The False AI Energy Crisis
Donald Trump and AI executives alike have sounded the alarm about a looming AI-driven energy shortage. Both benefit from the concern.
By Matteo Wong
Over the past few weeks, Donald Trump has positioned himself as an unabashed bull on America’s need to dominate AI. Yet the president has also tied this newfound and futuristic priority to a more traditional mission of his: to go big with fossil fuels. A true AI revolution will need “double the energy” that America produces today, Trump said in a recent address to the World Economic Forum, days after declaring a national energy emergency. And he noted a few ways to supply that power: “We have more coal than anybody. We also have more oil and gas than anybody.”
When the executives of AI companies talk about their ambitions, they tend to shy away from the environmental albatross of fossil fuels, pointing instead to renewable and nuclear energy as the power sources of the future for their data centers. But many of those executives, including OpenAI’s Sam Altman and Microsoft’s Satya Nadella, have also expressed concern that America could run out of the energy needed to sustain AI’s rapid development. An electricity shortage for AI chips, Elon Musk predicted last March, would arrive this year.
Both Trump and the oil and gas industry—which donated tens of millions of dollars to his presidential campaign—seem to have recognized an opportunity in the panic. The American Petroleum Institute has repeatedly stressed that natural gas will be crucial in powering the AI revolution. Now the doors are open. The oil giants Chevron and Exxon have both declared plans to build natural-gas-powered facilities connected directly to data centers. Major utilities are planning large fossil-fuel build-outs in part to meet the forecasted electricity demands of data centers. Meta is planning to build a massive data center in Louisiana for which Entergy, a major utility, will construct three new gas-powered turbines. Both the $500 billion Stargate AI-infrastructure venture and Musk’s AI supercomputer reportedly already or will rely on some fossil fuels.
If one takes the dire warnings of an energy apocalypse at face value, there’s a fair logic to drawing from the nation’s existing sources, at least in the near term, to build a more sustainable, AI-powered future. The problem, though, is that the U.S. is not actually in an energy crunch. “It is not a crisis,” Jonathan Koomey, an expert on energy and digital technology who has extensively studied data centers, recently told me. “There is no explosive electricity demand at the national level.” The evidence is ambiguous about a pending, AI-driven energy shortage, offering plenty of reason to believe that America would be fine without a major expansion in oil, coal, or natural-gas production—the latter of which the U.S. is already the world’s biggest exporter of. Rather than necessitating a fossil-fuel build-out, AI seems more to be a convenient excuse for Trump to pursue one. (The White House and its Office for Science and Technology Policy did not respond to requests for comment.)
Certainly, data centers will drive up U.S. energy consumption over the next few years. An analysis conducted by the Lawrence Berkeley National Laboratory (LBNL) and published by the Department of Energy in December found that data centers’ energy demand doubled from 2017 to 2023, ultimately accounting for 4.4 percent of nationwide electricity consumption—a number that could rise to somewhere between 6.7 and 12 percent by 2028. Some parts of the country will be affected more than others. Northern Virginia has the highest concentration of data centers in the world, and the state is facing “the largest growth in power demand since the years following World War II,” Aaron Ruby, a spokesperson for Dominion Energy, Virginia’s largest utility, told me. Georgia Power, similarly, is forecasting significant demand growth, likely driven by data-center development. In the meantime, Microsoft, Google, and Meta are all rapidly building out power-hungry data centers.
But as Koomey, who co-authored the LBNL forecast, argued, that forecasted growth does not seem likely to push the nation’s electricity demands past some precipice. Overall U.S. electricity consumption grew by 2 percent in 2024, according to federal data, and the Energy Information Administration predicted similar growth for the following two years. A good chunk of that growth has nothing to do with AI, but is the result of national efforts to electrify transportation, heating, and various industrial operations—factors that, in their own right, will continue to substantially increase the country’s electricity consumption. Even then, the U.S. produced more energy than it consumed every year from 2019 to 2023, as well as for all but one month for which there is data in 2024. An EIA outlook published last month expects natural-gas-fired electricity use to decline through 2026. John Larsen, who leads research into U.S. energy systems and climate policy at the Rhodium Group, analyzed the EIA’s power-plant data and found that 90 percent of all planned electric-capacity additions through 2028 will be from renewables or storage—and that the remaining additions, from natural gas, will be built at two-thirds the rate they have been over the past decade.
