And now four, equal to the number of European inference providers on Hugging Face.
Nebius AI (the Netherlands) had joined the loop in February. Nscale (the United Kingdom), in May. They’re now joined by Public AI (Switzerland)*… and by Scaleway.
These providers – along with about a dozen others – are integrated at several levels within the Hugging Face ecosystem :
- Inference widget on model pages
- Playground
- Dataset pages (text-to-SQL conversion)
- Python and JavaScript SDKs
- REST API
The OpenAI (Python) client is another access option, but only for text autocompletion.
Scaleway & Cie for prototyping, the hyperscalers for production
By default, Hugging Face routes each request to the appropriate provider and handles billing; for now, it takes no commission (there was a time when revenue-sharing agreements were discussed). It is, however, possible to fix which providers you use by supplying API keys.
Free Hugging Face accounts come with $0.10 of credits per month, with no option to spend beyond that. PRO accounts grant $2 in credits and unlock usage-based billing. For enterprise plans, it’s $2 in credits per seat. The message is clearly geared toward prototyping—indeed CPU inference (AWS, Google and Microsoft handle deployment aspects) is the focus here.
On the Scaleway side, the gateway to the Generative APIs service has been implemented. The payoff is a selection of around ten models, including Qwen, Llama, a Gemma, a DeepSeek and gpt-oss-120b.
| Provider | Auto-complete (LLM) | Auto-complete (VLM) | Feature extraction | Text-to-image | Text-to-video | Speech-to-text |
|---|---|---|---|---|---|---|
| Cerebras | ? | |||||
| Cohere | ? | ? | ||||
| Fal AI | ? | ? | ? | |||
| Featherless AI | ? | ? | ||||
| Fireworks | ? | ? | ||||
| Groq | ? | ? | ||||
| HF Inference | ? | ? | ? | ? | ? | |
| Hyperbolic | ? | ? | ||||
| Nebius | ? | ? | ? | ? | ||
| Novita | ? | ? | ? | |||
| Nscale | ? | ? | ? | |||
| Public AI | ? | |||||
| Replicate | ? | ? | ? | |||
| SambaNova | ? | ? | ||||
| Scaleway | ? | ? | ||||
| Together AI | ? | ? | ? |
Also worth checking:
The sovereign DNS4EU project is taking shape… with Scaleway involved
MLPerf: the quest for representative AI benchmarks
SLM or LLM? For agent-based systems, NVIDIA has picked a side
CyberSOCEval: a cyber analysis testbed for LLMs
* A non-profit organization that promotes the work of public-model developers such as the Swiss AI Initiative, AI Singapore, AI Sweden, and the Barcelona Supercomputing Center
