Policy

U.S. Hyperscale Data Centers Consume State-Level Electricity

New peer-reviewed research quantifies the power draw and carbon footprint of 403 AI-focused facilities across America.

Omega Editorial · November 16, 2026 · 3 min read

Key takeaways

  • 403 U.S. hyperscale data centers consume 68-99 TWh of electricity annually, equivalent to Wisconsin's or Michigan's total usage
  • More than half of data center electricity comes from fossil fuels, producing 25.7 million metric tons of CO2 per year
  • Four states—Virginia, Oregon, Ohio, and Iowa—account for 50% of all hyperscale data center electricity consumption
  • Regional energy mix creates sevenfold differences in carbon emissions for the same electricity consumption
  • Voters rejected data center expansion in 40 of 41 local ballot initiatives during November 2026 midterm elections

More than 400 hyperscale data centers across the United States collectively consume as much electricity as entire states like Wisconsin or Michigan, according to new research published in Nature Sustainability. The facilities, which power large-scale AI training and deployment, also produce approximately 25.7 million metric tons of carbon dioxide annually—roughly equivalent to Denmark's total emissions.

The study, conducted by researchers at Harvard's Data Science Initiative, provides facility-level estimates of electricity consumption and emissions for 403 hyperscale data centers operating between May 2024 and April 2025. Using satellite imagery and EPA data, the team calculated that these facilities use between 68 and 99 terawatt-hours of electricity per year, with more than half of that power coming from fossil fuels, primarily natural gas and coal.

Geographic concentration and grid disparities

Four states—Virginia, Oregon, Ohio, and Iowa—account for half of all electricity consumption by the studied data centers. Virginia alone hosts 142 hyperscale facilities consuming approximately 21.4 TWh annually, cementing its reputation as "data center alley." Iowa's 26 facilities, while fewer in number, show substantially higher average electricity demand per facility at 7.2 TWh combined.

The carbon impact varies dramatically by location due to regional energy mix differences. Oregon, despite hosting more hyperscale data centers than Iowa or Ohio, produces significantly fewer emissions because its grid relies more heavily on clean energy. Wyoming's electricity mix produces approximately seven times more carbon dioxide per unit of electricity than Oregon's, according to lead author Gianluca Guidi.

Why it matters

This research fills a critical gap in understanding AI's environmental footprint at a time when data center capacity is projected to double in the next three years. Without transparent, facility-level data, policymakers and communities cannot make informed decisions about infrastructure development or hold tech companies accountable for sustainability commitments. The study also reveals that location choices have enormous climate consequences—building in regions with cleaner grids could cut emissions by a factor of seven for the same computational workload.

Voter pushback and future growth

Public concern about data center expansion has already manifested at the ballot box. In November 2026 midterm elections, voters approved measures to limit or prevent data center development in 40 out of 41 local initiatives nationwide, citing economic and environmental concerns.

Yet the industry continues to expand rapidly. The number of hyperscale data centers worldwide doubled from just over 500 in 2019 to more than 1,100 in 2024, with more than half located in the United States. Research firm Synergy estimates U.S. data center capacity will double again by 2028, driven primarily by AI demand. BloombergNEF projects that without clean energy alternatives, U.S. data centers will consume more natural gas than most countries by 2035.

The researchers acknowledge their estimates likely undercount actual impacts, as they lacked data on every hyperscale facility and more centers have come online since their study period ended in April 2025. The team plans to continue quantifying environmental effects including air pollution, water consumption, and health impacts on nearby communities.

The findings were first reported by TIME, with senior author Francesca Dominici noting that despite widespread public statements about data center environmental impacts, "there was basically very little, if none, peer-reviewed data" before this study.

  • #data centers
  • #ai infrastructure
  • #carbon emissions
  • #electricity consumption
  • #hyperscale computing
  • #energy policy

This is an original analysis by the Omega editorial team. Source reporting: AI Watch.

Want systems like this working for your business?

Book a Call

More in Policy