AI carbon footprint in US East - N. Virginia: 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 East - N. Virginia?
A 1210-token query to a class B model (gpt-4o) served from US East - N. Virginia emits about 0.036 g CO₂e, and a million tokens emit 0.0261 kg CO₂e.
Rank 9 of 16 for class B, lowest carbon first (full table). 3.6x the level of EU North - Sweden, Asia Pacific - Singapore is 1.5x higher.
Which grid supplies the electricity in US East - N. Virginia?
US East - N. Virginia maps to SERC Virginia/Carolina, a US eGRID subregion, at 303.8 g CO₂e per kWh for 2023. That is 1.3 times the EU-27 average (236 g) and 0.6 times the global average (483.2 g). The source is eGRID2023 SRVC 596.3 lb/MWh CO2e total output x 1.1232 (Ember lifecycle basis).
Cloud regions on this grid, 5 in all (1 AWS, 2 Azure, 2 Google Cloud).
| Provider | Region id | Region name | Location | Mapping confidence | Mapping note |
|---|---|---|---|---|---|
| AWS | us-east-1 | US East (N. Virginia) | N. Virginia | high | Dominion territory; subregion settled in paper (anchor). Polygon test at Ashburn VA also SRVC. |
| Azure | eastus | East US | Virginia | high | Provider states state only; Virginia spans SRVC and RFCE. Site widely reported at Boydton VA (not in the official list); polygon test at Boydton (36.67,-78.39) and Ashburn = SRVC. CONFIRMED 2026-10-14 by Data Center Map (datacentermap.com, operator pages for Microsoft and Amazon AWS, read 14 October 2026): Microsoft Boydton campus, 101 Herbert Dr, Boydton VA 23917 (52 listings, filed under Richmond); polygon test at Boydton (36.67,-78.39) = SRVC; agrees with the assumed site, confidence raised from medium to high. |
| Azure | eastus2 | East US 2 | Virginia | high | Provider states state only; Virginia spans SRVC and RFCE. Site widely reported at Boydton VA (not in the official list); polygon test at Boydton (36.67,-78.39) = SRVC. CONFIRMED 2026-10-14 by Data Center Map (datacentermap.com, operator pages for Microsoft and Amazon AWS, read 14 October 2026): Microsoft Boydton campus, 101 Herbert Dr, Boydton VA 23917 (52 listings, filed under Richmond); polygon test at Boydton (36.67,-78.39) = SRVC; agrees with the assumed site, confidence raised from medium to high. |
| Google Cloud | us-east1 | us-east1 | Moncks Corner, South Carolina | high | Polygon test at Moncks Corner (33.20,-80.00) = SRVC. |
| Google Cloud | us-east4 | us-east4 | Ashburn, Virginia | high | Polygon test at Ashburn (39.04,-77.49) = SRVC; same as AWS us-east-1 anchor. Points ~0.25 deg away fall in RFCE/RFCW, so site is near a boundary, but Loudoun County is Dominion. |
The note for AWS us-east-1 reads "Dominion territory; subregion settled in paper (anchor). Polygon test at Ashburn VA also SRVC.".
Which grids have an intensity close to SERC Virginia/Carolina?
| Difference g | Grid | Code | g CO₂e per kWh | Ratio |
|---|---|---|---|---|
| 1.0 | Chile | CHL | 304.8 | 1.003 |
| 1.5 | RFC East | RFCE | 305.3 | 1.005 |
| -2.9 | Nicaragua | NIC | 300.9 | 0.990 |
| 6.1 | Guatemala | GTM | 309.9 | 1.020 |
| -8.7 | Liberia | LBR | 295.1 | 0.971 |
| -12.0 | Peru | PER | 291.8 | 0.961 |
| 14.1 | Suriname | SUR | 317.9 | 1.046 |
| 19.5 | Italy | ITA | 323.3 | 1.064 |
| 19.9 | WECC Northwest | NWPP | 323.7 | 1.066 |
| -21.3 | Ireland | IRL | 282.5 | 0.930 |
Emission factors for US East - N. Virginia 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) | 0.00661 | 0.00027 | 0.00014 | 0.00060 | 0.008 | 0.01 | 0.053 | 0.073 |
| Class B (mid-size) | 0.02166 | 0.00101 | 0.00046 | 0.00293 | 0.026 | 0.036 | 0.172 | 0.24 |
| Class C (frontier) | 0.08065 | 0.00388 | 0.00169 | 0.01284 | 0.099 | 0.13 | 0.642 | 0.89 |
| Model class | Central | Class range low | Class range high | High case | B200 | Development |
|---|---|---|---|---|---|---|
| Class A | 0.007625 | 0.004783 | 0.01096 | 0.01206 | 0.006718 | 0.009018 |
| Class B | 0.02606 | 0.01714 | 0.03135 | 0.0406 | 0.01866 | 0.02953 |
| Class C | 0.09907 | 0.04171 | 0.1459 | 0.1533 | 0.1017 | 0.11 |
How much does siting matter for a model served from US East - N. Virginia?
Electricity is 83.1% of the class B total here (class A 86.7%, class C 81.4%), with hardware at 3.9%, construction 1.7% and training 11%. Electricity carries most of the 0.0261 kg, so the grid is the biggest lever.
The example company (2,323 million class B tokens a year) emits 61 kg CO₂e from US East - N. Virginia. Moving to EU West - Ireland would save 3.5 kg (5.8%) or Netherlands would save 5.9 kg (9.7%).
| Region | Grid g per kWh | Class A | Class B | Class C |
|---|---|---|---|---|
| EU West - Ireland | 282.5 | 1.08 | 3.53 | 13.1 |
| Netherlands | 268.2 | 1.8 | 5.9 | 22 |
How much water does AI usage consume in US East - N. Virginia?
A class B query of 1210 tokens from US East - N. Virginia consumes about 0.24 mL of water and a million tokens 0.172 litres, or 0.144 litres without the water attributed to hydropower.
| Input | Value |
|---|---|
| On-site WUE, L per kWh | 0.06 |
| WUE denominator | IT load |
| WUE basis as published | withdrawal |
| Withdrawal to consumption | 0.6667 |
| EWIF, L per kWh | 2.385 |
| EWIF without hydro, L per kWh | 1.981 |
| AWARE factor | 1.62 |
| Confidence | published-site |
| Source | AWS; AWS Region U.S. East (Northern Virginia) us-east-1; 2025 |
| Model class | On site | Grid electricity | Total | Total without hydro | AWARE-weighted on site |
|---|---|---|---|---|---|
| Class A | 0.000726 | 0.0519 | 0.0526 | 0.0438 | 0.00118 |
| Class B | 0.00238 | 0.17 | 0.172 | 0.144 | 0.00385 |
| Class C | 0.00885 | 0.633 | 0.642 | 0.535 | 0.0143 |
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 "0.0261 kg CO₂e per million tokens, class B, US East - N. Virginia, SAEF, 19 October 2026" and link somaai.earth/ai-footprint/factors/us-east.