Hyperscale data centers in the United States consumed approximately 82 terawatt-hours of electricity between May 2024 and April 2025, generating 25.7 million metric tons of CO2 emissions. The 403 facilities identified by researchers account for 1.8 percent of all U.S. electricity use—more than the annual consumption of 148 individual nations.
Fossil fuels powered 54 percent of that demand, a structural vulnerability as operators race to expand AI infrastructure while facing grid capacity limits and corporate net-zero commitments.
The carbon footprint—equivalent to Denmark's annual emissions—exceeds the total output of roughly 113 countries. Gianluca Guidi, a postdoctoral fellow at Harvard T.H. Chan School of Public Health and lead author of the *Nature Sustainability* pa noted the trajectory is not predetermined. "We have choices about where we build and how we power this infrastructure," Guidi said.
The research team used satellite imagery to identify the 403 facilities and estimated power consumption from electrical capacity data. Carbon emissions were calculated using EPA power-grid data specific to each center's location during the study period.
MIT System Dynamics director John Sterman verified the study's scope and comparisons. The findings underscore a mounting economic constraint: as hyperscalers compete for limited renewable energy and face grid interconnection bottlenecks, the cost of AI expansion increasingly hinges not on chip availability but on securing decarbonized power and managing the capital intensity of grid upgrades.
