Nvidia said on Thursday, September 3, that it will buy Hugging Face, the platform used by more than 18 million artificial intelligence developers, for $12.93 billion, about 65.7 billion reais. It is one of the largest acquisitions ever made by the chipmaker, which is betting that support for open-weight AI models will keep demand for its computing power high even as its biggest customers design their own chips to rely less on the supplier. Nvidia shares traded slightly lower after the announcement, according to a Reuters report carried by Folha de S.Paulo. Hugging Face is best known as the largest public library of AI models, the place where companies and independent developers post, test and share systems anyone can download and adapt.
On the other side of the counter, Hugging Face was founded in New York in 2016 by French entrepreneurs Clément Delangue, Julien Chaumond and Thomas Wolf. TechCrunch reports the site hosts 3 million models, 1 million applications and 500,000 datasets. The startup had raised just over $395 million in its lifetime; its last round, $235 million in 2023, was led by Salesforce Ventures, with Google, Amazon, IBM and Nvidia itself among the investors. A year ago it turned down a $500 million offer from Nvidia, according to the Financial Times. In August, The Information put its annualized revenue at $150 million. Nvidia is now paying roughly 33 times everything the company ever raised, and more than 80 times its estimated annual sales.
Nvidia's bet
Nearly all open models already run on Nvidia hardware, according to CEO Jensen Huang, and Nvidia itself has published more than 500 models and 250 open datasets on the platform. The purchase gives the company direct control of the channel where that community collaborates, tests and shares tools, which Reuters says could yield valuable data to close the technology gap with top U.S. and Chinese labs. TechCrunch points to another gain: packaging Hugging Face's offering with Nvidia's idle computing capacity to sell it to enterprises.
"Hugging Face will remain an open platform for the entire AI ecosystem. Developers will choose the models they want, the frameworks they want, the clouds and inference service providers they want and the computing platforms they want. Nvidia compute will not be required to build on or deploy through Hugging Face," Nvidia CEO Jensen Huang wrote in a company blog post.
Demand for open-weight models has surged as companies resist the high cost of deploying the technology, Reuters reports. In that market, Chinese firms such as DeepSeek and Z.ai have gained ground with models that rival the best U.S. systems at tasks like code generation, at lower cost, feeding fears that American companies could become dependent on Beijing-built technology. Nvidia already plays in this field with its Nemotron model, said on its latest earnings call that it has put more than $50 billion into frontier AI labs, and struck a $6 billion deal in August with coding startup Poolside, the Wall Street Journal reported.
The breach in the background
Hugging Face became a household name in tech in July, when autonomous AI agents tied to an unreleased OpenAI model escaped a testing environment and broke into the platform. Delangue said at the time that an open Nvidia model helped defend the site after proprietary tools failed. On X, he said the community had proved the company could be an alternative to closed APIs, "but for it to happen at a larger scale, it needs more compute, more support, more collaboration, and more visibility. That's why we went to talk to Jensen."
For developers who build on the platform, the public promise is continuity: access stays open and Nvidia hardware will not be required. For companies that consume AI, the reading is economic: demand for open models is driven by high deployment costs, according to Reuters, and the industry's biggest chip supplier now owns the main catalog of those models. The tension sits with Nvidia's own largest customers, who are investing in their own chips precisely to depend less on it.