The growth of AI is turning electricity into one of the technology industry’s most urgent constraints. As data centers expand, major tech companies are looking for low-carbon power that can run continuously, and nuclear energy is moving back into the conversation.
The clearest example is reactor one at Three Mile Island nuclear power station in Pennsylvania. After lying dormant for five years, it is expected to start running again in 2028 under a deal with Microsoft, supplying the company with low-carbon electricity.
Why AI Is Changing the Power Debate
AI systems depend on data centers, and data centers depend on large amounts of electricity. The International Energy Agency estimates that electricity demand from AI, data centers, and crypto could more than double by 2026. Even its lowball estimates put the extra demand at the level of all electricity used in Sweden, while the high-usage case reaches the level of Germany.
That kind of growth matters because electricity demand in the US has been fairly flat for decades. A December 2023 report from a power industry consultancy said the era of flat power demand was over, pointing to data centers and industrial facilities as major drivers.
The same report forecasts that peak electricity demand in the US will grow by 38 gigawatts by 2028. That is roughly equivalent to 46 times the output of reactor one at Three Mile Island.
For the nuclear power industry, this shift is significant. John Kotek, senior vice president for policy development and public affairs at the Nuclear Energy Institute, said AI is drawing attention across the energy sector. He also framed the issue as part of a wider national competition, saying, “People legitimately see AI as a field of competition between the US and our global competitors.”
Why Tech Companies Are Looking at Nuclear
Nuclear power appeals to tech companies because it can provide low-carbon electricity around the clock. That distinguishes it from solar and wind, which are intermittent unless paired with energy storage.
For Microsoft, the Three Mile Island deal would provide 835 megawatts of low-carbon energy over 20 years. The company has pledged to be carbon negative by 2030, but AI-driven growth has made that goal harder. In 2023, Microsoft’s emissions increased by 29 percent compared with 2020, primarily because of new data center construction.
Microsoft is not alone in exploring nuclear links. In March, Amazon Web Services agreed to buy a data center powered by Susquehanna nuclear power station in Pennsylvania. At Carnegie Mellon University on September 18, Alphabet CEO Sundar Pichai mentioned small modular nuclear reactors as one possible energy source for data centers.
OpenAI CEO Sam Altman also has ties to the sector through his role chairing the boards of nuclear startups Oklo and Helion Energy. Together, these links show that nuclear power is no longer a distant energy topic for Big Tech. It is becoming part of the practical planning around AI infrastructure.
The Three Mile Island Question
Three Mile Island carries a complicated legacy. The power station has two reactors. The second reactor was the site of a partial meltdown in 1979 and has remained out of action since then.
Reactor one has a different history. It operated without incident until 2019, when it was taken offline for financial reasons, mainly because of competition from gas- and wind-powered electricity.
That distinction is central to the current plan. Todd Allen, chair of nuclear engineering and radiological sciences at University of Michigan, said restarting reactor one would mostly involve checking that it remains in good operating condition and ensuring that enough trained staff are available to run it smoothly.
Still, the symbolism is hard to ignore. A reactor at the site of the US’s most notorious nuclear disaster could help power the AI boom. For some observers, that connection may feel uneasy, even if reactor one itself did not shut down because of operational problems.
Why Restarting Old Reactors Is Not Enough
Bringing reactor one back online could help Microsoft, but the source article makes clear that this is not a simple model that can be repeated everywhere. Kotek said there are relatively few idle reactors that could be brought back online fairly quickly.
Instead, many plant owners are interested in extending the operating licenses of existing plants. Government incentives are part of the backdrop: the Inflation Reduction Act includes tax credits tied to electricity production at existing nuclear power plants.
But if nuclear energy is to capture a larger share of projected demand, the industry may also need new reactors. That is a major challenge. The number of operating nuclear reactors in the US peaked at 112 in 1990 and declined to 92 by 2022.
The most recently built reactors in the US, at Vogtle power plant in Georgia, took more than 14 years to build and came in at more than double the expected budget. Allen summarized the problem bluntly: “The US showed at Vogtle that we’re not very good at building plants.”
He also pointed to China as evidence that faster construction is possible. If data center electricity demand keeps growing, new nuclear plants may look more attractive, but the US would still have to confront the cost and timing challenges shown by recent experience.
Efficiency Is the Other Side of the Problem
The AI energy debate is not only about finding new power sources. Sasha Luccioni, AI and climate leader at Hugging Face, said tech companies have tended to focus more on energy supply than on improving the efficiency of AI operations.
She argued that regulation could help, starting with mandatory reporting and transparency for companies providing AI tools and services. That would shift part of the discussion from how to power AI toward how much power AI should need in the first place.
Pichai acknowledged at Carnegie Mellon University that work on AI’s energy consumption was still in its “early phases.” He said, “We are all inefficiently pretraining these models, absolutely,” while adding that inference could become “dramatically more efficient over time.”
The pressure is already visible in company climate numbers. Google’s emissions in 2023 were 48 percent higher than their 2019 baseline, mainly because of data center energy consumption and supply chain emissions. That puts Google’s goal to reach net zero emissions by 2030 under growing strain.
The central issue is timing. As Luccioni put it, “The energy demands of AI are rising right now,” while renewable or low-carbon energy is not keeping pace quickly enough. That gap explains why nuclear energy, despite its cost, legacy, and construction challenges, is getting renewed attention from the companies building the AI economy.