The AI boom is reviving America’s nuclear dream
Google, Microsoft, Amazon and Meta are pouring money into reactors and long-term nuclear power deals as they race to secure the electricity needed to run the next generation of AI data centers.
In 2022, tech giants were among the world's leading corporate champions of efforts to reduce greenhouse-gas emissions and transition to a carbon-free economy. Meta was committed to net-zero emissions, Google was matching 100% of its electricity consumption with renewable energy, Microsoft was pursuing a plan to eliminate all the carbon it had emitted since its founding, and Amazon had pledged to reach net zero by 2040.
Then, in late 2022, OpenAI launched ChatGPT, and the economics of the technology industry's energy transition began to change dramatically.
The AI revolution sparked by ChatGPT has created enormous global demand for computing power. To meet that demand, technology companies have committed hundreds of billions of dollars to building data centers, which consume vast amounts of electricity to power advanced chips, cooling systems and other energy-intensive equipment.
As a result, the carbon emissions of some major technology companies have risen sharply as their electricity consumption has surged and, in many locations, grids have relied on fossil fuels to meet the additional demand.
But the companies have not abandoned their climate ambitions. They have also recognized that existing power grids, particularly in the United States, may struggle to accommodate the enormous amount of electricity that the AI industry is expected to consume.
At the same time, the limitations of renewable energy have become more apparent in the context of data centers. Solar and wind power are increasingly inexpensive and can provide large amounts of electricity, but they are intermittent. Data centers, by contrast, need electricity around the clock, with extremely high reliability. Batteries can help bridge some gaps, but long-duration storage at the scale required remains expensive and technically challenging.
That has pushed the technology industry back toward an energy source that had largely fallen out of favor in the United States: nuclear power.
Nuclear reactors do not produce direct carbon emissions while generating electricity, and unlike solar and wind farms, they can provide a continuous supply of power. For technology companies facing an unprecedented need for reliable electricity, that combination has made nuclear energy increasingly attractive.
It is a remarkable comeback for an industry that was once considered the energy of the future.
In the 1950s and 1960s, nuclear power was widely viewed as the foundation of a new era of abundant electricity. But major accidents, most notably Three Mile Island in Pennsylvania in 1979 and Chernobyl in 1986, along with growing environmental concerns and the unresolved problem of radioactive waste, severely damaged the industry's reputation.
In the United States, new reactor construction largely ground to a halt.
Now, the AI boom is helping revive the nuclear industry. The technology giants are investing in existing reactors, signing long-term power purchase agreements and placing bets on a new generation of small modular reactors, or SMRs, which proponents say could eventually be built faster, more cheaply and with improved safety characteristics than conventional reactors.
Big Tech goes nuclear
Virtually every major technology company with significant AI ambitions is now exploring nuclear energy in some form.
Microsoft signed a 20-year power purchase agreement in 2024 with Constellation Energy tied to the planned restart of Unit 1 at the Three Mile Island nuclear power plant. Unit 1 was shut down in 2019 for economic reasons, while Unit 2, the reactor involved in the 1979 accident, has been offline since the accident. Unit 1 is expected to return to operation in 2028 with a capacity of 835 megawatts.
Microsoft has also backed a radically different nuclear technology: fusion.
In 2023, Microsoft signed an agreement to purchase electricity from Helion Energy, a fusion company whose investors include OpenAI CEO Sam Altman. Fusion is sometimes described as the "Holy Grail" of energy because, rather than splitting heavy atoms as conventional nuclear reactors do, it seeks to generate energy by fusing lighter atoms.
Scientists have pursued commercially viable fusion for decades. The technology promises enormous amounts of energy without the same long-lived radioactive waste associated with conventional fission reactors, but it has yet to demonstrate commercial viability at scale. Helion says it aims to produce fusion electricity as early as 2028, although many scientists believe widespread commercial fusion remains much further away.
Google has taken a different approach, focusing heavily on SMRs. In October 2024, it signed an agreement with Kairos Power to develop, build and operate a fleet of small modular reactors. The first is expected to begin operating by 2030, with additional reactors planned through 2035.
Amazon is pursuing both existing nuclear plants and SMRs.
In 2024, Amazon purchased a data center in Pennsylvania from Talen Energy in a $650 million deal and plans to power it using electricity from the nearby Susquehanna nuclear power plant. The facility is expected to reach full capacity in 2032.
About six months later, Amazon invested $500 million in X-Energy, which is developing SMR technology. The company has plans to deploy reactors in Washington state and Virginia, with the first expected to come online early in the next decade.
