ESRS E5-1 — resource use and circular economy¶
What ESRS E5-1 requires¶
ESRS E5 covers resource use and the circular economy. Datapoint E5-1 calls for disclosure of resource inflows including critical raw materials, with attention to circular-economy mechanisms.
For AI inference, the relevant resource flow is the embodied resource depletion of the hardware (GPUs, host servers, networking, datacentre construction materials), amortised across the inference workload.
What we provide¶
Embodied resource depletion is tracked via the CML-IA Abiotic Depletion Potential (ADP) indicator. ADP measures the depletion of finite resources weighted by their global reserves; metals like indium, tantalum, and certain rare earths used in GPU manufacture have high ADP factors.
Per receipt, we estimate the embodied ADP share allocated to that query via:
Hardware lifetime inferences = expected lifetime (default 3 years per EcoLogits 2025 update) × utilisation × queries per second.
Worked example¶
For the same Q1 2026 example:
| Region | Queries | Embodied ADP share (kg Sb-eq) |
|---|---|---|
| scaleway-par-1 | 3,100,000 | 0.42 |
| atnorth-sto-1 | 600,000 | 0.08 |
| atnorth-isl-1 | 200,000 | 0.03 |
| ovh-gra-1 | 300,000 | 0.04 |
| Total Q1 | 4,200,000 | 0.57 kg Sb-eq |
Disclosure language:
AI inference embodied resource depletion, Q1 2026: 0.57 kg Sb-eq (CML-IA Abiotic Depletion Potential, allocated by GPU-hours consumed). Hardware lifetime assumption: 3 years per EcoLogits 2025 update. Methodology: Vetted Inference v0.4.2 with ecoinvent v3.10 hardware inventories.
Limitations and notes¶
- The CML-IA ADP indicator is one of several resource-depletion indicators; alternatives include the European Commission's "Critical Raw Materials" framework and the EF (Environmental Footprint) "resource use, minerals and metals" indicator. We default to CML-IA but provide EF on request for customers using EF as their reporting framework.
- Hardware lifetime is a key assumption; sensitivity analysis is published per region.
- Embodied amortisation is allocated by GPU-hours consumed by the customer's workload — not by query count alone — to reflect that long-context queries consume more hardware-time per query.
Circular-economy mechanisms¶
Where customer requests touch circular-economy considerations:
- Hardware refurbishment. Our upstream providers' hardware refurbishment programmes are documented per-provider. Refurbishment extends effective lifetime and reduces per-query ADP allocation.
- End-of-life. Provider e-waste handling is documented in the sub-processor list and in the evidence pack appendix.