AI Compute Carbon Estimator
Estimate the order of magnitude of training + inference emissions by GPU / PUE / grid factor. The built-in factors are illustrative — the result is not a Scope 2 disclosure figure; replace them with assured grid factors and measured vendor energy data before reporting.
Training Workload
Inference / Online Serving (annual)
Factor basis: grid emission factors come from official publications (China MEE, US EPA eGRID, UK DEFRA and others), each labelled with its data year and basis in the dropdown. GPU figures are vendor spec-sheet TDP and exclude host, network, storage and cooling overheads; PUE and utilisation are your own inputs. The result is therefore an order-of-magnitude estimate and is not suitable for formal disclosure or assurance without verification.
Every grid factor is labelled with its publishing authority and data year, and links to the source; any entry without a source is illustrative and must not be cited as IEA, national grid-authority or any other official dataset figures. GPU power uses spec-sheet TDP magnitudes and excludes host, network, storage and cooling overheads; PUE and utilisation are assumptions you enter — most of the uncertainty in the result comes from those two, not from the grid factor.
An order-of-magnitude estimate from official grid factors and your own assumptions — not suitable for formal disclosure or assurance.
Equivalent to (illustrative conversion)
Reduction scenarios (illustrative)
The scenarios below are illustrative recalculations from your inputs, used to compare the relative magnitude of each path; they do not represent any verified abatement result.
- Renewable procurement (market-based Scope 2 to zero)−205.3 tCO₂e
- Improve data-centre PUE to 1.15−28.2 tCO₂e
- Migrate to the cleanest grid in the factor library (United Kingdom — transmission and distribution losses (DEFRA/DESNZ 2026 factor set), 0.01299 kgCO₂e/kWh)−200.9 tCO₂e
All three levers are recomputed from the inputs above rather than fixed percentages: renewable procurement zeroes market-based electricity emissions; the PUE lever holds IT load constant and changes only facility efficiency; the grid lever re-prices the same kWh at the lowest factor in the library.
- ① Judge the magnitude: see whether training or inference dominates, and set internal abatement priorities first.
- ② Compare scenarios: vary grid region, PUE or utilisation to compare deployment options relatively.
- ③ Before any formal disclosure, replace nameplate TDP with measured energy data, confirm the grid-factor year matches your reporting period, and have your ESG owner review it.
Output from this tool is not a Scope 2 disclosure figure and cannot be entered directly into a CSRD / ISSB S2 / TCFD report.
Content as of 2026-09-12 · corpus 2026.09|Compiled from public official sources with links and review dates for compliance decision support. Rules change; have licensed professionals review before external disclosure, regulatory filing or contractual commitments.