Big Tech backs nuclear power

The race to build bigger artificial intelligence systems has run into a more basic problem than chips: electricity. Reuters reported on April 10 that major technology companies are now putting money directly into nuclear projects because data centers need large blocks of steady power, not just more servers. (reuters.com) A data center is just a warehouse full of computers, and those computers draw power every hour they run. The International Energy Agency said in April 2025 that electricity use from data centers worldwide was about 415 terawatt-hours in 2024 and could more than double to around 945 terawatt-hours by 2030. (iea.org) Nuclear power appeals to these companies for the same reason a hospital keeps backup generators nearby: it runs day and night. The United States Energy Information Administration says nuclear plants supplied about 19% of U.S. electricity in 2024 and had a capacity factor above 92%, which means they produced near full power most of the time. (eia.gov, eia.gov) The new twist is that technology companies are not waiting for utilities to solve this alone. Reuters said their deals now provide reactor developers with financing and with buyers for future electricity, which makes it easier to raise the much larger sums needed to build plants. (reuters.com) Google made one of the clearest examples in October 2024 when it signed a deal with Kairos Power for up to 500 megawatts of advanced nuclear capacity by 2035, with the first deployment targeted for 2030. Kairos said it would build and operate a series of plants and sell Google electricity, grid services, and environmental attributes under power purchase agreements. (kairospower.com) Amazon moved the same direction that month with three nuclear agreements tied to its cloud computing arm, Amazon Web Services. X-energy said Amazon anchored a financing round of about $500 million, and Utility Dive reported the broader package supported more than 600 megawatts in Washington and Virginia. (x-energy.com, utilitydive.com) Microsoft chose a different route by backing an old plant instead of a new design. In September 2024, Constellation said it planned to restart Unit 1 at Three Mile Island, renamed the Crane Clean Energy Center, under a 20-year power purchase agreement with Microsoft, with restart targeted for 2028 at roughly 835 megawatts. (constellationenergy.com, politico.com) Meta first opened a bidding process in December 2024 for developers that could deliver 1 to 4 gigawatts of new U.S. nuclear capacity for its artificial intelligence and sustainability goals. By January 2026, industry outlets were describing the resulting partnerships as adding up to as much as 6.6 gigawatts across deals with Vistra, TerraPower, and Oklo. (sustainability.atmeta.com, powermag.com) That changes what “artificial intelligence infrastructure” means. Reuters said the money trail now runs past chip factories and server racks into transmission lines, reactor licensing, and the local utility paperwork needed to connect giant loads to the grid. (reuters.com) The bottleneck is not only generating power but moving it. Reuters reported this week that Constellation warned delays to transmission projects could push back the Three Mile Island restart, which shows that even a signed power deal is useless if the wires are not ready when the reactor is. (reuters.com) The catch is that most of these advanced reactors do not exist at commercial scale yet. The United States Nuclear Regulatory Commission has licensed only a small number of advanced reactor projects so far, and each plant still has to clear siting, safety, fuel, construction, and grid-connection hurdles before a data center gets a single megawatt. (nrc.gov) So the artificial intelligence boom is starting to look less like a software story and more like an industrial one. The companies spending tens of billions on chips are now also shopping for turbines, substations, permits, and reactor builders, because a model cannot answer a question if the server behind it has no power. (reuters.com)