Environment & energy
PrivatAI does not operate its own data centres. The energy and CO₂ figures below come from our two infrastructure providers — Scaleway for AI inference in France, Hetzner for hosting in Germany. This page reproduces them, with source and date.
Where PrivatAI runs
- AI inference — Scaleway, region fr-par (France). This is where essentially all energy use occurs: the GPU computation of your requests.
- Application, database, web — Hetzner, Falkenstein (Germany). A single server; a very small share compared with inference.
AI inference: what one request costs
Scaleway publishes a per-token environmental footprint for its Generative APIs — including the amortised manufacturing impact of the GPU nodes, data-centre efficiency (PUE) and the location-based carbon intensity of the French grid. Location-based means: calculated from the electricity physically on the grid, not from purchased guarantees of origin.
Hosting: what Hetzner publishes
Limits of these figures
No neutrality claim. We do not offset, and we do not describe the service as climate- or carbon-neutral.
No measurement of our own. Every figure on this page comes from our providers and has not been independently verified by us. Sources are linked below.
Not a complete footprint. Covered: inference and hosting. Not covered: email delivery (Resend), payment processing (Stripe), your own devices and network, and the training of the models we serve — that happens before our use and is not quantified by the model makers.
Sources
Methodology. Last reviewed by Scaleway on 2026-05-19.
Statements on hydropower, PUE, cooling and hardware lifecycle; certificates downloadable there.
Source of the per-token figures for gpt-oss-120b in region fr-par. Retrieved in August 2026.
Last updated: August 2026. Provider values change; we review this page every six months.