US lawmakers address 'astounding' AI datacenter power demand projections

Federal lawmakers addressed a range of concerns about the cost and grid reliability impacts of datacenter power consumption during an April 1 hearing before members of the US House Committee on Oversight and Government Reform.

"The power needed for these datacenters is creating a huge problem for our community," said Rep. Suhas Subramanyam (D-Va.), who was elected in 2024 to represent a congressional district encompassing all or parts of five counties in Northern Virginia. The district includes Loudoun County, Va., which has earned the nickname "Datacenter Alley" due to the rapidly expanding presence of artificial intelligence companies such as Alphabet Inc., Meta Platforms Inc., and Amazon.com Inc.

Rep. Eric Burlison (R-Mo.), chairman of the panel's subcommittee on Economic Growth, Energy Policy, and Regulatory Affairs, convened the April 1 hearing to explore the economics of AI datacenters and electricity demand.

Projected power demand from cloud computing and AI activity has been a major new market opportunity for utilities, with S&P Global Market Intelligence 451 Research recently forecasting 59,000 MW of new datacenter power demand by 2029.

During his remarks, Subramanyam's staff displayed estimates that datacenters will contribute to a $276 annual increase in Northern Virginians' power bills within the next five years and consume roughly three times as much electricity by 2050 as the entire state currently uses.

"Many years ago, when these datacenters were approved, they seemed like a great idea at the time. We got to talk about more property taxes and revenue for the counties," Subramanyam said. "But our community is paying the price now, and we're a cautionary tale for the rest of the country."

Subramanyam noted that datacenter-related transmission projects in the towns of Ashburn and Leesburg, and Loudoun, Rappahannock and Fauquier counties are now running into local opposition.

"The reality is that this infrastructure has to be done in concert with ratepayers and with local residents," Tyson Slocum, director of the consumer watchdog group Public Citizen, said during the hearing.

Demand expected to outpace efficiency gains

Expert witnesses peppered lawmakers with attention-grabbing statistics when asked to characterize expected demand growth from datacenters.

Mark Mills, executive director of the National Center for Energy Analytics, said a single datacenter can now use more power than a typical steel mill.

"A single refrigerator-sized computer rack in a datacenter — and there are thousands of racks — a single one uses more electricity than 50 Teslas or any other electric car," Mills said.

Rep. Julie Fedorchak (R-ND), who was previously a state utility regulator in North Dakota, noted that the 13-state PJM Interconnection LLC is projecting systemwide peak demand for its Mid-Atlantic footprint to grow from approximately 150 GW today to 185 GW in five years.

"That is astounding," said Fedorchak, who launched a working group in February to establish a legislative framework for addressing AI power demand growth.

Although data processing chips have achieved material efficiency gains in recent years, growing demand for AI-enabled products and services will continue to eclipse those gains "for quite a long time," Mills said.

Nuclear considerations

Rep. Gary Palmer (R-Ala.) argued that the US is well-positioned to meet datacenter demand with small modular nuclear reactors located at the sites of retired coal- and natural gas-fired power plants.

But Mills said that small modular nuclear reactor technology is likely still at least a decade away from becoming commercially viable. "They don't exist yet," he said.

Meanwhile, Rep. Scott Perry (R-Pa.) pointed to Constellation Energy Corp.'s plan to restart its Three Mile Island Unit 1 nuclear plant in Pennsylvania as part of a long-term deal to supply Microsoft Corp. with carbon-free electricity. The Three Mile Island facility, renamed the Crane Clean Energy Center, is located in Perry's congressional district.

"Why did they pick Three Mile Island?" Perry asked. Neil Chilson, head of AI policy at the Abundance Institute, responded that datacenters "support some of the most services that we all use all the time, so they need to be up and running all the time."