A chipmaker buys the model registry
Nvidia said on 3 September that it will acquire Hugging Face, the platform on which most of the world’s openly available AI models are published, for $12.93bn. According to France 24, $11.9bn of that goes to Hugging Face’s investors and up to $1bn is an equity-based retention programme for employees who join Nvidia.
Hugging Face hosts more than three million models, 500,000 datasets and one million applications, Nvidia said in its announcement. The company puts the platform’s reach at 18 million developers, researchers and creators, with more than 200,000 companies using it to discover, evaluate, customise and deploy AI.
The price is close to three times Hugging Face’s last disclosed valuation. Euronews reports the company was worth $4.5bn after a $235m round in 2023 — a round Nvidia itself invested in.
“NVIDIA compute will not be required”
The sentence the announcement leans hardest on is a commitment about neutrality. Hugging Face will stay an open platform, Nvidia said, with developers free to choose “the models they want, the frameworks they want, the clouds and inference service providers they want and the computing platforms they want” — and, explicitly, “NVIDIA compute will not be required to build on or deploy through Hugging Face.”
Nvidia framed the deal as continuous with what it already does on the site. It describes itself as the largest open-model contributor to Hugging Face, having published more than 500 models and over 250 open datasets there.
Jensen Huang, Nvidia’s chief executive, said he was “honored that Clem came to me as he considered the next chapter of Hugging Face and believed NVIDIA would be a great home for the company, its community and the future of open models” — a reference to Hugging Face co-founder and chief executive Clément Delangue, who France 24 and Euronews report opened the conversation over the summer.
Why a registry is worth $12.93bn
Hugging Face is not a model lab. Its value is positional: it is where an open model goes to be found, benchmarked, quantised, fine-tuned and pulled into a pipeline. A weights file that is not on Hugging Face effectively does not exist for most of the ecosystem, and the platform’s libraries decide which accelerators a model runs on out of the box.
That is the same layer Nvidia’s competitors depend on. AMD, Intel, Google’s TPUs and Apple silicon all reach open-model developers through Hugging Face integrations, and all of them now do so on infrastructure their largest rival owns.
The scepticism
That is where the objections land. Axel Rudolph of IG Group told France 24 that Nvidia is “buying strategic influence”, and Harold Byun, chief executive of BlueRock, raised the possibility that the company could gradually neglect rival hardware rather than block it outright — a slower failure mode than any promise in a press release is written to cover.
Nvidia’s announcement did not name a closing date, and it did not list the regulatory approvals the transaction will need. Given the size of the deal and Nvidia’s position in AI accelerators, that omission is the first thing worth watching.
What to watch next
The test of the neutrality pledge is not the wording, it is the maintenance. Watch whether Hugging Face’s inference stack and hardware integrations keep first-class support for non-Nvidia accelerators over the next several releases, and watch which competition authorities open a review before the deal closes.
