Hyperscale Data Centers Use as Much Power as States
The 403 largest U.S. data centers used between 68 and 99 terawatt-hours of electricity in a year, approaching what Indiana, Michigan or Tennessee use, and produced about 25.7 million metric tons of carbon dioxide, Harvard researchers report in Nature Sustainability on Monday, November 16. About 54% of their power came from fossil fuels.
The count is the point. Until now the AI buildout's power bill was estimated from the top down. This study counted it building by building, and the total it found is two to three times the last estimate for 2018, before the study period even captures the newest construction.
America's largest data centers together use about as much electricity as an entire state, according to a facility-by-facility study published on Monday, November 16, in Nature Sustainability. Researchers led by the Harvard T.H. Chan School of Public Health and the Harvard Data Science Initiative used satellite imagery to build a list of 403 hyperscale data centers across the United States, the largest class of facility, which supports cloud computing, large language model training and the global deployment of AI tools.
The team estimated each facility's annual electricity consumption between May 2024 and April 2025 under a range of operating scenarios, then used Environmental Protection Agency data to attribute carbon emissions to each site based on the mix of power plants in its regional grid. Depending on how hard the facilities were running, they used between 68 and 99 terawatt-hours of electricity. The upper estimate approaches what Indiana, Michigan or Tennessee use in a year. The demand equaled about 1.8% of total U.S. electricity consumption and about 25.7 million metric tons of carbon dioxide, two to three times higher than an earlier estimate for 2018.
The fuel mix is where the study's argument sits. About 54% of the electricity powering the facilities was generated by fossil fuels, mainly natural gas at 38% and coal at 15%, and about a quarter of the electricity came from grid regions dirtier than average. The geography is concentrated: just four states, Virginia, Oregon, Ohio and Iowa, accounted for more than half of the electricity use studied, and many facilities sit in areas where electricity is generated heavily by coal-fired power plants.
The authors are direct about what the numbers are for. “Data centers are growing incredibly fast, driven largely by AI, and decisions about where to build them and how to power them are being made right now. But there's very little facility-level information about their environmental footprint,” said lead author Gianluca Guidi, a postdoctoral fellow in Harvard's Department of Biostatistics. Corresponding author Francesca Dominici said the goal is to give the public, policymakers, investors and developers a more concrete, up-to-date picture than was available before.
The study carries stated limits. The researchers didn't have enough data to cover every hyperscale facility in the country, and the data didn't let them separate power used for general-purpose computing from power used for AI-specialized operations at each site. They also didn't measure health effects directly. Dominici said that's the next step: the fossil plants serving this demand also release fine particulate matter, nitrogen oxides and sulfur dioxide, which are linked to heart and lung disease, so people living near and downwind of those plants may bear health costs even when they're far from the data center itself.
And the snapshot is already old. The study period ended in April 2025, and the authors say electricity demand and emissions today are likely much higher. “What we're seeing is already striking, but it's only a snapshot of a rapidly expanding industry,” Guidi said. “The data center boom hasn't slowed down since our study period ended in April 2025. New facilities have been built, existing campuses have expanded, and demand for AI computing continues to grow.” The 68-to-99 terawatt-hour range is, by the researchers' own account, a floor that the industry has already built past.
Sources
The conversation.
Loading comments…
A little more perspective.
Reading is free. Verified members can comment and reply for $2 a month.