Council of Councils
Think Global Health
Online Store
David M. Hart is a senior fellow for climate and energy at the Council on Foreign Relations.
U.S. technology companies are rapidly pushing into nuclear power as they compete to develop the next generation of artificial intelligence (AI) tools and services. Their investments, including several deals announced in late 2024, could help them meet their ambitious climate goals and revive a U.S. energy sector that has long failed to deliver on its promise. While high costs and public concern could derail these initiatives, the upside for the planet could be big if these deals help launch new power generation technologies that can be used worldwide.
What are the Microsoft, Google, and Amazon nuclear deals?
More on:
Microsoft has agreed to buy all the power produced by a previously closed nuclear reactor at Three Mile Island in Pennsylvania for twenty years after its scheduled restart in 2028. Google has struck a deal to buy power from seven so-called small modular reactors (SMRs) to be built by a start-up called Kairos Power between 2030 and 2035. These reactors would be of a new design that is smaller and intrinsically safer than the current fleet.
And most recently, Amazon announced a $500 million investment in X-Energy, which plans to build a fleet of SMRs in the 2030s. Amazon is also planning to buy power directly from Susquehanna nuclear plant in Pennsylvania during the current decade, and is working with a utility in Virginia to site the new SMRs in the future.
Why do tech firms need so much power?
In two words: artificial intelligence. These “hyperscalers”—companies that operate massive computing and data storage systems for themselves and their customers—are in a race to seize competitive advantage in building out AI. They believe AI will create massive new businesses, and being the first to market could determine their success.
AI programs run on computers housed in large-scale data centers, which are power-intensive. Estimates of how much new demand for power AI will create vary widely. A recent projection by researchers affiliated with Columbia University concluded that AI data centers will quadruple their share of U.S. electricity sales from 1 percent in 2022 to 4 percent in 2027. Data centers as a whole could use three times that share. Many analysts expect such exponential growth to continue into the 2030s.
U.S. grid operators are dealing with a surge in demand after many years of stability. Along with emerging AI technologies, new factories, electric vehicles, and the cryptocurrency boom are contributing to this growth.
More on:
Satisfying this increased demand under the existing industry paradigm is likely to be a challenge because the U.S. grid is managed by a patchwork of authorities that have difficulty coordinating their plans. Changing its speed and direction is like steering a supertanker—it’s an extremely slow process.
How could nuclear power be part of the answer?
Nuclear power could help the tech industry overcome two of its major energy challenges: its extraordinary future power demands, and its need to achieve its ambitious climate goals.
Google, for instance, has set a goal of around-the-clock carbon-free energy on every grid where it operates by 2030. The company has nearly thirty large data centers, mostly concentrated in the United States and northern Europe. To progress toward this goal, it has made deals to buy an enormous amount of electricity generated by solar and wind resources. These renewables, however, are intermittent sources of power, while data centers need a steady supply to run all the time. The tech companies are currently reliant on the grid whenever the wind isn’t blowing or sun isn’t shining. Moreover, while it is getting cleaner, the U.S. grid still gets more than half of its power from natural gas and coal plants; and it is not yet on track to hit the Joe Biden administration’s goal of eliminating emissions by 2035.
Existing nuclear plants, such as those contracted by Amazon and Microsoft, are attractive because they can provide a lot of stable, carbon-free power during the current decade. Meanwhile, SMRs, like those that will be supported by Amazon and Google, offer the promise of less expensive stable, clean power, in a few years’ time.
What are the potential roadblocks?
Nuclear power has long been expensive and unpopular in the United States. Civilian reactor technology was spun out of the U.S. weapons program in the 1950s. The majority of the nation’s approximately 100 nuclear plants began construction in the 1960s and 1970s. But safety concerns sparked local opposition, which spiraled into burdensome regulatory requirements, costly delays, and a decadeslong dry spell. Three Mile Island, of course, was the site of the most famous nuclear accident in U.S. history, where a partial reactor meltdown in 1979 punctuated the industry’s fall from grace.
These tech firms’ new nuclear power deals could bump into the same barriers. First, no SMRs have been built yet, so their promised low costs are still just theoretical. In terms of safety, while SMRs are designed to avoid the risks of older reactors, a slip-up at a new site or with an existing refurbished reactor would be a huge setback. Few communities have considered whether they would want to risk having one in their backyard. Finally, all nuclear plants produce dangerous, long-lived radioactive waste, and there is no permanent site at which to store it in the United States.
What’s the upside?
The benefits to the world could be big. The tech giants’ purchases of SMRs demonstrate that they have a high-degree of confidence that this technology will be safe and reliable, and, eventually, affordable. If that’s the case, SMRs could become a viable source of stable, clean power worldwide in the 2040s. That would be highly consequential, as the vast majority of global emissions come from outside the United States.
At the same time, the advancements that nuclear power allows these companies to make with regard to AI could also help the world overcome many of the challenges of climate change. AI advocates envision the new technology contributing to grid management, urban design, and scientific discovery, to name a few—pretty much everything humanity needs to navigate this wicked problem. Google has already built an AI-powered system that could provide better and cheaper weather forecasts and climate models.
A nuclear power renaissance may not be essential for AI to get off the ground; the tech giants are hedging their bets with other energy resources. But nuclear power’s success would help allay climate-driven doubts about AI. Then, it’ll be up to AI and the hyperscalers to live up to the hype and deliver a better future for the planet.
This work represents the views and opinions solely of the author. The Council on Foreign Relations is an independent, nonpartisan membership organization, think tank, and publisher, and takes no institutional positions on matters of policy.