A Harvard-led study estimates that 403 US hyperscale data centers used about 82 terawatt-hours of electricity between May 2024 and April 2025, roughly 1.8% of total US consumption. New Scientist compares that with Missouri’s annual electricity use of 78 TWh. Office-park comparisons aren’t going to cut it. Source
The research, published November 16, 2026 in Nature Sustainability, estimates electricity consumption at 68 to 99 TWh across operating scenarios, depending on how hard the facilities were running. The Michigan-sized comparison in some coverage belongs at the upper end of that range, not at the 82-TWh estimate. Source Source Source
Hyperscale facilities are the largest class of data centers, supporting services including cloud computing and AI. Data gaps kept the study from covering every US hyperscale site or separating AI’s electricity use from other computing. Don’t hand the chatbot the bill for the whole building. Source
How the researchers estimated the electricity bill
The team assembled its facility list using satellite imagery and industry research, then modeled electricity consumption under different operating assumptions. To estimate associated carbon dioxide emissions, researchers used Environmental Protection Agency data on the power plants supplying each facility’s regional grid. These are modeled facility-level estimates, not a comprehensive collection of publicly disclosed electricity bills. Source Source
That regional-grid accounting attributed roughly 54% of electricity generation to fossil fuels and associated the studied demand with about 25.7 million metric tons of CO₂. The CO₂ estimate covers the generation of that electricity, not necessarily emissions released at the data centers themselves. Source Source
Virginia, Oregon, Ohio and Iowa accounted for more than half of the electricity use studied, and the researchers found wide regional differences in reliance on carbon-intensive generation. Source
A large footprint, with important blind spots
Data scientist Alex de Vries-Gao of the Free University of Amsterdam told New Scientist that the study relies on hard-to-verify commercial data and assumptions about facility power use and operating capacity. “You can’t really validate the results,” he said. He nevertheless said the findings were consistent with international estimates of data-center electricity use. Source
Johns Hopkins University researcher Yury Dvorkin also criticized the analysis as relatively “coarse,” pointing to omitted network constraints affecting grid emissions. Corresponding author Francesca Dominici acknowledged the study’s assumptions and called for greater transparency from operators. Source
WIRED reports that the study also excludes facilities using on-site power plants, known as behind-the-meter generation. Source
The study period ended in April 2025. Lead author Gianluca Guidi says expansion since then means consumption and emissions could already be substantially larger in 2026. Source
Comments