RESEARCH · RESEARCH · #3833
Harvard study estimates 403 US hyperscale data centers used ~82 TWh (54% fossil), emitting ~25.7 MtCO2
A Harvard team used satellite imagery to identify 403 hyperscale US data centers and estimated their electricity use from May 2024–Apr 2025 at 68–99 TWh (central estimate ~82 TWh), with about 54% of that power from fossil fuels (38% natural gas, 15% coal), implying ~25.7 million metric tons CO2 per year. The paper warns this is a snapshot before rapid AI-driven growth; the IEA separately projects global data center emissions could reach ~350 Mt CO2/yr by 2035 as operators build on-site gas plants, while the Antalya AI Pledge at COP31 remains voluntary and nonbinding.
KEY POINTS
- A Harvard team used satellite imagery to identify 403 hyperscale US data centers and estimated their electricity use from May 2024–Apr 2025 at 68–99 TWh (central estimate ~82 TWh), with about 54% of that power from fossil fuels (38% natural gas, 15% coal), implying ~25.7 million metric tons CO2 per year.
- The paper warns this is a snapshot before rapid AI-driven growth; the IEA separately projects global data center emissions could reach ~350 Mt CO2/yr by 2035 as operators build on-site gas plants, while the Antalya AI Pledge at COP31 remains voluntary and nonbinding.
- The finding quantifies how current hyperscale data center growth — and operators’ use of on‑site fossil generation to speed buildout — can substantially raise CO2 emissions and strain grids as AI demand expands.
WHY IT MATTERS
The finding quantifies how current hyperscale data center growth — and operators’ use of on‑site fossil generation to speed buildout — can substantially raise CO2 emissions and strain grids as AI demand expands.