Meta's Nuclear Deal Underscores the Looming Gap for AI's Power Needs

Yahoo

2 days ago

(Bloomberg) -- Meta Platforms Inc.'s flashy deal this week for nuclear power was just the latest sign of the technology industry's need for more energy — and fast. And while the top US reactor companies are eager to sign more contracts, the big question remains: Will new generation come online quickly enough?
ICE Moves to DNA-Test Families Targeted for Deportation with New Contract
The Global Struggle to Build Safer Cars
NYC Residents Want Safer Streets, Cheaper Housing, Survey Says
The Buffalo Architect Fighting for Women in Design
Data-center demand is rapidly disrupting the power grid as tech companies clamor for electricity to run artificial intelligence. That's revealing a fundamental mismatch with the energy industry, which typically requires years or decades to plan and build new power plants. Even with a few recent high-profile nuclear deals, it's becoming clear that the existing fleet of reactors will be unable to meet the surging demands from this first wave of the AI boom.
It's been more than a decade since the US last broke ground on a large-scale nuclear power plant that came online. And while there's talk of restarting a few shuttered plants and upgrades at ones that are already in operation, that's still not going to be enough to generate the amount of power that AI will demand. In fact, it's not clear if any power source will be adequate, according to Adam Stein, director of the Breakthrough Institute's nuclear energy innovation program.
'There's not going to be enough electricity, in general, to make the tech industry happy in the next few years,' Stein said. Building out new nuclear is possible with 'enough pre-planning and a supply chain in place. Unfortunately, at this moment, we don't have that,' he said.
Nuclear is especially well suited for technology companies with ambitious climate goals because it's a clean-energy source that runs around the clock, unlike intermittent wind and solar. But there's a finite number of conventional reactors in the US — 94 to be precise — and no new ones are currently planned. The US atomic industry is also notorious for project delays and running over budget. Meanwhile, dozens of companies are developing new reactor designs that are expected to be faster and cheaper to build, but they are years away from commercial availability.
Power demand in the US is expected to soar as much as 15% in some regions by the end of the decade, and data centers are a key driver. Texas alone has said it will need the equivalent of 30 conventional nuclear reactors to meet the growing demand from AI. That will easily outstrip the capacity of the grid.
'The acceleration of AI is pushing our energy infrastructure to the limit,' said Syed Bahauddin Alam, an assistant professor of nuclear, plasma and radiological engineering at the University of Illinois Urbana-Champaign.
What's happening now is the first wave of AI, according to Alam. Algorithms that were developed in the early 2020s are being deployed now, with the goal of testing and perfecting the models' ability to think and solve problems. That may run through 2030 to 2035, and will require vast amounts of energy. More nuclear may be available, but not until 'the tail end of the first wave,' Alam said.
That hasn't stopped tech companies from seeking partnerships to secure nuclear power.
Constellation Energy Corp., the biggest US nuclear operator, this week agreed to sell power from an Illinois nuclear plant to Meta in a 20-year contract that's likely to start in mid-2027. That came after the company in September said it would spend $1.6 billion to restart the shuttered Three Mile Island reactor in Pennsylvania to deliver power to Microsoft Corp.
Constellation said it can do more such deals. It has several other reactors receiving subsidies that are set to expire in the next several years, and it is already in talks with multiple potential customers that are eager to follow Meta's example, said Chief Financial Officer Daniel Eggers. The company is also planning upgrades at several existing plants that could add the equivalent of an entire new reactor.
'We have a lot of capacity that would be available,' Eggers said. 'If a customer wanted to procure that output, we could do that.'
But it's not clear that will be enough, at least for the next several years, according to Chris Gadomski, an analyst with BloombergNEF. He warned at an event in April that slow project deployment and a lack of construction today signal that nuclear is likely to miss the first wave of power demand from hyperscalers.
Moving forward, one of the key factors for the industry will be the availability of federal funding.
Nuclear reactors are incredibly expensive to build, and even next-generation cheaper models would benefit from government support.
This week, details of the Trump administration's fiscal year 2026 budget request showed that it plans to use a $400 billion green bank to finance billions of dollars in loans for projects including nuclear reactors. The move came after President Donald Trump in May signed orders meant to accelerate the construction of nuclear power plants.
Still, even if a major new nuclear plant went into construction immediately, it would likely be years before it would be able to send electricity to data centers.
'Nothing is going to help them for this first wave of AI demand,' said Stein of the Breakthrough Institute.
Cavs Owner Dan Gilbert Wants to Donate His Billions—and Walk Again
YouTube Is Swallowing TV Whole, and It's Coming for the Sitcom
Millions of Americans Are Obsessed With This Japanese Barbecue Sauce
Is Elon Musk's Political Capital Spent?
Trump Considers Deporting Migrants to Rwanda After the UK Decides Not To
©2025 Bloomberg L.P.