Blog

Load Growth Irrational Exuberance Crashes into DeepSeek

Electricity , Emerging Tech

The development of DeepSeek AI has demolished the justification for urgent, massive expansion of U.S. power generation fueled by natural gas.

Two years ago, the U.S. Energy Information Administration projected “stable” growth in electric power demand through 2050, driven by economic growth, reshoring of industry, and economy-wide electrification. Total energy consumption would increase by as much as 15%, with generation capacity more than doubling and renewables increasingly displacing fossil fuels. 

But since then, a “data center power frenzy” has taken over electricity forecasting, grid planning, and public policy discussion.

For example, a 2024 report from the Lawrence Berkeley National Laboratory found that U.S. data centers consumed roughly 4.4% of the nation’s electricity in 2023 but could double or perhaps triple their use, to as much as 12% of all power by 2028. The Electric Power Research Institute projected that data centers could consume up to 9% of U.S. electricity generation by 2030—more than doubling their current consumption. Grid planners nationally bought into the frenzy, nearly doubling the five-year load growth forecast from 2.6% to 4.7%. The situation was likened to adding another California’s worth of electricity demand in this decade.

The president of the Electric Power Supply Association, however, warned of irrational exuberance in overestimating load growth, citing PJM forecasts. A report from the Ohio River Valley Institute agreed, citing analysis suggesting that load growth from data centers will likely be only a third of the amount forecast by PJM.

But the response to the forecasts has been predictable. Some policymakers, and grid managers like PJM, are pushing gas. Despite renewables and storage comprising over 97% of proposed new generation projects in the interconnection queue, and those projects generally having shorter lead times compared to projects like gas plants and comparable or better reliability, PJM has tried to stampede gas-fired generation. They have even proposed to allow gas plants to jump to the head of the line to meet data center demand-driven load growth.

U.S. tech companies fed the frenzy by taking the approach that the way to scale up AI is to throw money at it—using ever larger datasets, bigger models, and more computational power; amassing more chips, more massive data centers, and using exponentially more energy.

Major AI players even began taking energy matters into their own hands, signing deals to bring retired nuclear power plants back on-line, tapping into existing nukes, and investing in new nuclear technology. Power companies and oil and gas companies, too, are gearing up to build new behind-the-meter gas plants for data centers.

The irrational exuberance reached a crescendo on the day after the inauguration, when President Trump announced the Stargate Project, an investment of up to $500 billion (largely repackaging and branding existing plans) in building data centers across the U.S. (some of which may employ some solar and storage).

But less than a week later, two years of frenzied forecasting, planning, and general exuberance crashed into DeepSeek.

A Chinese AI company’s breakthrough that one U.S. industry player called “a Sputnik moment,” DeepSeek’s open-source AI model outperforms current industry standards using a small fraction—perhaps as little as 2%—of the chips, hardware—and energy.

DeepSeek has suddenly demolished the justification for urgent, massive expansion of U.S. power generation using natural gas.

What now? To be sure, in the long run, making AI cheaper will likely spur wider adoption and use, and hence increase power requirements. But in the short to medium run, it should be harder to justify a stampede to gas.

Grid planners and policymakers should turn down the panic meter, take into account chip improvements, efficiency gains in AI computing, and the potential of clean energy and storage before green-lighting more gas plants. They should refocus on clearing clean generation backlogs and deploying grid enhancing technologies. Those moves will allow already strong clean energy growth to accelerate, growing the economy, creating jobs, and reducing energy costs.

John Quigley

Senior Fellow, Kleinman Center

John Quigley is a senior fellow at the Kleinman Center and previously served on the Center’s Advisory Board. He served as Secretary of the PA Department of Environmental Protection and of the PA Department of Conservation and Natural Resources.