Same AI request produces 8.5 times more CO2 in Texas than Sweden
inference
| Source: HN | Original article
An AI inference request generates 8.5 times more CO₂ when processed in Texas than in Sweden, highlighting regional energy impact differences.
Berget AI, a Swedish inference provider, has unveiled an open‑source calculator that quantifies the carbon cost of individual AI queries. The tool, released on 21 October 2026 under an MIT licence, shows that an identical request to the open‑weight model Kimi K3 generates roughly 0.4 g of CO₂ in Sweden but 3.2 g in Texas – a difference of about 8.5 times. The calculation draws on grid‑average emissions data, hardware manufacturing impacts and real‑time regional electricity mixes, and was developed in partnership with the Stockholm Environment Institute, the nonprofit WattTime and the Swedish climate platform Klimatkollen.
The finding underscores how the environmental footprint of AI services is shaped more by where the computation runs than by the model itself. In regions powered by coal‑heavy grids, such as parts of Texas, each inference can emit several grams of CO₂, whereas countries with greener electricity, like Sweden, keep the per‑request impact under a gram. For enterprises that run large volumes of AI workloads, the disparity translates into substantial differences in total emissions and may influence procurement decisions, data‑center placement and routing strategies.
Stakeholders will now watch for how cloud providers and AI platform operators respond. Possible next steps include offering carbon‑aware routing options, integrating the calculator into service‑level agreements, or adjusting pricing to reflect regional energy intensity. Regulators and sustainability auditors may also cite the tool when assessing the carbon claims of AI products, potentially prompting broader industry moves toward greener inference footprints.
Sources
Back to AIPULSEN