The same AI request emits 8.5 times more CO2 in Texas than in Sweden
AI infrastructure company Berget AI says the estimated carbon emissions for the same request differ sharply between two data-center locations.
TL;DR
- Berget AI reports an 8.5-fold difference in estimated emissions for the same AI request between Texas and Sweden.
- The company attributes the gap to differences in electricity supply and the emissions intensity of the local grid.
- The comparison is a company analysis; the public release does not by itself establish a universal per-request footprint.
Berget AI’s October 24 release compares estimated emissions for an identical AI request processed in Texas and Sweden. It reports that the Texas case produces 8.5 times as much CO2. [1]
The comparison centers on location rather than a change in the prompt or model. Electricity generation mix and grid carbon intensity can alter the emissions associated with the power used by a data center. [1]
The result makes geography a material variable in the company’s accounting of AI inference. Hacker News linked to the release and drew discussion around the difference. [1] [2]
The release is a vendor analysis, and the available material does not establish that the estimate applies to all models, workloads or facilities in either region. The underlying assumptions and boundaries should be read as part of the company’s calculation. [1]
Why it matters
The comparison highlights that inference emissions depend partly on where computing takes place and how local electricity is generated. It also shows why single global estimates can conceal large location-specific differences.
Editor's note
Berget AI is the source of the 8.5-fold estimate; Hacker News is the discussion venue, not an independent measurement.