AI carbon footprint in US South - Southeast: emissions per token and per query
Factor version soma-ai-ef-2026.10-dataset-v3.0.0. Last updated . Location-based grid averages, not renewable contracts. Estimates from published benchmarks.
Part of the SOMA AI Emission Factors cluster.
How much CO₂ does an AI query emit when served from US South - Southeast?
A 1210-token query to a class B model (gpt-4o) served from US South - Southeast emits about 0.048 g CO₂e, and a million tokens emit 0.0351 kg CO₂e.
Rank 13 of 16 for class B, lowest carbon first (full table). 4.9x the level of EU North - Sweden, Asia Pacific - Singapore is 1.1x higher.
Which grid supplies the electricity in US South - Southeast?
US South - Southeast maps to SERC South, a US eGRID subregion, at 431 g CO₂e per kWh for 2023. That is 1.8 times the EU-27 average (236 g) and 0.9 times the global average (483.2 g). The source is eGRID2023 SRSO 846.0 lb/MWh CO2e total output x 1.1232 (Ember lifecycle basis).
No provider region in the dataset's cloud mapping table resolves to the SERC South grid. The page is reached by its region name and not by a provider region id.
Which grids have an intensity close to SERC South?
| Difference g | Grid | Code | g CO₂e per kWh | Ratio |
|---|---|---|---|---|
| -3.7 | South Korea | KOR | 427.3 | 0.991 |
| 5.6 | French Polynesia | PYF | 436.6 | 1.013 |
| 5.6 | Togo | TGO | 436.6 | 1.013 |
| -7.5 | Palestine, State of | PSE | 423.5 | 0.983 |
| -7.6 | Madagascar | MDG | 423.4 | 0.982 |
| 10.1 | NPCC NYC/Westchester | NYCW | 441.1 | 1.023 |
| 11.1 | SPP North | SPNO | 442.1 | 1.026 |
| 11.6 | Czechia | CZE | 442.6 | 1.027 |
| 11.9 | Russian Federation (the) | RUS | 442.9 | 1.028 |
| -13.7 | Sri Lanka | LKA | 417.3 | 0.968 |
Emission factors for US South - Southeast by model class
| Model class | Electricity kg | Hardware kg | Construction kg | Training kg | Total kg CO₂e | g per query | Water L | mL per query |
|---|---|---|---|---|---|---|---|---|
| Class A (small) | 0.00938 | 0.00027 | 0.00014 | 0.00060 | 0.010 | 0.014 | 0.052 | 0.072 |
| Class B (mid-size) | 0.03073 | 0.00101 | 0.00046 | 0.00293 | 0.035 | 0.048 | 0.170 | 0.24 |
| Class C (frontier) | 0.11442 | 0.00388 | 0.00169 | 0.01284 | 0.133 | 0.18 | 0.635 | 0.88 |
| Model class | Central | Class range low | Class range high | High case | B200 | Development |
|---|---|---|---|---|---|---|
| Class A | 0.01039 | 0.006437 | 0.01504 | 0.01663 | 0.009116 | 0.01279 |
| Class B | 0.03513 | 0.02273 | 0.04257 | 0.05558 | 0.0247 | 0.0419 |
| Class C | 0.1328 | 0.05292 | 0.1982 | 0.2091 | 0.1366 | 0.156 |
How much does siting matter for a model served from US South - Southeast?
Electricity is 87.5% of the class B total here (class A 90.3%, class C 86.1%), with hardware at 2.9%, construction 1.3% and training 8.4%. Electricity carries most of the 0.0351 kg, so the grid is the biggest lever.
The example company (2,323 million class B tokens a year) emits 82 kg CO₂e from US South - Southeast. Moving to US South Central - Texas would save 9.2 kg (11%) or EU Central - Germany would save 11 kg (14%).
| Region | Grid g per kWh | Class A | Class B | Class C |
|---|---|---|---|---|
| US South Central - Texas | 375.3 | 2.82 | 9.23 | 34.4 |
| EU Central - Germany | 363.6 | 3.41 | 11.2 | 41.6 |
How much water does AI usage consume in US South - Southeast?
A class B query of 1210 tokens from US South - Southeast consumes about 0.24 mL of water and a million tokens 0.17 litres, or 0.13 litres without the water attributed to hydropower.
| Input | Value |
|---|---|
| On-site WUE, L per kWh | 0.123 |
| WUE denominator | total site electricity |
| WUE basis as published | withdrawal |
| Withdrawal to consumption | 0.6667 |
| EWIF, L per kWh | 2.309 |
| EWIF without hydro, L per kWh | 1.743 |
| AWARE factor | 1.78 |
| Confidence | derived-site |
| Source | Meta; Stanton Springs (GA) (site); 2024 |
| Model class | On site | Grid electricity | Total | Total without hydro | AWARE-weighted on site |
|---|---|---|---|---|---|
| Class A | 0.00179 | 0.0503 | 0.0521 | 0.0397 | 0.00319 |
| Class B | 0.00586 | 0.165 | 0.17 | 0.13 | 0.0104 |
| Class C | 0.0218 | 0.613 | 0.635 | 0.485 | 0.0388 |
Where do these numbers come from and how do I cite them?
Figures per query are one plain exchange, before the Tier 2b reasoning uplift. 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, data-centre construction 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). The modelled range of the models inside a class, around the class value, is −41% to +48% for class A, −39% to +24% for class B and −67% to +55% for class C.
Dataset: AI Inference Emission and Resource Factors for Corporate GHG Inventories (SOMA AI Emission Factors (SAEF)), Llopis-Garcia, G., SOMA AI, 19 October 2026, CC BY 4.0, concept DOI 10.5281/zenodo.20443585, which resolves to the latest version (this version 10.5281/zenodo.23371447, earlier version 2 values at 10.5281/zenodo.22767475). Method: Accounting for Purchased AI Inference in Corporate GHG Inventories: A Token-Based Method, arXiv:2606.10660, under peer review.
Partners can fetch any row from the AI factors API, and the whole table is a CSV download.
Cite as "0.0351 kg CO₂e per million tokens, class B, US South - Southeast, SAEF, 19 October 2026" and link somaai.earth/ai-footprint/factors/us-south.