Infrastructure·
US Hyperscale Data Centers Consume More Power Than Most Nations, Study Finds
Harvard research reveals 403 hyperscale facilities used 82 TWh annually — 1.8% of US power — with over half sourced from fossil fuels, emitting 25.7 million tons of CO2.

A new Harvard study published in Nature Sustainability quantifies the electricity appetite of America's hyperscale data center fleet. Researchers identified 403 facilities via satellite imagery and estimated their combined draw at roughly 82 terawatt‑hours between May 2024 and April 2025 — about 1.8% of total U.S. generation. That figure exceeds the annual consumption of 148 individual countries, including Austria and Israel.
Because 54% of that power still comes from gas and coal, the associated carbon footprint reaches 25.7 million metric tons of CO₂, on par with Denmark's yearly emissions and higher than 113 nations. Lead author Gianluca Guidi emphasizes that the environmental impact is not inherent to AI hardware but to the grid mix powering it.
Why it matters for AI infrastructure
For GPU cloud operators and enterprise buyers, the study underscores a strategic lever: siting new clusters in regions with cleaner grids — or pairing builds with dedicated renewables — can slash emissions per training run without changing a single line of code. As demand for AI GPU capacity accelerates, energy procurement becomes a core competitive and sustainability differentiator.
- aigpu
- ai gpu
- ai gpu cloud
- aigpu dubai
- data centers
- energy consumption
- carbon emissions
- hyperscale
By AiGpu Editorial · Editorial rewrite based on public reporting (The Verge AI)
← All articles