The 1979 accident at Three Mile Island brought U.S. nuclear construction to a halt for 40 years. Now, Microsoft is spending $16 billion to restart the plant’s sister reactor, betting that only atomic energy can keep its artificial intelligence data centers running around the clock.
Microsoft signed a 20-year, $16 billion contract for the Crane Clean Energy Center, the renamed Three Mile Island Unit 1. The plant is expected online in late 2027, and its entire output will feed Microsoft’s AI data centers. The deal is the largest single corporate nuclear investment in U.S. history.
Microsoft is not alone. Every major AI company—Microsoft, Google, Amazon and Meta—signed at least one nuclear deal in 2026. Their total commitments now amount to nearly 10 gigawatts of capacity, enough to power roughly 7 million homes. The rush marks the most significant corporate push into nuclear energy since the industry stalled four decades ago.
The AI electricity crisis
Global electricity demand is projected to grow by more than 10,000 terawatt-hours by 2035, equivalent to adding the entire current consumption of all advanced economies. Data centers will drive more than one-fifth of that growth in wealthy countries by 2030.
By the end of the decade, U.S. AI data processing alone is projected to consume more power than the aluminum, steel, cement and chemical industries combined. The U.S. grid connection queue already holds more than 2,600 gigawatts of proposed generation, and average wait times for new connections are nearing five years.
Rafael Mariano Grossi, director general of the International Atomic Energy Agency, said nuclear power is uniquely suited to the moment. “Only nuclear energy can meet the five needs of low-carbon power generation, round-the-clock reliability, ultra-high power density, grid stability and true scalability,” he told UN News.
Grossi added: “AI is the engine of the future, but an engine without fuel is almost useless. Nuclear energy is not only an option, but also an indispensable core component of the future energy structure.”
He noted that growth curves for AI and nuclear power are moving in tandem. “AI technology and the construction of artificial intelligence data centers are advancing simultaneously, and the number of new nuclear reactors in the world also ranks first in the world during the same period.”
The IAEA has revised its global nuclear capacity projections upward for the fourth consecutive year, citing AI demand as a primary driver. Grossi said the agency now expects between 500 and 700 new reactors worldwide by 2050, a sharp increase from earlier forecasts. He described the convergence of AI and nuclear as "the most significant energy development of the decade."
The deals behind the boom
Meta signed agreements worth up to 6.6 gigawatts of nuclear power, according to NucNet. One of those deals is a 20-year power purchase agreement with Constellation for the output of the Clinton Clean Energy Center, a single-reactor nuclear plant in Illinois. A photo of the Clinton plant was taken on July 25, 2025. Meta also signed a separate agreement for up to 4 gigawatts of new nuclear capacity from developers yet to be named, making it the largest single corporate nuclear commitment by volume.
Google signed deals with two nuclear developers and a separate contract for 1,800 megawatts of new capacity. Manuel Greisinger, Google’s AI infrastructure lead, said the company is pursuing multiple nuclear pathways to secure reliable, carbon-free power for its data centers. Google's agreements include a partnership with Kairos Power, a developer of molten salt reactors, and a separate deal with TerraPower, the Bill Gates-backed advanced nuclear company. Greisinger said Google expects its first nuclear-powered data center to be operational by 2030.
Amazon invested $700 million into X-energy, a nuclear reactor developer, to build up to a dozen small modular reactors. SMR Intel reported that the investment is one of the largest corporate bets on advanced nuclear technology. Amazon also secured options to purchase power from those reactors once they are built, with the first units expected to come online in the early 2030s. The company has said it plans to use the reactors to power its AWS cloud computing data centers.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
The Department of Energy pledged up to $17.5 billion in loans for 10 new Westinghouse reactors, with construction start targeted by 2030. In June 2026, the U.S. and Canada each announced plans to build 10 new nuclear reactors, the biggest coordinated nuclear push in decades. The DOE loans are part of a broader federal effort to revive the domestic nuclear supply chain, which has atrophied since the 1970s. The agency said it expects the loans to support up to 20,000 construction jobs and 5,000 permanent operating positions.
Skeptics warn of cost and timeline gaps
The last large-scale U.S. reactors completed were Georgia’s Plant Vogtle Units 3 and 4, finished in 2023 and 2024. They came online seven years late and $36.8 billion over budget—more than double the original estimate, making them the most expensive power plant ever built. The Vogtle project was originally estimated at $14 billion and was supposed to take four years. Instead, it took more than a decade and required billions in additional financing from Georgia ratepayers.
Small modular reactors, which are compact factory-built units designed to be faster and cheaper, have not completed a single commercial project in the U.S. yet. Restarts like Three Mile Island take two to three years, while new builds take roughly 10 years. The Nuclear Regulatory Commission has not licensed a single small modular reactor design for commercial operation, and the first applications are still under review.
Molly Langabeer and M.V. Ramana, researchers at the Bulletin of the Atomic Scientists, published a skeptical analysis in July 2026. They argued that small modular reactors “will be unable to provide any significant boost to the electricity consumption of data centers anytime soon—especially within the next several years, when AI-related energy demand is expected to increase most rapidly.”
They also said the tech investment so far is “completely mismatched to the actual cost of building even one reactor, let alone the dozens” that would be needed to meaningfully power the AI boom. Langabeer and Ramana estimated that building a single small modular reactor could cost between $3 billion and $5 billion, meaning the total corporate investment of roughly $20 billion across all companies would cover only four to seven reactors at best.
The Carnegie Endowment for International Peace assessed that manufacturing, licensing and workforce for new reactor designs remain unproven, and that timelines are out of sync with AI growth. Critics have described the nuclear push as nothing more than a well-funded press release. The endowment's analysis noted that the U.S. has not built a single new nuclear reactor from scratch in more than 30 years, and that the domestic supply chain for large forgings, specialized valves and reactor pressure vessels has largely disappeared.
Restarts buy time, new builds are a bet on the future
The article notes that skeptics are half right—new reactors are years away, but restarts are faster. Natural gas is filling the gap until the restarts come online. U.S. natural gas generation hit a record high in 2025, with data center demand accounting for roughly 15 percent of the increase.
The Crane Clean Energy Center restart is expected in late 2027. Constellation, the plant's owner, has already begun hiring and retraining former nuclear workers, and has placed orders for long-lead components such as steam generators and control rod drives. The company said it expects the restart to cost roughly $1.6 billion, a fraction of the cost of building a new reactor.
The DOE’s 10 new Westinghouse reactors are targeted for construction start by 2030, meaning they would not deliver power until the 2040s at the earliest. Westinghouse has said it plans to build the reactors at existing nuclear sites to take advantage of already-licensed land and cooling infrastructure. The company has identified four potential sites, including the Savannah River Site in South Carolina and the Idaho National Laboratory.
Big Tech has become nuclear power’s most important backer in a generation. The pairing of AI and nuclear may decide which companies win the AI race. It remains an open question whether the plants will arrive on Silicon Valley’s optimistic timeline. Some analysts have suggested that the tech companies may need to build their own dedicated natural gas plants as a bridge, a move that would undercut their carbon reduction pledges.