OVHcloud AI Endpoints: our review

OVHcloud AI Endpoints gives access to more than forty open models — Llama, Qwen, DeepSeek and others — through an API billed per token consumed. The publisher is a French hosting provider, founded in 1999, that builds this offering on its own infrastructure rather than on cloud capacity rented from a third party.

The catalogue and per-token billing

The principle: one API call per request, one rate per million tokens depending on the model chosen. A small model like Qwen3.5-9B costs around €0.10 per million input tokens and €0.15 for output; a much heavier model like Qwen3.5-397B-A17B climbs to €0.60 and €3.60. A few content-moderation models, still in beta, are free. The catalogue covers text, vision, embeddings for document indexing, and speech synthesis or recognition. A €200 credit is offered on first Public Cloud projects, enough to test before committing.

What's been verified about hosting

AI Endpoints' product page doesn't say where the models run; it's the official launch announcement, published on OVHcloud's corporate site, that specifies it: the service is deployed from the Gravelines data centre, in France, and "data is hosted in Europe and protected against non-European regulations." The publisher adds that it never uses data to train or improve its models, and announces zero retention beyond what serves billing. On certification, however, caution is warranted: the product page mentions an "underlying platform certified ISO27000, SOC and health data", but no SecNumCloud qualification is claimed for this specific service. OVHcloud has indeed qualified other parts of its Public Cloud for SecNumCloud, and states it wants to extend the approach to its most recent AI offerings — with no qualification date published for AI Endpoints so far.

Where it falls short

Three reservations. First, SecNumCloud certification is missing for this specific service, which rules AI Endpoints out of markets that formally require it today. Second, as with any provider of this kind, there's no interface: it's a technical building block to integrate, not a product for a non-developer user. Third, the price gap between models is significant and no cost simulator appears on the catalogue page: better to cost out real usage before budgeting a project.

Our recommendation

For a technical team that wants a French provider with processing announced in Europe, AI Endpoints is a serious option to compare against Scaleway before choosing, notably as part of an internal AI agent. If your need is immediate SecNumCloud certification, or a no-code product to install, this isn't the right choice: our automation services offer more suitable support.