Nvidia is buying Hugging Face for $12.93 billion

Nvidia is buying Hugging Face for $12.93 billion, the largest acquisition in the chipmaker's history.

Hugging Face is the platform where most open-source AI models, datasets, and developer tools are hosted and downloaded. It functions as the default app store and version-control hub for the open-source AI world: developers go there to grab model weights (the trained parameters that make a model work), fine-tune them, post evaluations, and publish derivative releases. Nvidia's chips power most of the training and inference behind that activity. Owning Hugging Face gives Nvidia a position in the layer directly above its own hardware.

The deal, announced Sept. 3, prices Hugging Face at roughly triple its last private mark of $4.5 billion in a 2023 funding round that Nvidia itself co-led. A year ago, Hugging Face turned down a $500 million Nvidia investment offer that would have valued the company at $7 billion. The same company that refused a smaller check at a lower price is now selling outright at nearly double that implied valuation. Whatever changed between the two offers, it was not the price floor.

Nvidia CEO Jensen Huang said in the announcement that "Hugging Face will remain an open platform" and that "Nvidia compute will not be required to build on or deploy through Hugging Face." Developers, the statement continued, will still be able to choose their models, frameworks, clouds, inference providers, and compute. Whether that firewall survives contact with Nvidia's quarterly incentives is the question every open-source maintainer is now asking. Business Insider reported that the deal included explicit language preserving the platform's open character.

OpenAI, Anthropic, and Google, the three largest closed-AI labs (labs that keep their model weights private and sell access through an API rather than publishing them), are all running or commissioning custom AI silicon designed to reduce their dependence on Nvidia GPUs. The largest customers of Nvidia's chips are now publicly funding substitutes. Buying Hugging Face lets Nvidia own the open-model commons, the part of the AI ecosystem that is structurally least likely to migrate to a closed lab's custom chip, and seed that layer with Nvidia-friendly defaults: optimized inference runtimes, fine-tuning recipes, and benchmarks that already target CUDA, Nvidia's proprietary GPU programming layer.

The deal also buys Nvidia leverage in a different negotiation. The closed labs can build their own chips, but they still depend on the broader open-model ecosystem for tooling, evaluation suites, and a community of independent developers. Whoever hosts that commons now has a seat at every standards conversation about what "open" means in AI, including the definitions regulators and enterprise buyers are starting to write into procurement language.

The risk for Hugging Face's community is the obvious one. A platform that promised neutrality is now owned by a hardware vendor with a vested interest in keeping open-source AI running well on its own chips. The seller's promise is not enforceable, only credible, and the first concrete test will be whether Hugging Face's inference and fine-tuning defaults stay vendor-neutral or quietly drift toward CUDA-only paths.

Nvidia's bet is that owning the substrate of open-source AI is worth more than selling chips into it. The open-model community's bet is the opposite, and the deal will be settled on every pull request, every model card, and every choice of where to host from here.