Factor version soma-ai-ef-2026.09-lifecycle-v2, dated . Location-based grid averages, not renewable contracts. Estimates from published benchmarks.
A typical query of 400 tokens to a mid-size AI model (class B, the gpt-4o class) served from the global-average grid emits about 0.029 g CO₂e, and a million tokens emit 0.0721 kg CO₂e, counting electricity, the share of server hardware and amortised training. The same query served from EU Sweden emits 0.0052 g and from Japan 0.035 g, a 6.6 times difference from region alone.
The table below is the full factor set: three model classes across 16 regions, in kg CO₂e per million tokens (all three terms) and litres of water per million tokens. Each row links to a page with the per-query figures, the split between electricity, hardware and training, and the comparison with the other regions or classes.
| Region | CO₂e A | CO₂e B | CO₂e C | Water A | Water B | Water C |
|---|---|---|---|---|---|---|
| EU Belgium | 0.007 | 0.029 | 0.050 | 0.138 | 0.558 | 0.710 |
| EU Finland | 0.005 | 0.020 | 0.039 | 0.138 | 0.558 | 0.710 |
| EU France | 0.004 | 0.016 | 0.033 | 0.138 | 0.558 | 0.710 |
| EU Germany | 0.017 | 0.066 | 0.097 | 0.115 | 0.464 | 0.590 |
| EU Ireland | 0.013 | 0.053 | 0.080 | 0.060 | 0.241 | 0.307 |
| EU Netherlands | 0.013 | 0.050 | 0.077 | 0.140 | 0.565 | 0.719 |
| EU Sweden | 0.004 | 0.013 | 0.030 | 0.244 | 0.985 | 1.253 |
| Global average | 0.018 | 0.072 | 0.104 | 0.144 | 0.582 | 0.740 |
| Japan | 0.022 | 0.087 | 0.123 | 0.087 | 0.350 | 0.446 |
| Singapore | 0.021 | 0.083 | 0.119 | 0.129 | 0.520 | 0.661 |
| UK (London) | 0.011 | 0.045 | 0.071 | 0.090 | 0.362 | 0.461 |
| US Central (Iowa) | 0.019 | 0.074 | 0.108 | 0.113 | 0.458 | 0.583 |
| US East | 0.013 | 0.051 | 0.078 | 0.100 | 0.404 | 0.514 |
| US South | 0.017 | 0.069 | 0.101 | 0.144 | 0.582 | 0.740 |
| US Texas | 0.015 | 0.061 | 0.090 | 0.060 | 0.242 | 0.308 |
| US West (Oregon) | 0.013 | 0.053 | 0.081 | 0.144 | 0.582 | 0.740 |
Electricity-only values (the term most other sources publish) are lower: for class B they run from 0.006 kg in EU Sweden to 0.080 kg in Japan per million tokens. The methodology page shows the electricity term on its own.
Providers do not publish model sizes, so the class comes from the published tier and benchmark behaviour. Misclassifying a model is the largest single source of error, about four to five times, which is why SOMA maintains the mapping. Reasoning models (o1, o3, DeepSeek-R1, extended thinking modes) generate many more tokens per answer; the token count captures that when it comes from billing.
EU Sweden, EU France and EU Finland have the lowest carbon per token for every model class, because their grids run largely on hydro, nuclear and wind. Japan and Singapore are the highest. Water follows a different pattern: cooler grids can use more on-site water per kWh, so the lowest-carbon region is not always the lowest-water region.
Electricity term: GPU energy per token from the ML.ENERGY Leaderboard v3, a facility overhead of 1.20, and grid carbon intensity from EPA eGRID 2023 and Ember 2023. Embodied hardware and amortised training terms are documented on the methodology page. Water combines on-site cooling water (Li et al. 2025) with the water embedded in grid electricity (Reig et al. 2020). Uncertainty on the electricity term is ±40% for class A and ±50% for classes B and C.
Dataset: AI Inference Emission and Resource Factors for Corporate GHG Inventories, Guillermo Llopis, SOMA AI, September 2026, CC BY 4.0, concept DOI 10.5281/zenodo.20443585 (this version 10.5281/zenodo.22767475). Method: Accounting for AI Inference in Corporate GHG Inventories: A Four-Tier Methodology for Scope 3 Category 1 Reporting, arXiv:2606.10660, under peer review.
Partners can fetch any row from the AI factors API.
Read how to account for AI usage in a Scope 3 inventory under CSRD and why published estimates of a ChatGPT query's footprint differ by 150 times.