Harvard Study Estimates US Hyperscale Data Centers Used 68-99 TWh, About 54% From Fossil Fuels
A Harvard-led study estimates US hyperscale data centers used 68 to 99 terawatt-hours of electricity from May 2024 to April 2025, with about 54% of that power from fossil fuels and emissions near 25.7 million metric tons.
Newsnibs Staff
- 2026-11-19
- Updated 2:42 AM UTC
A Harvard-led study estimates that 403 US hyperscale data centers used between 68 and 99 terawatt-hours of electricity from May 2024 to April 2025, with a middle estimate of about 82 TWh. That is about 1.8% of all electricity used in the United States. The facilities' carbon dioxide emissions come to about 25.7 million metric tons, within a range of 21 to 31 million.
The emissions figure is two to three times higher than estimates for 2018, one of which put that year at 10.5 million metric tons. About 54% of the electricity came from fossil fuels, mainly natural gas and coal. Breakdowns of that share differ. One gives 38% natural gas and 15% coal, and another counts 21% nuclear within it. One account puts renewables at about 25% of the energy.
The researchers are at the Harvard T.H. Chan School of Public Health and the Harvard Data Science Initiative. Gianluca Guidi, a postdoctoral fellow in biostatistics, is the lead author, and Francesca Dominici is the corresponding author. The study appeared in Nature Sustainability under the title "Assessing the carbon emissions and energy consumption of US hyperscale data centres." The team used satellite imagery to identify the facilities, and one account says it combined that imagery with operator disclosures. EPA data was used to attribute emissions according to each regional grid's fuel mix. The Department of Energy publishes only aggregated national figures, so public facility-level data is scarce.
Virginia, Oregon, Ohio and Iowa together account for more than half of the studied electricity use. By one account, Virginia alone has 142 of the facilities and roughly a quarter of the load. The average facility has about 40 megawatts of installed capacity and a footprint of roughly 2,000 square meters. Comparisons for the total vary. The range has been likened to the annual use of states such as Indiana, Michigan and Tennessee, each around 100 TWh, and one comparison sets it against Chile's 83.2 TWh.
The study period ended in April 2025, and the authors say the picture has changed since. "Data center boom has not slowed since April 2025," Guidi said, adding that "decisions about where to build them and how to power them are being made right now." One account says the number of large facilities has more than doubled since the data ends, with 2,000 more planned or under development. Dominici said the findings bear on "critical environmental and public health decisions." Guidi also asked: "Are we confusing what's possible or what's profitable with what's actually good for society?"
Guidi has said actual current figures are likely higher. Ana Valdivia, a professor at the University of Oxford, argues the study underestimates real use because it excludes off-grid generation, such as xAI's infrastructure in Tennessee. One account makes a similar point about unreported on-site generators. The study acknowledges that it cannot separate general-purpose computing loads from AI-specialized loads at individual sites.
Individual projects are large. Meta's planned AI facility in Louisiana could demand up to 7 gigawatts, according to one account. That is more than 175 times the median current facility, and the same account says Meta is reportedly building 10 new gas power plants. One analysis found that 99 proposed gas plants meant to power data centers directly could emit 318 million metric tons of CO2 per year, which would raise US power-sector emissions by about 20%.
At the UN COP31 climate talks in Turkey, UN climate chief Simon Stiell launched a voluntary pledge for companies to power AI with cleaner energy. "Claims about future emissions savings cannot excuse rising pollution now," he said. The pledge has not yet drawn signatures from the US tech companies driving the AI buildout.
Photo by imgix on Unsplash