Nvidia agrees to buy Hugging Face for $12.93 billion and pledges to keep the platform open

NVIDIA Corporation told the U.S. Securities and Exchange Commission on 3 September that it had signed a definitive agreement on 2 September to acquire Hugging Face Inc. The filing lists an approximately $11.9 billion purchase price payable to Hugging Face stockholders, subject to adjustments, and an equity retention program of up to about $1.0 billion for staff who join Nvidia. The company and trade press put the headline value at $12.93 billion.

The deal is expected to close in the first half of 2027, after regulatory review. It is Nvidia’s largest outright software-platform acquisition. The $6.9 billion purchase of Mellanox in 2020 remains the template for how the chipmaker uses a buyout to own a chokepoint. A later $20 billion purchase of Groq, reported in some accounts as Nvidia’s largest deal, sits in a different category: silicon. Hugging Face is distribution.

Hugging Face runs the main public repository for open-source machine-learning models, datasets and demo apps. Jensen Huang wrote that the platform holds more than 500,000 datasets and four times as many models. Other briefings put the user base above 18 million and the model count between two and three million, with more than 200,000 companies pulling weights from the site. Annualized revenue is about $150 million, according to The Information, which makes the multiple on sales extreme.

Clément Delangue, Julien Chaumond and Thomas Wolf founded the company. Delangue had turned down a large Nvidia investment last year at a $7 billion valuation in order to stay independent. In a joint interview with Huang on CNBC, Delangue said he reached out over the summer and that the papers were finished in weeks. Huang said Delangue told him he was ready for a “next chapter.” All three founders are due to join Nvidia after closing.

AMD, Intel and Qualcomm backed Hugging Face’s most recent funding round. Salesforce had looked at a purchase before Nvidia moved. Those names will collect part of the $11.9 billion. They will also have to decide whether to keep shipping models to a hub owned by the company that sells most of the training chips.

Huang has pledged that Hugging Face will “remain an open platform for the entire AI ecosystem” and that users will not be required to run Nvidia hardware. The 8-K repeats the pledge in legal language: Hugging Face would continue to let model makers upload and download weights of their choosing and would support other silicon vendors. That commitment is the point regulators will test. A platform that ranks, hosts and recommends models sits upstream of every open-weight lab. Control of that ranking is a form of industrial policy.

The timing is not accidental. Open-weight models have closed much of the quality gap with closed systems on many tasks. Labs that cannot buy clusters at Nvidia’s scale still need a place to publish. If that place is inside Nvidia, the chipmaker captures both the GPU sale and the distribution layer that decides which models get copied. Huang frames the purchase as a way to “speed up the spread of open models.” Competitors will frame it as a bid to own the commons.

Closing in 2027 gives antitrust offices in Washington, Brussels and Beijing time to demand structural promises. The filing already contains one: the platform stays open and multi-vendor. Whether that promise survives product roadmaps is the question that will sit on Hugging Face’s homepage after the logo changes.