Nvidia buys Hugging Face for $12.93B, vows to keep it open
The registry 3M models and 18M developers rely on now belongs to a chip vendor. Nvidia pledges neutrality; the community isn't sold.
Copy markdown
The deal, confirmed by Jensen himself
Nvidia will buy Hugging Face for $12.93B, confirmed in a Sept 3 post authored by CEO Jensen Huang after Clement Delangue approached him. HF hosts 3M models, 1M apps, and 500K datasets for 18M+ developers, making it the default registry your `from_pretrained` calls already hit.
'Nvidia compute will not be required'
The openness pledge is explicit: developers keep their choice of models, frameworks, clouds, and accelerators, and building or deploying on the Hub won't require Nvidia hardware. Nvidia is already the platform's largest open contributor, with 500+ models and 250+ datasets, so nothing in your workflow changes today.
Why the community isn't buying the neutrality line
HF's whole value is hardware neutrality: AMD, Intel, and AWS chips are first-class via Optimum. Owning both the Hub and Inference Endpoints, Nvidia can make its NIM/CUDA path the default with fewer steps, nudging you toward its silicon without banning anyone, while rivals may quietly pull the integration engineering they no longer control.
Elsewhere: Claude Code 2.1.259 goes headless-friendly
The new release adds managed MCP servers for org-wide deployment and `--permission-prompts none` for unattended headless hosts, both wins for cleaner CI and scheduled runs. It also now recognizes GitLab merge requests, showing them as `MR !N`.