Skip to content
10 GPUs · 8 Grid Regions · 2026 Data

AI Carbon Footprint Calculator

Calculate the CO₂ emissions and energy cost of your GPU cluster. Factor in data center PUE, regional grid mix, and utilization hours — then compare every GPU by carbon efficiency.

1.0 (ideal)1.4 (typical)2.5 (legacy)
1h12h24h (always-on)
Selected config
8× H100 SXM5 · 5.6 kW raw · 7.8 kW w/PUE
Power draw
7.8 kW
with PUE overhead
CO₂ / month
2.2t
5645 kWh/month
CO₂ / year
26.5t
68678 kWh/year
Tok/s per kg CO₂
3.7M
tokens per kg CO₂ per day
Annual CO₂ equivalents
🌳
1,263 trees
Trees needed to offset
absorbing CO₂ for one year
🚗
5.8 cars/year
Equivalent car driving
average US passenger vehicle
✈️
43.3 flights
Transatlantic flights
London → New York, economy

Methodology

Energy

Power = GPU TDP × count × PUE. PUE (Power Usage Effectiveness) accounts for cooling, networking, and facility overhead. Typical hyperscalers run 1.1–1.3×; legacy on-prem averages 1.5–2.0×.

Carbon

CO₂ = kWh × regional grid intensity (kg CO₂/kWh). Grid intensity varies 10× between cleanest (Sweden at 0.045) and most carbon-heavy (India at 0.713) grids. Data from 2024 IEA estimates.

Efficiency

Tokens per kg CO₂ = throughput ÷ daily CO₂. A higher score means more AI work per unit of carbon — the key metric for sustainable AI scaling and ESG reporting.

Formula: kWh = (GPU_TDP_W × count × PUE × hours) ÷ 1000  |  CO₂_kg = kWh × grid_intensity  |  Throughput is estimated peak FP16 inference tokens/s. Actual numbers vary by model, batch size, and utilization rate.