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

The Specola’s log records a press release published by Berget.ai on October 23, comparing the carbon cost of identical artificial intelligence requests when processed in Texas versus Sweden. According to the release, the same computational request produces 8.5 times more carbon dioxide when run on infrastructure drawing from the Texas grid than when run on infrastructure drawing from the Swedish grid.

The release attributes this disparity to the differing composition of electricity generation feeding each location’s data centers, framing the gap as a function of grid carbon intensity rather than any difference in the computation itself. The item circulated on Hacker News, where it drew discussion among readers interested in the infrastructure costs of large-scale computing.

No further detail on methodology, measurement period, or the specific facilities involved accompanies the available material. The release does not specify which AI model or request type was used as the basis for comparison, nor does it state whether the measurement reflects a single data center pair or an averaged estimate across multiple sites in each region.

The finding is offered as a quantified data point rather than a policy recommendation. It joins a body of reporting treating the geography of computation, the siting of server farms relative to national power grids, as a measurable variable with direct bearing on the emissions profile of digital services, including those built on conversational and reasoning systems.