as some analysts note, is that this demand concentration creates fragility. If one or two major tech companies significantly reduced their AI infrastructure buildout — due to a model efficiency breakthrough, a regulatory shift, or a market downturn — the offtake commitments that underpin these projects could be renegotiated, creating stranded cost risk for investors and host communities.

For now, however, the direction of travel is clear. AI’s electricity hunger has turned an industry that was quietly fading into one of the most actively financed sectors in energy. The atom and the algorithm have found each other, and the implications for the global electricity grid will take decades to fully understand.