None of this discounts the fact that the AI industry is rapidly expanding. The near-term electricity-demand growth is likely real and “a little surprising,” Eric Masanet, a sustainability researcher at UC Santa Barbara and another co-author of the LBNL forecast, told me. More people are using AI products, tech companies are building more data centers to serve their customers, and more powerful bots may also need more power. Last year, Rene Haas, the CEO of Arm Holdings, which designs semiconductors, attracted much attention for his prediction that data centers around the world may use more electricity than the entire country of India by 2030. Some regional utilities have projected much higher demand growth into the late 2030s than nationwide estimates suggest. And chatbots or not, building enough electricity generation and power lines for transportation, heating, and industry in the coming years will be a challenge.
Still, tremendous uncertainty exists around just how power-hungry the AI industry will be in the long term. State utilities, for instance, are likely exaggerating demand, according to a recent analysis from the Institute for Energy Economics and Financial Analysis. That might be because utilities are overestimating the number of proposed data centers that will actually be built in their territories, according to a new Bipartisan Policy Center report that Koomey co-authored. And AI still could not turn out to be as world-changing and money-making as its makers want everyone to believe. Even if it does, the energy costs are not straightforward. Last month, the success of DeepSeek—an AI model from a Chinese start-up that matched top American models for lower costs—suggested that AI can be developed with lower resource demands, although DeepSeek’s cost and energy efficiency are still being debated. “It’s really not a good idea” to look beyond the next two to three years, Masanet said. “The uncertainties are just so large that, frankly, it’s kind of a futile exercise.”
If AI and data centers drive sustained, explosive electricity demand, natural gas and coal need not be the energy sources of choice. For now, utilities are likely planning to use some fossil fuels to meet short-term demand, because these facilities are more familiar and much quicker to integrate into the grid than renewable sources, Larsen told me. Plus, natural-gas turbines can operate around the clock and be ramped up to meet surges in demand, unlike solar and wind. But clean energy will also meet much of that short-term demand, if for no reasons other than cost and inertia: Solar panels, wind turbines, and batteries are becoming cost-competitive with natural gas and getting cheaper, while a growing number of industries are turning to renewable energy sources. The tech firms leading the AI race are major purchasers of and investors in clean energy, and many of these companies have also made substantial investments in nuclear power.
Using natural gas, coal, or oil to power the way to an AI future will not be the inevitable result of the physics, chemistry, or economics of electricity generation so much as a decision driven by politics and profit. AI proponents and energy companies “have an incentive to argue there’s going to be explosive demand,” Koomey told me. Tech firms benefit from the perception that they are building something so awe-inspiring and expensive that they need every possible source of energy they can get. Any federal blessing for data-center construction, as Trump granted Stargate, is a boon to production. Meanwhile, oil and gas companies want to sell more energy; utilities earn higher profits the more they spend on infrastructure; and the Republican Party, Trump included, has a pretense to satisfy demand to ramp up fossil-fuel production.
Of course, AI needn’t precipitate a national energy shortage to add to a different crisis. Microsoft and Google, despite promising to significantly reduce and offset their carbon footprints, both emit more greenhouse gases across their operations than they did a few years ago. Google’s emissions grew 48 percent from 2019 to 2023, the most recent year for which there is public data, and Microsoft’s are up 29 percent since 2020, an increase driven substantially by data centers. These companies want more power, and the fossil-fuel industry wants to supply it. While AI’s energy needs remain uncertain, the environmental damages of fossil-fuel extraction do not.
