Big Tech races to nuclear energy for AI data centers, but massive hurdles loom ahead

Big Tech Is Betting on Nuclear Energy to Fuel AI. But There Are Some Massive Hurdles

(Image Credit: PCMag Composite/Jeffrey Hazelwood; Just_Super/quantic69/Yuichiro Chino/via Getty Images) Over the past year, nuclear power has reentered the conversation, hailed as a way to advance energy-hungry AI technology without devastating the planet and causing our electrical bills to skyrocket. So tech companies are going nuclear -- announcing plans to add more reactors in an all-out battle to secure as much data center power as possible and one-up each other's AI ambitions. It won't be easy. And it might not work at all. Meta last month signed a new deal to expand three nuclear plants to power its data centers; it's also planning to reopen an abandoned Illinois reactor. Meanwhile, Microsoft plans to reopen Pennsylvania's Three Mile Island -- nearly half a century after an infamous partial meltdown at the plant. Amazon and Google have also invested in new reactor tech. President Trump has issued four executive orders to promote the nuclear industry, and this week exempted new reactors from environmental review in the hopes of standing them up faster, NPR reports. But many challenges remain. For example, the administration and Westinghouse also plan to open 10 new reactors in the US, with construction starting in 2030. But Westinghouse's last reactors in Georgia didn't fare so well: They were seven years late, $18 billion over budget, and bankrupted the company. This time, Westinghouse says it will use Google's AI products to streamline development. Essentially, AI will help in creating reactors to power AI. What could go wrong? To find out how feasible all of this is, I spoke to several experts in nuclear power. They point out major hurdles to growing our nation's nuclear capabilities, foremost among them high costs and long construction timelines. Then there's the potential public panic over a Chernobyl happening in their backyards. Plus, everything rests on new technology known as small modular reactors (SMRs), which is still in development and completely unproven at scale. It could be a long time -- decades, even -- before your conversations with ChatGPT are powered by nuclear energy, and nothing is guaranteed. Here's what Big Tech isn't telling you about the challenges ahead. AI Needs an Unimaginable Amount of Energy To power all of our data center demands with nuclear energy, we'll need significantly more capacity than is currently available or planned. Reactors' steady, consistent energy supply works well with AI models, which the public expects to be able to use to generate wacky images or vibe code at any time of day. But creating a meaningful amount of nuclear power is a tall order. According to a Goldman Sachs analysis, Big Tech companies would need 85 to 90 gigawatts (GW) of new nuclear capacity to power their AI data centers, but just 10% of that will be available by 2030. Meta is currently accepting proposals to build projects that supply 1 to 4 GW, which translates to under 0.05% of Goldman Sachs's estimate. "Nuclear energy is a very realistic and well-suited option for data centers -- the only issue is timing," says Gary Cunningham, director of market research at Tradition Energy, an energy procurement and sustainability solutions advisor. "New nuclear would take years to deploy." Mark Gribbin, chief legislative analyst at the Joint Legislative Audit and Review Commission, a state legislative oversight firm, echoes Cunningham's analysis. "The short answer is that nuclear is a realistic solution in the long term, but it won't solve the current and near-term energy demand crunch," he says. Gribbin spearheaded a report on data center demands in Northern Virginia, a region with more such centers than anywhere in the world. Virginia residents could see their electricity bills rise by $14 to $37 per month by 2040 (independent of inflation), largely due to a spike in data center demand, Gribbin's team finds. The costs of building new grid infrastructure get passed to customers, as they would with any other large-scale project, including wind and solar installations. "Modern data centers are huge, as big as a stadium," Gribbin tells us. Meta CEO Mark Zuckerberg envisions centers that would cover large parts of Manhattan. The power from Meta's Illinois reactor, run by Constellation Energy, "will go directly into the local energy grid managed by the utility, but support our operations in the region," a company spokesperson tells us. Big 'Not in My Backyard' Energy Will these old reactors be safe? None of the Big Tech companies is expressing any concern, at least publicly. But the 2011 disaster at the Fukushima plant in Japan, caused by a tsunami, renewed public safety concerns after decades of dormancy. (Chernobyl was in 1986.) Presumably, new plants will have better alarm detection and auto-shutoff features. Still, it's hard to imagine local communities embracing nuclear facilities in their backyards. Radiation exposure is a key concern in the event of an accident. It can cause a range of health conditions, such as cancer, according to the US Environmental Protection Agency. However, a decade after Fukushima, the UN found no statistically significant