On November 16, The Verge reported on a new paper in the journal Nature Sustainability about how much electricity huge U.S. data centers use, and how much carbon pollution that power likely produces. The study focuses on hyperscale facilities, the very large buildings that companies are racing to build, in large part to support generative AI.

Gianluca Guidi, a postdoctoral fellow in the Department of Biostatistics at the Harvard T.H. Chan School of Public Health and the study’s lead author, and his colleagues used satellite imagery to identify 403 hyperscale data centers in the United States. They estimated each site’s power use from its electrical capacity, then used Environmental Protection Agency data to approximate the carbon pollution that would come from the local grid between May 2024 and April 2025.

Altogether, The Verge reported, the 403 facilities likely consumed around 82 terawatt-hours of electricity that year, about 1.8 percent of all U.S. electricity consumption. John Sterman, director of the MIT System Dynamics group and the MIT Sloan Sustainability Initiative, who was not involved in the study, told The Verge that this was more electricity than about 148 nations — a majority of the world’s countries, including Austria, Israel, and Venezuela, but not the largest economies such as the United States and China — each use in a year. A terawatt-hour is a huge amount of energy: one trillion watt-hours.

The same research estimated that electricity demand from these data centers produced 25.7 million metric tons of carbon dioxide, which the paper compared to Denmark’s annual carbon footprint. Sterman said that exceeds the total carbon emissions of about 113 nations. About 54 percent of the electricity those hyperscale sites consumed likely came from gas and coal, according to the study as reported by The Verge. The mix matters: generating power in coal-heavy Wyoming led to roughly seven times more carbon emissions per kilowatt-hour than in hydropower-rich Oregon, the story said.

“We tend to discuss AI’s environmental footprint as though it were some fixed property of the technology, when in reality the same computing infrastructure can have dramatically different consequences depending on the electricity system supporting it,” Guidi told The Verge by email. Fengqi You, a professor in energy systems engineering at Cornell who also was not part of the study, pointed to “long-term carbon lock-in,” because power plants built to serve data centers may run for decades. The Verge noted that the study does not count local air pollution from new gas equipment, and that hyperscale sites are only a fraction of all U.S. data centers. Researchers also argued that operators often will not disclose energy use, which is why the team had to estimate it.

Sterman disclosed that he used help from Anthropic’s Claude to compare the paper’s findings with country-level electricity and emissions numbers. Gina McCarthy, a former White House national climate advisor and EPA administrator, said in a statement shared by the authors, “Communities and policymakers need reliable, independent data on what these facilities mean for our air, health, and the climate. This study is an important step.” The figures are estimates for a specific set of giant facilities in one year, not a scoreboard for every computer or every country.