Glonce

Nvidia to buy Hugging Face for about $12.9 billion

businessThe Glonce editorsThe Decoder🔥81Glonce rating81Impact85Novelty70Relevance95Surprise60Players90Credibility90
Nvidia to buy Hugging Face for about $12.9 billion

Nvidia plans to buy Hugging Face for about $12.9 billion, according to a blog post from Nvidia CEO Jensen Huang published on September 3, 2026, following reports of the deal that surfaced in late August. The total splits into roughly $11.9 billion in purchase price plus a stock program of up to $1 billion meant to keep Hugging Face staff on board. Nvidia's SEC filing states the deal will not close until the first half of 2027 and still needs regulatory approval.

Hugging Face is the leading hub for open AI models. Nvidia says more than 18 million developers use the platform to share over 3 million models, 500,000 datasets and 1 million applications, and that more than 200,000 companies rely on the service. Huang says the platform will stay open to everyone after the acquisition: Nvidia hardware will not be required, and other clouds and chips will keep working on it. Nvidia is already, by its own account, the platform's biggest contributor of open models, with over 500 models and more than 250 datasets uploaded there.

Nvidia and Hugging Face give different accounts of how the deal came together. Huang says co-founder Clem Delangue approached him while thinking through the company's next chapter. Co-founder Thomas Wolf writes on LinkedIn that it was Huang who made Delangue the offer, to build Hugging Face into an open, independent and hardware-neutral platform. Wolf calls Nvidia the best-fitting partner for the company's mission across open weights, robotics and science, and says nothing changes for users today.

The strategic logic, as the piece lays it out: Nvidia earns its money selling compute infrastructure, and anyone using the API from OpenAI or another AI provider has mostly ended up on Nvidia chips anyway. But the largest AI providers, Google, Amazon, OpenAI and, more recently, Anthropic, are now building their own accelerators instead. Open models are different: they run across many clouds, inside companies, at universities and in government agencies, none of which build their own chips. Huang summed up the logic to Axios in one line: "Free AI should be great for hardware."

Nvidia already runs open-model efforts of its own. Its Nemotron models ship with weights, large parts of the training data, and the training and post-training recipes; a related Nemotron Coalition brings in Mistral AI, Thinking Machines Lab, Perplexity, Black Forest Labs, Cursor, LangChain, Reflection AI and Sarvam to build another open model. None of this puts Nvidia in the lead: its models sit behind the strongest Chinese models, though they run much faster in some scenarios. Nvidia also offers its own NVFP4 versions of the Chinese models Kimi K2.6 and GLM-5.1, tuned for its chips. Owning Hugging Face would give Nvidia more influence over the platform where this competition plays out.

There is a hosting angle too. Hugging Face both distributes models and rents out compute to run them, a business Nvidia has tried before and backed out of: it scaled back its own DGX Cloud because it did not want to poach rental customers from its big buyers. Since then, Nvidia's Lepton marketplace has routed workloads to partners such as CoreWeave, Lambda and Nebius instead, with Hugging Face already plugged into Lepton through a cluster service. In late July, Nvidia reported $36 billion in commitments from its AI cloud-partner agreements: under these deals, Nvidia's own commitments shrink whenever a partner resells spare capacity to a third party, and Nvidia only partly insures partners against unused capacity. A large developer hub folded into that arrangement gives Nvidia an extra sales channel to lean on.

Key facts

  • Nvidia's offer totals about $12.9 billion: roughly $11.9 billion in purchase price plus a stock program of up to $1 billion to retain Hugging Face staff.
  • Hugging Face hosts more than 18 million developers sharing over 3 million models, 500,000 datasets and 1 million applications, with more than 200,000 companies relying on the service, per Nvidia's own figures.
  • Huang and Hugging Face co-founder Thomas Wolf give conflicting accounts of who proposed the deal: Huang says co-founder Clem Delangue approached him, while Wolf writes on LinkedIn that Huang made the offer.
  • Per the SEC filing, the deal will not close until the first half of 2027 and still needs regulatory approval; Huang says the platform will stay open, with no requirement to use Nvidia hardware.
  • Nvidia runs its own open-model efforts, including Nemotron and the Nemotron Coalition (with partners such as Mistral AI, Perplexity and Black Forest Labs), and reported $36 billion in AI cloud-partner commitments in late July.

Why it matters

Nvidia's business is built on selling compute, and the AI labs richest enough to design their own chips, Google, Amazon, OpenAI and now Anthropic, are doing exactly that, cutting into Nvidia's leverage over them. Open AI models are the counterweight: they run across many clouds, companies, universities and government agencies that do not build their own silicon, which is exactly the customer base Hugging Face reaches at scale. Buying the platform gives Nvidia direct influence over the showcase where its own open-model efforts, including Nemotron and its NVFP4-tuned versions of the Chinese models Kimi K2.6 and GLM-5.1, compete for attention. As Huang put it to Axios: "Free AI should be great for hardware."

Who it affects

Directly: Hugging Face's stated user base of more than 18 million developers, over 3 million models, 500,000 datasets, 1 million applications and more than 200,000 companies. Indirectly: Nvidia's cloud partners such as CoreWeave, Lambda and Nebius, which take on workloads through Nvidia's Lepton marketplace that Hugging Face already plugs into, and rival labs building in-house chips, Google, Amazon, OpenAI and Anthropic among them. Hugging Face's own co-founders, Clem Delangue and Thomas Wolf, are affected most directly of all, though the source does not say what role either will hold once the deal closes.

How to use it

Nothing changes for current users today, according to Wolf. Huang says the platform stays open to everyone after the deal: Nvidia hardware will not be required, and other clouds and chips keep working as before. That said, the deal itself is not done. Per the SEC filing, it is not expected to close before the first half of 2027 and still needs regulatory approval, so there is no near-term change for anyone using Hugging Face to plan around.

How solid is it

The deal terms and timeline come from primary sources: Huang's own announcement and Nvidia's SEC filing. But the account of how the deal came together is contested. Huang says Delangue approached him; Wolf, writing separately on LinkedIn, says Huang made the offer. The usage figures, 18 million developers, 3 million models and the rest, are Nvidia's own stated numbers, not independently verified, and the source gives no date for when they were measured.

Risks and caveats

The deal is not final: it still needs regulatory approval, and the SEC filing puts the close no earlier than the first half of 2027, leaving room for the terms to change or for the deal to fall through. The source does not name the regulator or jurisdiction involved, does not say whether Hugging Face's board or shareholders have signed off, and does not say what role Delangue or Wolf will hold afterward. Huang's promise to keep the platform open and hardware-neutral is a stated intention, not a term the source describes as binding, and Nvidia has reversed a similar hosting ambition before: it scaled back its own DGX Cloud rather than compete with the cloud partners it relies on.

“Free AI should be great for hardware.”

— Jensen Huang, to Axios