correlation between the disaster and cancer rates. Towns across the US are already protesting data centers located too close to their homes. We visited one Pennsylvania community that's losing land through eminent domain to make way for a new crop of them. Another Virginia community has suffered from constant roars and screeches coming from an Amazon data center, the Prince Williams Times reports. Meta's Georgia facility guzzled so much water that local kitchen taps almost went dry, the New York Times reports. "In Virginia, we already have a lot of local-level community opposition to solar and gas facilities," Gribbin says. "I expect you'd see the same local opposition for proposed new nuclear facilities, especially if the facility is located away from one of the state's two existing nuclear plants." The World Nuclear Association says that every industry has accidents, and nuclear power has had relatively few despite their high-profile nature. "The evidence over six decades shows that nuclear power is a safe means of generating electricity," says the pro-nuclear organization. "The risk of accidents in nuclear power plants is low and declining." It All Rests on 'Unproven' Small Modular Reactors Technological developments are also a key part of making nuclear energy a bigger reality. The reactors at Three Mile Island and Meta's Illinois plant are old. They are stereotypical-looking plants with giant, gray smokestacks billowing steam across the landscape. But experts say these reactors are not what would power a nuclear-rich future. Instead, small modular reactors (SMRs) are the new It Girl of nuclear energy. According to Idaho National Laboratory, they're anywhere from one-tenth to one-fourth the size of legacy reactors. They can be manufactured in a factory, making them easier to scale and deploy across grids nationwide. "There are a lot of companies throwing their hat into the 'build new nuke' circus, to attract investors and government funding, but the most likely candidates are not going to be the same tech used in prior builds, which are decades old, and instead will lean towards the small modular reactor designs being worked on," Cunningham says. There's just one glaring problem: SMRs have never been deployed at scale anywhere in the world. The technology is still in a costly development phase, and "there are no guarantees they ever actually get deployed at utility scale because that hasn't been done before," Gribbin says. Allison Macfarlane, a professor who chaired the US Nuclear Regulatory Commission in the 2010s, writes that the "primary obstacle for nuclear energy, particularly SMRs, is cost -- and the fact that they currently do not exist and are therefore unproven." Macfarlane also raises the issue of waste disposal. If we have reactors across the US, where do we put nuclear waste? The US, she writes, "has no long-term plan for nuclear waste disposal, as progress towards a deep geologic repository for disposal of high-level nuclear waste remains at an impasse." Plans for a dumping ground in Nevada's Yucca Mountain have yet to become a reality. A consultant working with the Joint Legislative Audit and Review Commission "thought the earliest SMRs could start being deployed at utility scale would be 2035," Gribbin says, but that "could be moved up if big-pocketed companies continue dumping resources into them." Amazon and Google's nuclear plans are both focused on SMRs, not legacy reactors. Google says its deal is the "world's first corporate agreement to purchase nuclear energy from multiple small modular reactors." It plans to bankroll the manufacturer, Kairos Power, to ensure its first SMR comes online by 2030, with more reactor deployments to follow by 2035. However, Macfarlane isn't optimistic that SMRs will ever happen, recommending investment in other "proven energy sources," such as wind, solar, geothermal, and battery storage. "SMRs are unlikely to be ready to meet significant electricity needs for another 20 years or more, by which time electricity markets will have evolved, with cheaper storage and renewables more widely available," she writes. "If, in the coming decade, nuclear power -- particularly SMRs -- proves economically unfeasible, investments in the sector will be for naught." But before we dismiss the potential for nuclear energy, it's worth noting that Europe has had more success integrating it into its electrical grid. According to the Nuclear Energy Institute, 65% of France's electricity comes from nuclear power. In the US, it's currently 19%, per the US Energy Information Administration. Britain's Rolls-Royce is developing next-generation SMRs and is the furthest ahead on the technology, Cunningham says. If the company is successful, the US could presumably use its products. It remains to be seen if any country can go full nuclear. To power US data centers with the tech would require a large-scale, all-hands-on-deck deployment of (currently non-existent) SMRs in every corner of the grid. That means a flood of funding, the use of legacy reactors in the meantime (hopefully without any accidents), and a lot of luck. In the meantime, the explosive growth of data centers will cause distress for local residents and the planet. One has to wonder whether all of this is worth it just to date an AI chatbot?

