OVHcloud brings sub-200ms inference to Hugging Face, starting at €0.04 per million tokens

Hugging Face is expanding its Inference Providers ecosystem with OVHcloud, bringing a distinctly European option to the serverless inference game. This isn’t just another name on the list — OVHcloud’s AI Endpoints are running on sovereign European infrastructure, which matters a lot if you care about where your data lives and how fast it moves. We’re talking sub-200ms time-to-first-token, a number that puts it squarely in the conversation for interactive apps and agentic workflows where latency kills the experience.

The integration is clean. Browse models on the Hub, pick one compatible with OVHcloud, and you’re off. You can either bring your own API key and get billed directly, or route through Hugging Face and just pay the standard provider rate with no markup. The pricing is aggressive: pay-per-token starts at €0.04 per million tokens. That’s cheap enough to make you wonder about margins, but OVHcloud is clearly betting on volume and its own infrastructure costs to make it work. Supported models include the usual suspects — gpt-oss, Qwen3, DeepSeek R1, Llama — so you’re not stuck in a walled garden.

One detail I appreciate: the preference ordering. In your Hugging Face settings, you can rank providers so the widget and code snippets on model pages automatically favor your pick. No hunting through dropdowns. On the billing side, PRO users get $2 in monthly inference credits that work across any provider, which is a nice perk for experimenting without commitment. The client SDK integration is straightforward for both Python and JavaScript, requiring nothing more than specifying a model string like “openai/gpt-oss-120b:ovhcloud.”

OVHcloud is positioning this as a production-grade option, not a sandbox. Structured outputs, function calling, multimodal — the features you’d expect from a frontier endpoint are all there. The real question is how the market shakes out when every cloud provider wants to be your inference backend and the real differentiator becomes reliability at scale. OVHcloud’s European data center footprint gives it a regulatory edge, but it’ll need to prove it can handle spikes without throttling when a model goes viral on the Hub.