However this is not the story I’ve been hearing this week in New York City, where thousands of government officials, climate activists, celebrities, investors, and executives have descended for the annual meeting-and-panel extravaganza that is Climate Week. For the Biden Administration officials, clean energy executives, and technological visionaries flitting between sponsored events, data center load growth is,as John F. Kennedy might have put it in one of his frequent flights of amateur Chinese linguistics, a danger and opportunity mixed into one.
“This can be a good-news story. The sky doesn’t necessarily need to be falling,” Kelly Sanders, assistant director for energy systems innovation at the White House Office of Science and Technology Policy, said during a panel discussion hosted by the think tank Third Way, referring to load growth from manufacturing and data centers. “This could actually be good for clean energy.”
And very good for anyone who can promise to deliver said clean energy, even if it’s years in the future. During a “fireside chat” at Geothermal House, a day-long summit on geothermal energy sponsored by Project InnerSpace, a geothermal nonprofit, Mike Schroepfer, the former CTO of Meta who is now a climate venture investor, said the demand for power from AI was “the best news ever.” He argued that having companies with big power needs and deep pockets was much better for clean energy development than having a stagnant grid that’s just trying to replace dirty power plants.
Among those in the same rah-rah camp, the general idea is that energy-hungry data centers can help get new clean energy sources like advanced geothermal through the project finance "valley of death" so they can eventually deliver affordable, clean power to the rest of us. “For the first time in history, demand for clean energy outstrips supply,” said Ally Yost, a senior vice president at Commonwealth Fusion Systems, during a panel discussion in New York City. “Those that have access to that clean power will be in a very profitable situation.”
“AI is a gift for fusion,” added Clay Dumas, a partner at Lowercarbon Capital, a Commonwealth investor. He even conceded that the skyrocketing demand was a “gift for fission,” from which fusion advocates are typically at pains to distinguish themselves. “There’s an intense interest and demand for clean electrons,” he said, referencing the recent deal to bring back a shuttered reactor at Three Mile Island, alongside a power purchase agreement with Microsoft.
That investors and executives at fusion companies were talking about meeting projected load growth is a good sign of how heady the financial and technology prospects have gotten for anyone who has a good story to tell (and some capital). Fusion’s claim to be the holy grail of energy has passed over time from aspiration to irony and back again, thanks to billions piled into the industry in the past few years.
This combination of dreaminess and realism prevailed at Commonwealth’s event, where Dumas said that when he first invested in the company, “there was an exciting story of how fusion or a company like CFS could provide 5, 10, 20% of the world’s primary energy and could become the biggest company in the history of capitalism,” Dumas said. (Perhaps not surprisingly, several former SpaceX employees work at Commonwealth.) Now the focus is on getting a power plant developed with technology that the industry insists will be ready to go online on a reasonable timeframe — something more like a decade than the standard 20 or 30 years.
But whether you’re splitting atoms or fusing them, the demand for clean power from data centers is coming in months and years, not decades. OpenAI chief executive Sam Altman reportedlytold the White House he wants to build 5-gigawatt data centers, which would take the equivalent of five large nuclear reactors to power. Even restarting an existing fission plant takes at least three years, while building a new one using existing typically takes around … well no one knows because there are no plans currently to do so.
“People are not going to be patient” if new clean power can’t be developed quickly, Juliann Edwards, the founder and chief development officer of The Nuclear Company, told me this week. “They're going to go build more gas plants.”
Kathleen Barrón, the chief strategy officer at Constellation, the country’s leading nuclear energy providers, said during the panel hosted by Third Way that conversations about new nuclear are “starting to happen,” and that the most important part of that process is coming up with a reasonable cost estimate. “Once you know what it costs, you can figure out what contributions will be,” she said, referring to the nasty problem of how to split up the expense among various stakeholders, including the government. Barrón pointed out that the second reactor at Vogtle was almost a third cheaper than the first — meaning that maybe the nuclear industry has a chance of getting a handle on costs. In the meantime, owners of existing plants will be happy to reopen and expand what they can, picking up generous incentives all along the way.
Edwards told me she’s been speaking with potential offtakers like Amazon and Meta, utilities, independent power producers, and investors in pursuit of having “binding contracts” for new plants by late 2026. But the hyperscalers committed to using clean power will need more than that.
Lucia Tian, a former official at the Department of Energy’s Loan Program Office who now heads of clean energy and decarbonization technologies at Google, estimated that Google’s clean energy needs would be largely served from existing renewable technologies “that we can deploy at scale,” which, paired with storage, would get the company to around 80% of its needs. “But in order to get that last 20%, we need a suite of technologies including nuclear, long-duration energy storage, fossil generation with carbon capture and storage.”
Behind each of these promising technologies is a unique deployment issue. Geothermal might work in the western United States only, for instance, and even then not before the late 2020s. As for nuclear, outside of reopening shuttered plants and uprating existing ones, Tian said, “the reality that everyone recognizes” is that if “I sign a deal today” for a small modular reactor or the existing AP1000 design, “it’s not going to come online before 2030.” This leaves “a strong role for CCS,” she added, referring to using natural gas with carbon capture and storage, an approach strongly encouraged by new Environmental Protection Agency rules for gas plants, but one that is by no means widespread today.
Making progress on a technology that’s been in development for decades and still involves extracting and burning a fossil fuel doesn’t quite meet the futuristic moment the data center and artificial intelligence boom has created in the present.
“Every day someone asks, can’t you foot the billion dollar risk of a nuclear reactor?” Tian said. The future will have to wait a bit longer, but the data centers are coming now.
NOAA/NWS
