AI carbon footprint by country
The energy a prompt needs is nearly the same everywhere. What changes is how the electricity is made. A data center in Sweden or France runs mostly on hydro and nuclear power, one in Poland runs mostly on coal.
Result
- Sweden
- 59.6 g CO2
- France
- 76.9 g CO2
- Brazil
- 256 g CO2
- European Union average
- 529 g CO2
- Germany
- 901 g CO2
- United States
- 953 g CO2
- World average
- 1.17 kg CO2
- Japan
- 1.2 kg CO2
- China
- 1.39 kg CO2
- Poland
- 1.64 kg CO2
- India
- 1.76 kg CO2
These are estimates. Real figures depend on the model, the data center and the length of the answer.
How this is calculated
- Ember, Global Electricity Review 2025. Carbon intensity of electricity in 2024 (Poland 2023). France and Sweden come from a study based on Ember data and are approximate. ember-energy.org
- OpenAI (2025). Sam Altman states an average ChatGPT query uses about 0.34 Wh and 0.32 mL of water. blog.samaltman.com
- Google (2025). A median Gemini text prompt uses 0.24 Wh, 0.03 g CO2e and 0.26 mL of water. arxiv.org
- IEA, Electricity 2024. Earlier estimate of 2.9 Wh per ChatGPT request, used here as the high case. iea.org
Questions
Where is AI the cleanest to run?
On grids with a lot of hydro, nuclear, wind or solar power. In this list Sweden and France have the lowest carbon intensity, at about 24 and 31 g of CO2 per kWh.
Do I choose where my prompts run?
Usually not. Most assistants route requests to whichever data center has capacity. Some providers let you pick a region or commit to one in their terms.