RESEARCH · RESEARCH · #3708
Nature Sustainability study: US hyperscale data centers consumed ~82 TWh and emitted 25.7 Mt CO2 (May 2024–Apr 2025)
A Nature Sustainability paper led by Gianluca Guidi (Harvard Chan) used satellite imagery and EPA data to identify 403 US hyperscale data centers and estimate they consumed roughly 82 TWh of electricity between May 2024 and April 2025—about 1.8% of US electricity—and produced an estimated 25.7 million metric tons of CO2, with ~54% of power coming from gas and coal. The authors stress that location and local grid mixes drive large variation in carbon intensity and warn of potential long‑term carbon lock‑in as data center buildouts supporting generative AI expand.
KEY POINTS
- A Nature Sustainability paper led by Gianluca Guidi (Harvard Chan) used satellite imagery and EPA data to identify 403 US hyperscale data centers and estimate they consumed roughly 82 TWh of electricity between May 2024 and April 2025—about 1.8% of US electricity—and produced an estimated 25.7 million metric tons of CO2, with ~54% of power coming from gas and coal.
- The authors stress that location and local grid mixes drive large variation in carbon intensity and warn of potential long‑term carbon lock‑in as data center buildouts supporting generative AI expand.
- The findings quantify a substantial, concentrated source of electricity demand and emissions tied to hyperscale facilities—often supporting AI—and show that where centers are sited and how they’re powered materially affects national emissions trajectories and policy choices.
WHY IT MATTERS
The findings quantify a substantial, concentrated source of electricity demand and emissions tied to hyperscale facilities—often supporting AI—and show that where centers are sited and how they’re powered materially affects national emissions trajectories and policy choices.