Why Clément Delangue Sold Hugging Face to Nvidia After Turning It Down

As of a few months ago, Hugging Face CEO Clément Delangue was still reluctant to have a single dominant investor in his company, which hadn’t raised outside funds in more than two years. “We don’t need to raise too much money, and we’ve decided not to in the past few years,” Delangue told Observer in June. Most notably, Hugging Face turned down a $500 million investment offer from Nvidia last year. The tables have turned 180. After over a week of speculation and reported talks, Hugging Face is now selling itself in its entirety to Nvidia for $12.9 billion, Nvidia announced today (Sept. 3).

The amount is nearly three times Hugging Face’s last private-market valuation three years ago. And it’s Nvidia’s second-largest acquisition on record, after its $20 billion purchase of chipmaker Groq in December.

In a joint interview with CNBC today, Nvidia CEO Jensen Huang said Delangue reached out to him over the summer, and the deal was finalized within weeks. Huang also said Delangue told him he was ready for his “next chapter” and that he “has other interests.”

Delangue said he and his two Hugging Face co-founders, Julien Chaumond and Thomas Wolf, will join Nvidia after the acquisition.

Hugging Face is a GitHub-like hosting platform used by millions of developers and organizations. It doesn’t build or train its own models, which makes its cost structure and fundraising needs fundamentally different than other headline-grabbing A.I. companies.

“We’re lucky to have a business that doesn’t require burning tons of money like the model builders. We’ve decided to take a more sustainable, reasonable approach,” Delangue told Observer in June.

The company’s last funding took place in August 2023, raising a $235 million round led by Salesforce with participation from almost every major A.I. chipmaker—Nvidia, Intel, AMD, Qualcomm, Google and Amazon. Since then, Delangue said the company had been able to “self-fund” with revenue.

The platform currently has more than 18 million users, including some 200,000 enterprise users. It hosts more than 3 million models, 500,000 datasets and 1 million applications. Nvidia itself is the largest user, having released more than 500 models and more than 250 open datasets on Hugging Face.

“Most people building A.I. today are using Hugging Face in some ways,” Delangue said in June.

Why open source is suddenly hot

For Hugging Face, the acquisition comes against a concerning backdrop. The company was recently at the center of a hacking incident involving OpenAI’s autonomous agents, raising concerns about the rapid evolution of powerful A.I. getting out of the hands of human control.

“We couldn’t defend ourselves with closed-source, proprietary APIs. We had to use open models to defend ourselves. It proved the importance of open models,” Delangue told CNBC today.

Delangue said he sees a future where closed, proprietary models and open models grow on parallel paths. With proprietary model makers like OpenAI and Anthropic set for massive IPOs, he felt he needed to “double down” on the open source side of A.I.

“We realized that Hugging Face and open-source A.I. in general was at the turning point, and that it needed more, more resources, more scale, more visibility,” he said of the motivation behind turning to Nvidia.

Aside from security considerations, growing concerns about the high costs of using proprietary models are making open-source models—whose capabilities are catching up with frontier models anyway—more appealing. Instead of using expensive frontier models such as OpenAI’s GPT and Google’s Gemini for all tasks, developers can use platforms like Hugging Face to choose the models best suited to their purposes without overspending.

“We don’t need to drive a Ferrari to go grocery shopping,” Lin Qiao, the founder and CEO of Fireworks AI, a Hugging Face competitor, recently explained to Observer. “There are so many tasks we solve day-to-day at varying levels of complexity… If you select the best model suitable for solving a particular task, you get the quality you need at the lowest cost.”

“Open models let startups, businesses, universities and public institutions build on advanced capabilities without training every model from scratch. They enable organizations to match the right model to the right job,” Huang wrote in a blog announcing the acquisition today.

After the ownership handover, Hugging Face will remain “an open platform for the entire A.I. ecosystem,” Huang said. He added that developers can still choose the models, inference service providers and the computing platforms they want. “Nvidia compute will not be required to build on or deploy through Hugging Face.”

For Huang, Nvidia is benefiting from the growth of both parallel paths Delangue mentioned. Ultimately, the more people use A.I., the bigger the demand for its chips. “The world will need both closed models and open models. And both closed models and open models are skyrocketing in use…and they’re simultaneously driving our sales,” he told analysts on Nvidia’s quarterly earnings call last week.