Hyperscale Data Centers Consumed Up to 2.2 Percent of US Electricity, Study Finds
A new tally of more than 400 large US data centers estimates they drew 68 to 99 terawatt-hours of power in a year, roughly matching the electricity use of an entire midwestern state.
By Marco Harris ·

The technology industry's sweeping bet on large language models has set off a rapid race to convert investment capital into hyperscale data centers. That buildout has moved so quickly that no one has been able to plan how to supply the needed electricity cleanly or with minimal impact on ratepayers. Even basic information about what is being built has been hard to come by.
A new study offers a clearer picture, tallying energy use and estimating emissions for hyperscale facilities operating in the United States between May 2024 and April 2025. The researchers could not separate AI computing from more traditional data services, but their work establishes a firm reference point for tracking the sector's footprint in the years ahead.
To find the facilities, the team began with a list of 675 sites drawn from a commercial database compiled from news reports, company statements, and satellite imagery. They removed any location that lacked a reported electrical capacity, as well as any they could not verify was a suitably large building outfitted with the cooling and electrical connection equipment such operations require.
That screening cut the list to 403 data centers, with an average electrical capacity of about 40 megawatts each. Because 272 facilities believed to be large data centers were excluded, the researchers describe their figures as a very conservative estimate. At the same time, they acknowledge that some of the sites included in the final count may not actually be AI data centers, which matters for anyone focused on that category specifically.
Uncertainty also surrounds how hard these facilities actually run. Drawing on earlier research, the study applies a range of utilization rates: conventional data centers tend to operate at an average of about 48 percent of capacity, while AI training workloads have been running considerably higher, at around 70 percent.
Combining those factors, the researchers estimate that the 403 data centers consumed between 68 and 99 terawatt-hours of electricity over the year studied.
That total works out to between 1.5 and 2.2 percent of all electricity used in the United States, an amount comparable to the entire consumption of Wisconsin or Indiana.
The findings land amid intensifying scrutiny of the AI boom's demands on the power grid. While the study cannot fully isolate AI's share of the load, it gives policymakers, utilities, and researchers a baseline for measuring how the industry's data center expansion affects national electricity use going forward.
More News · National news · Report a business · Got a Story?






