AI datacentres need power, not just chips
The AI build‑out is running into physical limits — power, land and capex — and investors already see data‑centre and utility constraints as central to scaling AI. That reality is visible in major infrastructure moves, including a partnership announced between NextEra Energy and TerraPower to explore small modular reactors to fuel Google and Microsoft AI facilities, which signals energy planning is becoming part of AI strategy. (marketsandmarkets.com) (markets.financialcontent.com)
The surprise in artificial intelligence right now is that the bottleneck is starting to look less like chips and more like electricity. On April 8, 2026, a release said NextEra Energy and TerraPower will explore small modular reactors for Google and Microsoft data centers, tying power planning directly to artificial intelligence expansion. (markets.financialcontent.com) A modern data center is basically a warehouse full of computers, plus cooling machines, batteries, networking gear, and backup systems that keep those computers from overheating or shutting off. The International Energy Agency says servers alone account for about 60% of electricity demand in modern data centers. (iea.org) Those buildings are getting hungrier because artificial intelligence work runs on specialized processors that stay busy for long stretches. The International Energy Agency estimates data centers used about 415 terawatt-hours of electricity globally in 2024, or about 1.5% of world electricity consumption. (iea.org) In the United States, the jump is already large enough to show up in federal projections. The Department of Energy said in December 2024 that data centers used about 4.4% of U.S. electricity in 2023 and could reach roughly 6.7% to 12% by 2028. (energy.gov) The same Department of Energy summary said U.S. data center electricity use rose from 58 terawatt-hours in 2014 to 176 terawatt-hours in 2023. It projected a further rise to 325 to 580 terawatt-hours by 2028. (energy.gov) That is why nuclear power keeps showing up in technology deals. Google said on October 14, 2024 that it signed what it called the world’s first corporate agreement to buy power from multiple small modular reactors, with Kairos Power aiming to deliver up to 500 megawatts by 2035. (blog.google) Microsoft made a similar move one month earlier, but with an older reactor instead of a new small one. Constellation said on September 20, 2024 that a 20-year power purchase agreement with Microsoft would support restarting Three Mile Island Unit 1, renamed the Crane Clean Energy Center, and return about 835 megawatts to the grid. (constellationenergy.com) Amazon also stepped in on October 16, 2024, funding early work on a small modular reactor project with Energy Northwest in Washington. The same day, X-energy said Amazon was part of a financing round of about $500 million and that the two companies want to bring more than 5 gigawatts of new projects online in the United States by 2039. (energy-northwest.com) (x-energy.com) The reactor design in the new NextEra-TerraPower plan is not a science-fiction sketch anymore. TerraPower says its Natrium design is a 345-megawatt sodium fast reactor paired with molten-salt energy storage, and the Department of Energy said in March 2026 that the Nuclear Regulatory Commission had issued the first construction permit for a commercial non-light-water reactor in the United States for that project in Wyoming. (terrapower.com) (energy.gov) The pattern across all of these deals is simple: technology companies used to shop for chips, buildings, and cloud capacity, and now they are shopping for power plants too. When Google signs for 500 megawatts, Microsoft backs an 835-megawatt restart, and Amazon chases multi-gigawatt nuclear buildouts, it means the next limit on artificial intelligence is no longer inside the server rack alone. (blog.google) (constellationenergy.com) (x-energy.com)
