Same AI query emits more CO2 in Texas than in Sweden

Swedish AI company Berget AI has worked with researchers from the Stockholm Environment Institute (SEI) and the organisations WattTime and Klimatkollen to develop a new method for calculating how much carbon dioxide different AI queries generate. The aim is to make the climate impact of the AI industry more transparent and to let companies compare different providers.

The method shows how strongly location affects emissions. A typical query to the AI model Kimi K3 produces about 0.4 grams of CO2 in Sweden, 1.6 grams in Ireland, 2.5 grams in Germany and 3.2 grams in Texas. These differences arise even though the hardware is identical, so the gap comes from where the computing is done and not from the equipment used.

Berget AI has published the method as open source, which means other companies can use it freely. A web-based calculator is also available, allowing users to compare emissions between different AI models and geographic regions.

The work matters because it gives businesses a concrete way to account for the emissions of the AI services they buy. By making the figures comparable across providers and regions, it lets a company see how the choice of where a model runs can change the carbon footprint of the same query by a factor of eight, as in the gap between Sweden and Texas.