Story
Nov 17, 2026
AI’s Hidden Power Bill Is Already the Size of a State
Researchers say hyperscale data centers’ enormous electricity demand is now measurable, while energy experts stress that the resulting emissions depend heavily on where facilities are built and how they are powered. Their findings sharpen calls for cleaner energy and more transparency as communities confront rapid expansion.
The electricity demand behind America’s AI boom is no longer just a projection. Researchers say a new facility-level study offers rare evidence of the scale—and shows that the climate cost depends greatly on the local power grid.1
The team identified 403 U.S. hyperscale data centers using satellite imagery and estimated their electricity use from May 2024 to April 2025. The study put consumption at 68 to 99 terawatt-hours, with a central estimate of about 82 TWh—roughly 1.8% of all U.S. electricity use. The facilities’ estimated emissions reached 25.7 million metric tons of carbon dioxide, comparable to Denmark’s annual total. More than half their electricity likely came from fossil fuels, chiefly gas and coal.2
The research addresses a major data gap, said senior author Francesca Dominici: “there was basically very little, if none, peer-reviewed data” on data centers’ environmental impacts. Its estimates may still be low: the study did not cover every hyperscale facility, and more have opened since the analysis ended.1
But the researchers’ central point is not that every data center carries the same footprint. Lead author Gianluca Guidi said Wyoming’s electricity mix produces about seven times more carbon per unit of power than Oregon’s. “Geography matters,” he said.1 A separate account of the study puts the difference at roughly sevenfold and reports that independent energy experts see long-term risks in building power infrastructure for rising demand. Cornell’s Fengqi You warned that decisions made now could lock in emissions for decades.2
That makes the question political as well as technical. MIT’s John Sterman urged policymakers to require operators to bring new clean energy online, while former EPA administrator Gina McCarthy called for reliable independent data on effects on air, health and climate. Communities are already pushing back: the first account reported that voters backed limits or bans in 40 of 41 local ballot initiatives this November.21
For the researchers, the takeaway is that AI’s infrastructure is physical—and its consequences can be shaped by where facilities go and what powers them.