Next-gen nuclear sees a tipping point as Meta and hyperscalers start dealmaking with Bill Gates' TerraPower, Sam Altman-backed Oklo and more | Fortune

Chris Levesque spent a career in the nuclear industry as a Navy submarine operator and a commercial nuclear executive before he joined Bill Gates' startup TerraPower a decade ago, only to realize he "didn't know what innovation was." The staid nuclear sector stalled for decades as natural gas and renewable energy came to dominate a power sector that feared nuclear for both safety concerns and its history of dramatic cost overruns. The only major U.S. expansion in nearly 30 years was the Vogtle project in Georgia, which took 15 years and cost over $35 billion -- more the double the planned budget and timeline. That exercise hardly created an appetite for more. "The U.S. (nuclear) safety record has been so good, but it created a culture where you were almost punished if you innovated," TerraPower CEO Levesque said. "We were rewarded for doing everything the same way it was done last time, maybe 1% better. But don't be a cowboy!" When Levesque joined TerraPower from Westinghouse, a stalwart of the nuclear industry, he found a company guided by a different mindset: What does nature allow? What does science allow? Roughly seven decades after the first nuclear power plant came online in the U.S., we may be witnessing a watershed moment for the industry as a new generation of small modular reactors (SMRs), along with surging demand from power-hungry AI data centers, and the Trump administration's expedited regulatory process, converge to set the stage for what Energy Secretary Chris Wright heralds as "the next American nuclear renaissance." In January, Meta partnered with Gates' TerraPower and Sam Altman-backed Oklo to develop about 4 gigawatts of combined SMR projects -- enough to power almost 3 million homes -- for "clean, reliable energy" both for Meta's planned Prometheus AI mega campus in Ohio and beyond. Analysts see Meta as the start of more Big Tech nuclear construction deals -- not just agreements with existing plants or restarts such as the now-Microsoft-backed Three Mile Island. "That was the first shot across the bow," said Dan Ives, head of tech research for Wedbush Securities, of the Meta deals. "I would be shocked if every Big Tech company doesn't make some play on nuclear in 2026, whether a strategic partnership or acquisitions." Ives pointed out there are more data centers under construction than there are active data centers in the U.S. "I believe clean energy around nuclear is going to be the answer," he said. "I think 2030 is the key threshold to hit some sort of scale and begin the next nuclear era in the United States." Smaller SMR reactors can be built in as little as three years instead of the decade required for traditional large reactors. And they can be expanded, one or two modular reactors at a time, to meet increasingly greater energy demand from 'hyperscalers,' the companies that build and operate data centers. "There's major risk if nuclear doesn't happen," Oklo chairman and CEO Jacob DeWitte told Fortune, citing the need for emission-free power and consistent baseload electricity to meet skyrocketing demand. "The hyperscalers, as the ultimate consumers of power are, are looking at the space and seeing that the market is real. They can play a major role in helping make that happen," DeWitte said, speaking in his fast-talking, Silicon Valley startup mode. "We're in a moment where we finally see this confluence of innovation in the industry to actually do things differently -- kind of for the first time since the advent of nuclear power." Thanks to the shale drilling boom, natural gas-fired power generation has dominated the power sector for much of this century, now comprising over 40% of the U.S. grid. But with gas prices on the rise, and orders for combined-cycle gas turbines backlogged, hyperscalers are looking for alternative and, ideally, cleaner solutions for their long-term energy needs. Wind and solar power, which make up more than 15% of the grid by electricity generation, have presented an attractive option for hyperscalers. But federal subsidies are ending and tariffs are further impacting costs. So nuclear power -- under 20% of the grid -- reenters the equation thanks to new technologies, growing bipartisan support, and eased regulatory permitting. And, with U.S. electricity demand expected to surge anywhere from 50% to 80% between 2023 to 2050, depending on projections, the need for more sources of energy is critical. "The electricity industry in general operates on a slower time constant than the tech industry, and the two industries are really crashing into each right now," Levesque told Fortune about the nuclear race to meet AI's demands. He contends his SMRs will compete economically with gas-fired power. TerraPower is currently constructing its first 345-megawatt, nuclear SMR plant in Wyoming -- the Kemmerer Power Station. It's slated for completion in 