Nvidia acquires Hugging Face, open AI hub

The acquisition comes as Nvidia faces a strategic shift. Big cloud providers and AI labs are designing their own accelerators, which could reduce their reliance on Nvidia chips. Open models, by contrast, run across diverse environments—clouds, companies, universities, government agencies—where customers typically don't build custom silicon. By owning Hugging Face, Nvidia gains influence over a key distribution channel for open models, potentially steering more compute workloads to its hardware.

Nvidia has been expanding its open-model efforts, releasing Nemotron with weights, training data, and recipes. It also collaborates with partners like Mistral AI and others in the Nemotron Coalition. However, it trails leading Chinese models in performance, though its models run faster in some scenarios. The company also benefits from Chinese open models by offering tuned versions for its chips. Hugging Face's hosting business, which rents compute time, could complement Nvidia's existing Lepton marketplace that routes jobs to partners like CoreWeave and Lambda.

Nvidia reported $36 billion in commitments from AI cloud partners in late July. The company's commitments shrink when partners sell capacity to third parties, and it partly insures partners against unused capacity. A large developer hub like Hugging Face could serve as an additional sales channel, potentially helping Nvidia secure more rental demand and stabilize its cloud ecosystem.