A group of 403 hyperscale data centers operating in the United States between May 2024 and April 2025 consumed as much electricity in that period as the state of Michigan, according to Harvard research published Monday in Nature Sustainability. The study also concluded that most of that power came from fossil fuels, and that emissions from data centers are now three times what they were before the AI boom.

The facilities examined are hyperscale data centers, the largest class of such buildings and a frequent target of local opposition around the country. Many are used for AI training or inference, though not all of the hyperscalers in the study serve AI workloads. To build their dataset, the Harvard team combined industry research with satellite data to identify the 403 sites, then modeled their energy use under multiple scenarios and calculated the carbon intensity of the regional grids serving each one.

The burden is concentrated geographically. Data centers in Virginia, Oregon, Ohio, and Iowa accounted for more than half of the energy use in the study; Virginia is home to Data Center Alley. Jon Koomey, an energy researcher who was not involved in the work, said hyperscalers were expanding rapidly well before the AI boom. He added that a peer-reviewed figure for annual consumption is valuable because such numbers are sometimes known only under nondisclosure agreements and never seen publicly.

Federal and state governments do not break out data center electricity use from the rest of the power sector, even as projections point to growing strain on the US grid. Individual facilities' consumption is typically treated as proprietary business information, and operators and utilities are not required to disclose specific figures to the public. The study may also already be behind the industry's pace: its average facility is 40 megawatts, small next to campuses now under development that reach hundreds or thousands of megawatts. Gianluca Guidi, a postdoctoral research fellow at Harvard's T. H. Chan School of Public Health and the study's lead author, said some of the massive campuses now proposed are substantially larger than the typical facility in the dataset.

The research does not account for behind-the-meter power, in which developers build generation onsite; Koomey called that a wild card for measuring total data center demand. Plans for such onsite plants have surged at proposed sites, with as many as 189 gigawatts in the pipeline earmarked for data centers, many relying on inefficient natural gas turbines. BloombergNEF estimated last week that building all of the behind-the-meter plants now queued in the US could raise national emissions by 10 percent.

Federal tracking could soon improve. The Energy Information Agency ran three pilot surveys of data center power use across 10 states this year, and EIA spokesperson Lee Tucker said those pilots are complete and the agency is using them to design a possible voluntary survey of the sector.

More technology news from TechManNews.