2030 and to start providing power to the grid in 2031. The company's new deal with Meta calls for two reactors to come online as early as 2032, powering data center facilities at a yet-to-be-determined location. The agreement includes the option for six additional modular reactors supporting Meta operations -- meaning there could be up to eight reactors totaling 2.8 gigawatts. "It's defining our order book," Levesque said of the Meta agreement. "We have other discussions going on too, and we're trying to scale as quickly as we can," he said, noting that the company expects to have about a dozen plants under construction when Wyoming plant comes online in 2031. "Several of those could be these Meta units." Oklo, which was founded in 2013 by husband and wife Jacob and Caroline DeWitte, plans to start construction on its first nuclear reactors this year in Pike County, Ohio -- about 85 miles from Meta's future "Prometheus" data center campus in New Albany, Ohio. The first reactors are targeted to come online as early as 2030, with the "powerhouse" facility incrementally scaling up to 1.2 gigawatts of electricity on 200 acres of land by 2034. In the meantime, Oklo already is building its first test reactor -- dubbed the Aurora Powerhouse -- with the Department of Energy's Idaho National Laboratory as part of the White House's executive order-created Nuclear Reactor Pilot Program. There are 11 such projects in the works at varying degrees of development and Oklo has three of them. No other company has more than one. Aurora is slated to come online in 2027 or 2028. "Obviously, Idaho is the first one, but Ohio is where we're planning a pretty major presence," DeWitte said. "We're going to be building a lot more there. We're eager to position ourselves to really double down and put down significant roots and start building there." It's a major milestone for the DeWittes, who met at the nuclear engineering department at the Massachusetts Institute of Technology. He hailed from the New Mexico nuclear environment while she grew up around oil and gas technology in Oklahoma. They met Sam Altman the same year they founded Oklo, when Altman was still with the startup incubator Y Combinator and had not yet started OpenAI. They became fast friends, especially since Altman was a believer in power demand growth and the need for clean, next-gen nuclear power. Altman became an investor and fundraiser and served as Oklo's chairman from 2015 until April 2025 -- Oklo went public in 2024. Altman still maintains an almost 4% ownership stake, but no longer leads the board -- a move meant to help Oklo sign more deals with hyperscalers who compete with OpenAI. "Hyperscalers are really good partners to help get new power generation built and on the grid sooner, because they're willing to move faster and they're willing to bring resources to bear," DeWitte said. "That helps all of us de-risk project certainty so it gets built, which translates to having power online sooner. That brings more capacity online, which is great, but that then helps us drive our costs down so that we can build more plants." Oklo now has a market cap hovering above $11 billion, up almost 50% in 12 months despite sizable fluctuations. Tried-and-true, old-school nuclear plants typically function with light-water reactors -- using ordinary water both to create pressure and serve as the reactor's coolant. TerraPower and Oklo both utilize differing versions of sodium-cooled reactors instead of water. The sodium transfers heat better, and their low-pressure systems require much less containment. After all, much of the cost of nuclear plants is for the massive amounts of concrete and steel needed for reactor containment. Levesque said the steel, concrete, and labor per megawatt is more than double what TerraPower's sodium system -- dubbed natrium -- requires. "It's still fission. We're still breaking uranium atoms to release heat, and then we make the electricity with the turbine," Levesque said. "Be we're moving to a plant that's cooled with liquid metal -- sodium -- instead of water, which lets us have a low-pressure plant, meaning everything in the plant is lighter -- lighter components, less piping, less structural concrete and steel." The sodium design also takes advantage of air-cooled chimney systems to keep the reactor safe when it's shut down, instead of requiring off-site electric and water systems for emergencies. Russia, China, and India have been more aggressive over the years in pursuing sodium-cooled reactor projects, but the U.S. is currently catching up. The sodium designs are loosely based on the 60-year-old designs of the Argonne National Lab's Experimental Breeder Reactor-II (EBR-II) in Idaho that first showed sodium-cooled fast reactors could work. But, by then, the traditional water reactors were well accepted, and no one commercially was going to risk anything else -- until now. "To put it bluntly, the industry got used to making things really expensive because it could," DeWitte