The largest technology companies are turning to nuclear power to feed the enormous electricity appetite of their artificial intelligence data centers, signing agreements to buy output from existing reactors and to help finance new ones. Industry coverage describes a wave in which Amazon, Microsoft and Google have rushed to sign nuclear deals, with reporting indicating that big tech contracted for more than 10 gigawatts of potential new nuclear capacity in the United States over about a year. The specifics of who receives power first, and when, vary considerably by deal, and many of the newest reactor projects are years from producing a single watt.
Why AI is driving companies to reactors
Training and running large AI models requires dense clusters of chips that draw power continuously, and operators want electricity that is both carbon-free and available around the clock. Intermittent sources like wind and solar do not, on their own, provide that steady baseload, which is why nuclear has re-entered the conversation. As one industry primer frames it, small nuclear reactors could eventually power data centers with clean energy, though it stresses that significant challenges remain, from cost to regulatory approval, per a guide to SMRs and nuclear-powered data centers.
The appeal is straightforward: a reactor delivers large, constant output from a small footprint, matching the profile of a data campus that never sleeps. The catch is timing. Restarting a shuttered reactor is faster than building a new one, and small modular reactors, the compact designs many deals reference, are still largely pre-commercial in the U.S.
The deals on the table
According to reporting compiled from these announcements, the arrangements fall into two broad categories. In the near term, companies are contracting for power from existing large reactors. Microsoft signed a long-term power purchase agreement tied to the planned restart of a reactor at Pennsylvania’s Three Mile Island site, rebranded the Crane Clean Energy Center, with roughly 835 megawatts of capacity targeted for the late 2020s. Because that plant already exists and is being brought back online rather than built from scratch, Microsoft is positioned to be among the first to actually receive nuclear power under these agreements.
In the longer term, companies are backing new reactor construction. Google agreed to buy energy from a fleet of small modular reactors being developed by Kairos Power, with reporting describing six or seven reactors and completion targeted across roughly 2030 to 2035. Amazon, meanwhile, announced agreements in October 2024 to support reactor projects, with reporting citing an initial phase around 320 megawatts and an option to scale toward roughly 960 megawatts. Additional coverage notes Meta issuing a request for proposals for new nuclear capacity. These figures come from press reporting rather than a single audited registry, so exact capacities and timelines should be treated as targets that could shift, and analyses such as IEEE Spectrum’s look at big tech and nuclear underscore how much still hinges on financing and licensing.
What is consistent across the accounts is direction: rather than simply buying grid power, these firms are entering into direct, long-dated commitments with reactor operators and developers, which is a notable shift for an industry that has historically relied on utilities and renewables.
What it means and what remains uncertain
For the nuclear sector, tech money is a potential lifeline, providing the guaranteed demand that has long been missing for both plant restarts and next-generation designs. For the public, the trade-offs are still being worked out. New reactors face licensing reviews, supply-chain constraints and high upfront costs, and even restarts must clear regulatory hurdles before energizing. None of the small modular reactor projects tied to these deals were operating at the time of this reporting, meaning the headline gigawatt totals describe contracted potential, not delivered power.
The practical questions to watch are whether the first restart comes online near its late-2020s target, whether small modular reactors can be licensed and built on schedule and budget, and how the added demand affects electricity prices and grid reliability for ordinary customers who share the same power system. If these projects deliver, they could reshape how large-scale computing is powered; if they slip, the AI industry will keep leaning on the existing grid in the meantime. Either way, the commitments signal that energy, not just chips, has become a central constraint on the growth of artificial intelligence.
More from Morning Overview
- The NSA issues an urgent warning to all phone users about a setting most people leave on
- 5,000-year-old wolves on remote island upend domestication theory
- Scientists may have found a Type II civilization
- EPA finally kills ‘stupid feature’ in 60% of U.S. cars after 14 years of driver rage
*This article was researched with the help of AI, with human editors creating the final content.
