Key facts
- The 403 hyperscale data centers studied used 68 to 99 terawatt-hours of electricity over the year to April 2025, the study said.
- More than half of the power used by the facilities came from fossil fuels, the researchers found.
- Emissions were two to three times a previous peer-reviewed estimate of hyperscale emissions in 2018, the study said.
- Co-author Francesca Dominici said the figure should be treated as a lower bound as more data centers have opened since 2025.
US hyperscale data centers emitted about 25.7 million tonnes of carbon dioxide between May 2024 and April 2025, roughly the annual emissions of Denmark, according to a peer-reviewed study by Harvard University researchers published in Nature Sustainability on November 16.
The study counted 403 hyperscale facilities, the largest data centers used primarily to handle growth in artificial intelligence. It said they consumed 68 to 99 terawatt-hours of electricity over the year, about 1.8% of total US electricity consumption, and that more than half of the power came from fossil fuels. Virginia, Oregon, Ohio and Iowa accounted for over half of the electricity use studied, and most of the facilities were powered by natural gas.
The emissions figure was two to three times a previous peer-reviewed estimate of hyperscale emissions in 2018, the researchers said. Francesca Dominici, a professor of biostatistics at Harvard and a co-author, said the estimate is likely conservative because many more data centers have opened since 2025. "Since 2025 there have been many, many more datacenters so we need to think of this level of emissions as a lower bound," she said.
Gianluca Guidi, a Harvard researcher on AI and public health and the study's lead author, said the aim was to bring transparency and data to the debate over data centers and AI. "We tried and managed to build a methodology that is scientifically accepted to build public trust," he said.
