U.S. President Donald Trump’s longtime ambitions of ushering in an artificial intelligence boom have only been supercharged by the emergence of Chinese start-up DeepSeek’s new AI model, which torpedoed markets last week and wiped hundreds of billions of dollars from AI chipmaker Nvidia’s market cap.
But the DeepSeek disruption has also underscored the deep uncertainty over just how much energy will be necessary to power Trump’s big AI push. The hulking data centers that underpin the technology are notoriously energy-hungry, prompting some predictions of explosive electricity demand in the coming years.
The sudden emergence of DeepSeek’s new model, DeepSeek-R1, which the company says is built more efficiently than its U.S. competitors, reveals just how hazy that demand outlook actually is in the long run—adding yet another complication to the ongoing U.S.-China tech race.
“We’re really at the beginning of this journey with AI,” said Tanya Das, the director of AI and energy technology policy at the Bipartisan Policy Center (BPC). Das compared the current moment to the internet boom of the late 1990s and early 2000s, when fears arose that the internet would crush the energy grid. That, of course, never really happened.
It’s “completely unclear where we’re going to land,” said Das, who served at the Energy Department during the Biden administration. “It’s unclear how much more efficient the chips are going to get and the algorithms and the software are going to get.”
Trump, for his part, is moving full speed ahead. His AI aspirations stretch back to his first presidency, when he unrolled a national AI strategy and established the National AI Initiative Office. Four years later, he has again made AI a centerpiece of his presidential agenda, including by using an emergency declaration to rapidly approve the construction of new AI power stations and vowing to fast-track power plants for AI data centers.
In a sign of the administration’s all-in approach, Trump’s new Environmental Protection Agency administrator, Lee Zeldin, cited AI as one of the agency’s top five priorities. “Those looking to invest in and develop AI should be able to do so in the U.S., while we work to ensure data centers and related facilities can be powered and operated in a clean manner with American-made energy,” Zeldin said in a statement. “Under President Trump’s leadership, I have no doubt that we will become the AI capital of the world.”
To usher in this transformation, Trump has touted a new joint venture by ChatGPT developer OpenAI, SoftBank, and Oracle to dramatically expand the United States’ AI infrastructure over the next four years. Known as the Stargate Project, the plan could see as much as $500 billion invested over that time frame.
OpenAI has said it “will begin deploying $100 billion immediately,” although exact financial details remain hazy—and tech billionaire and close presidential advisor Elon Musk has publicly cast doubt on whether the firms have enough money to make the investments.
Financial questions aside, DeepSeek-R1’s launch has only underscored the importance of this broader AI push for Team Trump. DeepSeek “should be a wake-up call for our industries,” the U.S. president told reporters last week.
DeepSeek “magnifies the importance of the United States being a haven for AI technology,” said Thomas Pyle, who is the president of the American Energy Alliance and headed the first Trump administration’s Energy Department transition team. “Just add this to the list of things that we are competing with China on.”
It’s harder to gauge exactly how much energy will be necessary to power the AI bonanza that Trump has set his sights on. AI is a notoriously power-hungry technology, consuming copious amounts of electricity and water to power and cool overheating data centers.
Researchers estimate that one ChatGPT query consumes nearly 10 times as much energy as a Google search. And a Washington Post analysis, for example, found that a 100-word AI chatbot-generated email uses a half-liter of water and enough electricity to power 14 LED light bulbs for one hour.
Scaled up, the analysis found that if 1 out of 10 Americans generated one such email per week for an entire year, it would take 435,235,476 liters of water—or the total water consumed by all Rhode Island households for one and a half days. It would also consume the equivalent amount of electricity used by all Washington, D.C., households for 20 days.
In recent years, the United States has not experienced an explosive growth in energy demand nationally, according to a new report by the BPC and Koomey Analytics released on Wednesday. But those pressures are felt differently at the state level, with both Georgia and Virginia grappling with surging data center electricity demand.
As the AI boom takes off, national data center energy demand could nearly triple by 2028, according to an Energy Department-backed study by the Lawrence Berkeley National Laboratory. The research, which was published in December 2024, found that data centers’ annual energy consumption could account for between 6.7 and 12 percent of total U.S. electricity use in the next three years.
But the DeepSeek revelations have added a new layer of uncertainty to projections about future U.S. energy demand, experts said.
“Although data centers’ electricity use appears to be growing again, exactly how that growth will play out in coming years is deeply uncertain, both because growth in the use of AI is uncertain and because progress in efficiency is uncertain,” the BPC-Koomey Analytics report said.
“On the efficiency side, this DeepSeek thing is an example of the kind of curveball that can come from new technology innovation,” said Jonathan Koomey, the president of Koomey Analytics and one of the report’s co-authors.
After the Chinese model triggered a frenzied market scramble last week, International Energy Agency (IEA) experts stressed just how little is currently known about AI’s future energy demands.
“This abrupt reaction highlights that the market currently does not yet have adequate tools and information to assess the outlook for AI-driven electricity demand,” IEA analyst Thomas Spencer told the Financial Times.
It’s not just the hazy demand outlook that has complicated the AI equation, either. No matter how quickly Trump and the world’s tech giants are eager to scale up their AI infrastructure, experts said a big question is whether utility companies will be able to match that pace.
“The reality of actually building that scale of electricity infrastructure is that it can’t happen as fast as what the IT guys would love,” said Koomey, who added that the utility industry operates at an “order of magnitude slower” than the tech sector.
Utility companies build power infrastructure expecting to use it for decades or even hundreds of years, said Joshua Rhodes, a research scientist at the University of Texas at Austin.
“The utility industry doesn’t move very fast, and all of a sudden it’s like someone just ran into the conference room shouting, ‘We need power right now,’” Rhodes said. “Not everyone’s just going to snap to attention for the new guy.”
U.S. tech giants remain undeterred. Hungry for more energy, last year Microsoft announced that it had inked a 20-year deal to revive the Three Mile Island nuclear power plant, the site of the worst nuclear accident in U.S. history. If all goes according to plan, the plant will reopen by 2028.
As Trump barrels ahead, experts say the hunt for more power will likely drive up fossil fuel production—with big implications for U.S. greenhouse gas emissions and national climate action.
Diana Furchtgott-Roth, the director of the Heritage Foundation’s Center for Energy, Climate, and Environment, said she expected the AI push to usher in greater production of natural gas, coal, and nuclear power in the United States.
“The AI field is moving very fast,” she said. “The Chinese new DeepSeek AI is not going to be the last innovation that we see.”