said. TerraPower even incorporated molten-salt energy storage, which essentially operates as a "thermal battery" to store excess power that can be deployed when electricity demand spikes. Levesque argued that eliminates the need for gas-fired peaker power plants commonly used to add extra power during demand surges. TerraPower's dual reactors offer 690 megawatts of baseload power, but Levesque said the storage addition allows them to deploy up to 1 gigawatt of dispatchable electricity on the hottest days or when other power plants suffer outages. Apart from all the construction supplies and labor constraints, another major expense for the plants is the enriched uranium that sources the nuclear fuel, especially when Russia dominates almost half of the global uranium enrichment market. The U.S. is actively working to build up its own uranium supply chains -- both from a mining and processing perspective -- but Oklo also is focused on nuclear fuel recycling to eventually eliminate much of those concerns. Only about 5% of the energy is used by a reactor, meaning the used nuclear fuel has the potential to be recycled. Oklo is working on fuel fabrication and building a $1.7 billion nuclear fuel recycling facility in Oak Ridge, Tennessee to come online as soon as 2030. Of course, the technology still must be perfected. Oklo may use plutonium as a bridge fuel and, in the meantime, even has a partnership with Energy Secretary Wright's previous oil and gas services company, Liberty Energy, to provide temporary, gas-fired power to data centers until Oklo's SMRs scale up. "Recycling is the big game changer in many ways because it enables you to actually extend the resource considerably," DeWitte said. With recycling, "The entire (uranium) reserves in the United States could power the country for over 150 years." The rebirth of the nuclear industry, and the way it's happening, has not been universally cheered. The White House's goal is to dramatically expand nuclear capabilities in the U.S. from about 100 gigawatts today to 400 gigawatts by 2050 -- enough to power almost 300 million homes (keep in mind that there are about 150 million homes in the entire country today). To meet the ambitious goal and accelerate development of next-generation nuclear technologies, Trump's new reactor program is combining with a federal rewriting of the nuclear safety rules -- placing more under the purview of the Department of Energy instead of the Nuclear Regulatory Commission. The DOE contends it is eliminating unnecessary excess regulations without sacrificing safety. But, while there's truth to overly burdensome bureaucracy, the Union of Concerned Scientists (UCS) and other outside observers remain concerned that safety is falling by the wayside to better serve the global AI race. "The Energy Department has not only taken a sledgehammer to the basic principles that underlie effective nuclear regulation, but it has also done so in the shadows, keeping the public in the dark," said Edwin Lyman, UCS director of nuclear power safety, in a statement. "These longstanding principles were developed over the course of many decades and considered lessons learned from painful events such as the Chernobyl and Fukushima disasters." Despite the fears, Oklo, Antares Nuclear, Natura Resources, and other startups in the reactor pilot program are pressing forward, contending their projects are much smaller and safer than the past disasters that unfolded in the former Soviet Union and Japan. The Energy Department just granted Antares preliminary safety approval for its Mark-0 demonstration reactor to come online this summer in Idaho. In February, Natura reached a deal to develop a 100-megawatt reactor project to help power oil and gas and water treatment facilities in West Texas' Permian Basin. Natura also has a DOE reactor project in the works at Abilene Christian University in Texas. Elsewhere, Kairos Power is building a DOE demonstration reactor in Oak Ridge, Tennessee, but Kairos also has a bigger deal to deal to develop 500 megawatts of SMR power to Google by 2035 for Tennessee, Alabama, and other sites. And Amazon backs x-Energy planning to build 5 gigawatts of SMR power by 2039, including about 1 gigawatt in Washington state. But this potential nuclear renaissance isn't just about varying SMR technologies. With the Trump administration's support, traditional nuclear developer Westinghouse is building 10, pre-licensed AP1000 reactors -- the same kind as Vogtle -- by 2030, each with 1.1 gigawatts of power. Even DeWitte acknowledges the need for both large and small reactors. "I'm not a fan of the small versus large debate," he said. "Large plays an important role in certain areas. It faces a really difficult capital allocation challenge. Smaller reactors need fewer dollars, so they're easier to find the capital, and then you build faster because they're smaller. They can iterate more quickly, both on cost and time. That's important because the learning cycles matter, and they compound."