Meta has also moved aggressively into nuclear power. In 2025, it signed a long-term agreement to purchase the entire output of Constellation Energy's Clinton nuclear plant in Illinois. The 1.1-gigawatt reactor is expected to supply Meta with electricity for 20 years beginning next June.
In January, Meta announced a similar agreement with Vistra covering three reactors in Ohio and Pennsylvania. It has also signed an agreement with TerraPower, whose investors include Bill Gates, to develop as many as eight SMRs with combined capacity of up to 2.8 gigawatts. The first reactors are expected to come online in 2032.
Oracle founder and CEO Larry Ellison has also said the company plans to build a data center powered by three SMRs.
"As crazy as this field is, this is what's happening right now," Ellison said.
Elon Musk's SpaceX has not announced plans to use nuclear power for its data centers, but Musk has expressed support for the technology. "With the latest technologies, you can build a nuclear reactor that can't tolerate a meltdown," he said in October 2024, while calling for significant regulatory reform.
For the AI industry, the attraction is straightforward.
"Nuclear fusion is the way forward," Altman said at the World Economic Forum in Davos in 2024.
Amazon CEO Andy Jassy has been even more direct about the underlying problem, saying last year that energy, rather than chips, could become the industry's "biggest bottleneck."
The nuclear race meets reality
The technology industry's nuclear ambitions are also being encouraged by the Trump administration, which has adopted an explicitly pro-nuclear agenda. Its policies include plans to accelerate reactor development, remove regulatory barriers, encourage the deployment of SMRs and increase U.S. nuclear capacity substantially over the coming decades.
The attraction for data-center operators is clear. Their facilities need extremely high levels of reliability, making a continuous source of low-carbon electricity particularly valuable.
Nuclear power can provide that stability while producing far fewer carbon emissions during operation than fossil-fuel plants. But that does not mean the nuclear solution is simple.
Traditional reactors are extraordinarily complex and expensive to build. The U.S. Nuclear Regulatory Commission maintains strict safety requirements, and licensing a new reactor can take many years and cost hundreds of millions of dollars before construction is even complete.
Under the existing system, developing and building a nuclear facility can take eight to 12 years or longer, far slower than the pace at which the AI industry is adding data-center capacity.
The Trump administration has therefore ordered the NRC to reform its licensing process and dramatically shorten the time required to approve new reactors, with an eventual goal of decisions within roughly 18 months.
SMRs are supposed to address some of these problems. Because they are smaller and designed to be manufactured in standardized modules, proponents argue that they could eventually be produced more efficiently and deployed more quickly than traditional reactors.
But the technology remains largely unproven commercially.
In recent years, companies including TerraPower and NuScale Power have delayed or canceled projects because of soaring construction costs, financing challenges and difficulties securing supplies of enriched uranium, some of which have been exacerbated by the war in Ukraine.
And even if SMRs begin operating on schedule, they are unlikely to provide cheap electricity immediately. First-of-a-kind reactors are expected to be significantly more expensive than established solar and wind projects.
The economics could improve if manufacturers are able to produce large numbers of identical reactors and benefit from economies of scale. But reaching that point could take well into the next decade.
That creates a fundamental tension for the technology industry: AI needs enormous quantities of electricity now, while the nuclear technologies being promoted as the long-term solution may not arrive at scale for another decade or more.
The waste problem hasn't gone away
There is another problem that the AI boom cannot simply make disappear: nuclear waste.
Nuclear reactors produce little direct carbon dioxide during operation, but they generate radioactive waste that must be safely isolated for extremely long periods.
The United States still lacks a permanent repository for high-level nuclear waste. A proposed storage facility at Yucca Mountain in Nevada was authorized by Congress in 1987, but the project was never completed and was effectively abandoned in 2011.
For now, much of the country's nuclear waste remains stored at temporary facilities, often at or near the reactors where it was generated.
That reality continues to fuel opposition from local communities and environmental groups to new nuclear projects and reactor restarts.
For decades, these problems helped make nuclear power politically and economically difficult in the United States.
But the AI boom has changed the equation.
The world's largest technology companies are now among the wealthiest and most powerful corporations ever created. They have enormous incentives to secure reliable electricity, enormous amounts of capital to invest in energy infrastructure and growing political influence in Washington.
That does not make nuclear power cheap, safe or easy. It does, however, mean that obstacles that once appeared almost insurmountable are now being confronted with unprecedented financial resources and political momentum.
