The same AI request emits 8.5 times more CO2 in Texas than in Sweden

A new way to count AI's carbon

Berget AI, a Swedish AI provider, released a free method to measure the carbon emissions of every single AI request. It was built with researchers from the Stockholm Environment Institute (SEI) and the non-profits WattTime and Klimatkollen. Berget AI wants other AI companies to report emissions per request, so buyers can compare providers using real numbers.

Key numbers

• A typical request to the open model Kimi K3 emits about 0.4 grams of CO₂ in Sweden.
• The same request emits about 1.6 grams in Ireland, 2.5 grams in Germany and 3.2 grams in Texas.
• That makes the Texas figure 8.5 times the Sweden figure, on identical hardware.
• Open-weight models made up 29% of tokens on Vercel's AI Gateway in June 2026, up from 11% in April.
• The International Energy Agency projects data-centre electricity use to more than double to about 945 TWh by 2030.

Why location matters

Electricity comes from different mixes of power sources in each place. Some grids use more fossil fuels, so each request there causes more CO₂. Berget AI uses the average grid mix where its servers run, and it counts the emissions from making the hardware too. The method also checks each request separately, because results change with the model, the time of day and whether part of the prompt is cached.

Why it matters

Most buyers rarely get emission numbers from their AI providers. Until now, public figures were mostly averages estimated from outside. Berget AI says that with per-request numbers, buyers can pick lower-carbon options. Researchers at SEI say that the location of the grid and the infrastructure strongly affect how sustainable AI can be.

What to watch

• Whether other AI providers start publishing emissions for each request.
• How the partners extend the method to cover water and energy use, which they plan to add.
• Whether rising AI demand cancels out gains from cleaner power, as the IEA warns could happen.