ESRS E1-6 — dual-method climate disclosure¶
What ESRS E1-6 requires¶
ESRS E1-6 mandates disclosure of GHG emissions across scope 1, 2, and material scope-3 categories. For scope 2, dual-method disclosure (location-based and market-based) is required, with explanation where they materially diverge.
For AI inference scope-3 (a category-1 purchased-services scope-3 for most enterprises), the relevant outputs are:
- Location-based scope-2 — GHG emissions calculated using grid-average intensity at the location where electricity is consumed
- Market-based scope-2 — GHG emissions calculated using contractual instruments (RECs, GoOs, PPAs) where applicable
- Total scope-3 attributable to the service — including embodied amortisation
- Methodological notes — boundary, emission factors, calculation lineage
What we provide¶
For every receipt, we compute both:
- Location-based co2e:
energy_per_query × grid_intensity_at_location, using ENTSO-E live or regional ISO/TSO data - Market-based co2e:
energy_per_query × residual_grid_intensity, where the residual mix excludes contractually-allocated renewable supply
For our infrastructure (Scaleway, atNorth, OVH, Hetzner), the providers' contractual instruments are documented:
| Provider | Region | Contractual claim | Residual mix used for market-based |
|---|---|---|---|
| Scaleway | PAR-1, PAR-2 | "100% renewable" via French nuclear + REC mix | EU residual mix per AIB |
| atNorth | STO-1 | Nordic green electricity supply contracts | Nordic residual mix per AIB |
| atNorth | ISL-1 | Geothermal direct supply | Direct (no residual) |
| OVHcloud | GRA, RBX | "100% renewable" via mix of contracts | EU residual mix per AIB |
| Hetzner | FSN-1, HEL-1 | Mix of REC + grid; documented per-DC | EU residual mix per AIB |
We default to location-based on the receipt because it is the comparable, auditor-preferred number. The market-based number is available in the lineage URL and in the quarterly evidence pack.
Worked example¶
Suppose your account ran 4.2M queries in Q1 2026, distributed across regions:
| Region | Queries | Energy (kWh) | Location-based co2e | Market-based co2e |
|---|---|---|---|---|
| scaleway-par-1 | 3,100,000 | 715 | 41.5 kg | 12.8 kg |
| atnorth-sto-1 | 600,000 | 138 | 1.7 kg | 1.7 kg |
| atnorth-isl-1 | 200,000 | 46 | 0.8 kg | 0 kg (direct geothermal) |
| ovh-gra-1 | 300,000 | 69 | 4.0 kg | 1.4 kg |
| Total Q1 | 4,200,000 | 968 | 48.0 kg | 15.9 kg |
This translates to your CSRD report as:
AI inference scope-2 emissions, Q1 2026 (period to be aggregated to annual): location-based 48.0 kg CO₂e, market-based 15.9 kg CO₂e. Methodology: Vetted Inference v0.4.2 four-tier estimator, parametric tier with conformal-calibrated 90% intervals. Boundary: comprehensive (accelerator + host CPU/DRAM + provisioned idle + datacentre PUE). Emission factors: ENTSO-E live + JRC NEEFE 2024 (location-based); AIB residual mixes 2024 (market-based). 90% interval per receipt; aggregated interval available on request.
Where the divergence comes from¶
Location-based and market-based diverge by a factor of 3 in this example, dominated by Scaleway's renewable-procurement claims (its contracted mix is much greener than the French average grid). ESRS E1-6 requires both numbers; the explanation belongs in your sustainability-reporting narrative.
A reviewer's question to expect: "What is the residual-mix methodology and where is it documented?" Our answer: AIB (Association of Issuing Bodies) European Residual Mix 2024, per regional disclosure. Documented per-region in the evidence pack appendix.
Worked aggregate uncertainty¶
Per-receipt 90% intervals propagate to the aggregate. For independent (per-receipt) uncertainty, the aggregate's interval narrows roughly as 1/√N where N is the number of receipts. For 4.2M queries, the aggregate's relative interval is roughly 0.05% — much narrower than the per-receipt ±25%. Your auditor will accept the aggregate as the disclosure number.
For correlated uncertainty (model-family or region-wide bias), the conformal calibration handles this directly.