America’s 403 largest data centers were estimated to consume between 68 and 99 terawatt-hours of electricity during the May 2024 through April 2025 study period. The figures come from a facility-level analysis published on the preprint server arxiv.org. The central estimate of roughly 82 terawatt-hours represents approximately 1.8% of total U.S. electricity consumption for the period.
Those figures are modeled estimates across load scenarios, not direct meter readings across every facility. The 31-terawatt-hour gap between the low and high ends reflects modeled uncertainty about facility load, and readers should treat the central estimate accordingly.
Fossil Fuels Still Dominate the Mix
The central scenario attributes approximately 54% of electricity consumed by these facilities to fossil-fuel sources, undercutting assumptions that hyperscale growth runs primarily on new renewables.
About 54% of the electricity attributed to these facilities came from fossil-fuel sources under the central scenario, with natural gas and coal as major components. That share challenges any assumption that hyperscale data-center growth is being supplied primarily by new renewable generation.
The central estimate places attributable carbon dioxide emissions at approximately 45 million metric tons. Across the modeled scenarios, that range spans roughly 37 million to 54 million metric tons.
Geography Turns a National Percentage Into a Local Problem
Virginia, Oregon, Ohio, and Iowa carry the heaviest modeled demand, and concentration at the state level can dwarf what a modest national share suggests.
Northern Virginia’s “Data Center Alley” shows how a modest national share can still mean enormous pressure on individual utilities, transmission infrastructure, and nearby communities. Residents near the fossil-fuel power plants supplying these facilities may experience elevated concentrations of fine particulate matter, nitrogen oxides, and sulfur dioxide, even when they live far from any data center.
How Researchers Built the Estimate
Researchers linked each facility to its supplying power plants and estimated emissions from those plants’ fuel mix, using a previously validated but modeled allocation method.
The study linked each of the 403 facilities to its supplying power plants, then estimated emissions based on those plants’ fuel mix. A key assumption holds that supplying plants contributed equal shares to each facility’s energy mix. Researchers describe this as a previously validated approach; it remains a modeling assumption, not a contract-level observation of actual electricity flows.
The analysis also did not consistently distinguish AI-specialized operations from general-purpose cloud services. The figures should not be read as a precise measure of AI’s energy footprint alone.
Two Studies, Two Different Scopes
A broader earlier assessment covered more than five times as many facilities and is not directly comparable to the newer hyperscale-only estimate.
A separate earlier assessment covered 2,132 U.S. data centers across a different time period and estimated they accounted for more than 4% of U.S. electricity consumption, with more than 105 million metric tons of carbon dioxide equivalent emissions. The two results are not directly comparable: they cover different facility populations, definitions, and time windows.
The newer hyperscale-only study covers a narrower, more specifically defined set of facilities and produces a more granular baseline as a result.
What the Data Cannot Yet Answer
The study identifies a consumption and emissions baseline, but the central policy and accountability questions remain unresolved.
Researchers noted that future work will examine differences between grid-connected facilities and those relying on on-site generation. That distinction matters because on-site gas turbines can shift pollution from regional power plants closer to data-center communities.
The study does not resolve who pays for new generation and transmission capacity. Nor does it determine whether renewable-energy contracts reflect actual hourly grid supply, or whether communities near data centers and their supplying plants receive enforceable pollution protections. It provides a baseline. The accountability questions remain open.






























