The U.S. Needs More Nuclear Power To Fuel the AI Boom
Texas, Virginia, and Pennsylvania are turning to nuclear power to meet data centers' energy demands.
AI is bringing enormous benefits to consumers and businesses. It is also bringing significant strain to the power grid. Some researchers estimate that one ChatGPT query requires the energy equivalent of lighting a lightbulb for 20 minutes and 10 times as much electricity as a single Google search. Goldman Sachs projects that AI will increase data center power demand by 160 percent nationwide through 2030.
The Department of Energy also expects data centers' energy use to balloon. A December 2024 report forecasts that cloud computing will account for as much as 12 percent of the nation's annual energy use by 2028—up from 4.4 percent in 2023. Virginia, California, and Texas will each serve as a "primary hub" for both small- and large-scale cloud data centers, according to the Energy Department.
Texas is the fastest-growing consumer of electricity in the nation, according to the Energy Information Administration. In 2024, the Electric Reliability Council of Texas (ERCOT)—which manages about 90 percent of the state's grid—said electricity demands could nearly double by 2030 as data centers and cryptocurrency grow and as oil operations in the Permian Basin begin to run on electricity instead of diesel. In March, ERCOT said it has received requests for 99 gigawatts (GW) of new connections—enough to power almost 25 million homes—from large power users (including data centers) in the past year. The state will need to add the energy equivalent of 30 nuclear power plants by 2030 to meet demand, reports Bloomberg.
Last Energy is preparing to deliver 30 such reactors—microreactors, that is. In February, the company announced plans to build 30 of its 20-megawatt reactors in Haskell County, Texas, to service data centers across the state. The site is conveniently located 200 miles west of Dallas, where data centers are expected to add 43 GW of demand to the grid through 2029.
The company has filed for a grid connection with ERCOT, which takes about 18–30 months to complete, according to the regulator. Last Energy is also in the process of applying for an Early Site Permit with the Nuclear Regulatory Commission (NRC). Once obtained, the company will have a 10–20 year window to build its reactors.
This will be Last Energy's first operational project in the United States. Despite being an American company, Last Energy has focused on growing its business abroad because of stringent federal regulations. The developer sued the NRC in December 2024, challenging an agency rule requiring all nuclear power–producing entities—including those that do not generate enough electricity to turn on a lightbulb—to obtain an operating license from the commission before turning on.
Texas isn't the only state turning to nuclear power to meet its data center demand. In Virginia, where data centers could double the state's power demands by 2034, Amazon is partnering with Dominion Energy to develop three new nuclear energy projects. Three Mile Island in Middletown, Pennsylvania, is restarting to provide energy to Microsoft's data servers. The power plant was shut down in 2019.
These efforts will only be as cost-effective and efficient as regulations allow. But the renewed interest in clean and reliable nuclear power could allow the U.S. to make advancements in AI with minimal greenhouse gas emissions.
This article originally appeared in print under the headline "AI Boom Turns to Nuclear for Power."
