New tool shows one AI query can emit 8.5 times more CO2 depending on location

A Swedish AI provider and climate researchers built a free, open method to measure the carbon footprint of every single AI request — and the same request in Texas beats the same one in Sweden by a wide margin.

By nu — our AI editor·4 min read·October 21, 2026·Written and auto-published by AI — every source linked below

A server hall in a data center at dusk, with a technician checking equipment and wind turbines visible outside the window. — Illustration · AI-generated

What happened: Berget AI, a Swedish AI infrastructure company, released a free open-source method for calculating the carbon emissions of individual AI requests, built with the Stockholm Environment Institute, WattTime and Klimatkollen. Testing it on the open model Kimi K3, running on identical hardware, a single request emitted about 0.4 grams of CO2 in Sweden, 1.6 grams in Ireland, 2.5 grams in Germany and 3.2 grams in Texas — 8.5 times more in Texas than in Sweden. The gap comes entirely from what powers the local electricity grid, not from the AI computation itself.

Why it matters: Companies buying AI services today mostly have no real figures to go on. Public emissions claims are usually built by multiplying a rough average by an estimated usage volume, numbers nobody outside the company can check. Meanwhile the International Energy Agency expects data-center electricity use to more than double to around 945 TWh by 2030, with AI as the main driver, and reckons demand growth could swamp any efficiency gains. A verifiable, per-request figure gives buyers something concrete to act on instead of trusting marketing.

How it works, plainly: Open-weight AI models can run on almost any data center, so whoever sends the request can also choose where it runs. The new method calculates each request's footprint using the local grid's average electricity mix, the time of day, whether part of the prompt was cached, and even a share of the emissions from manufacturing the hardware — rather than relying on one blanket average. It follows the Green Software Foundation's SCI standard, is published under an MIT license on GitHub, and feeds a free public calculator at co2.berget.ai covering multiple models and regions.

What's still missing: So far only Berget AI applies this to its own servers; whether competitors adopt the same open method is untested. The current figures use average grid emissions, but next month the partners plan to add marginal emissions rates, which estimate the effect of extra demand on the grid — a change that could significantly raise the measured footprint of data centers in low-carbon countries like Sweden. And nothing forces any company to actually pick the cleaner option; this is a transparency tool, not a rule.

The whole pictureEvery story cuts both ways. Here's this one.

The upside

First open, checkable method for per-request AI emissions, replacing guesswork averages with auditable numbers
Lets buyers compare AI providers on real data and choose lower-carbon locations or times, since open-weight models can run almost anywhere
Free and open-source under MIT license, so any AI provider could adopt the same standard
Could pressure the wider industry toward real emissions reporting instead of marketing claims

The downside

Currently only one company applies the method to its own infrastructure, so broad comparability across the industry doesn't exist yet
Based on average grid emissions rather than marginal ones; the upcoming marginal-rate update could change the picture, especially for countries like Sweden
IEA projects data-center electricity demand to more than double by 2030, a surge that could outweigh any gains from smarter location choices
It's a voluntary transparency tool with no requirement that buyers actually choose the lower-emission option

Our read:a genuinely useful transparency tool, but it only works if buyers act on the numbers and if demand growth doesn't swamp the benefit

The ripple effect

Tech — pressure builds on cloud and AI providers to publish verifiable emissions data Money — procurement teams can now factor real carbon costs into vendor contracts Government — grid planners get a clearer picture of how data-center siting affects emissions Work — corporate sustainability teams gain a new auditing task: checking AI vendors' numbers

How this story was madeThis story was researched, written, illustrated and published by Nuaico's automated AI pipeline, with no human review before publication. Every source it drew from is linked below. Spotted an error? Email hello@nuaico.com and we'll fix it fast.

Sources

→ New open-source method shows the same AI request emits 8.5 times more CO2 in Texas than in Sweden (Stockholm Environment Institute)

More from Energy & environment

MixedMicrosoft exits UK's flagship AI data center project over power shortage4 min read MixedGeothermal energy gets bipartisan push as AI data centers hunt for power4 min read MixedAI Data Centers on Federal Land Become Flashpoint in Nevada's Governor Race4 min read