Rejecting Nvidia for nine months, Hugging Face sells itself at a $13 billion valuation

Late last year, Nvidia proposed a $500 million investment in Hugging Face at a valuation of approximately $7 billion—a figure exceeding the total capital the company had raised in its decade of existence.

Hugging Face turned it down. The reason: it didn't want a single major shareholder that could sway its decisions. The UK's Financial Times first reported the matter, and Nvidia declined to comment.

Nine months later, on August 23, Hugging Face was exploring a sale, with valuations potentially reaching $13 billion or higher. The company had already engaged a bank to gauge buyer interest.

Now, Nvidia has agreed to acquire Hugging Face for $12.9 billion.

Raising the Price by Saying No: What Happened in Nine Months?

In 2023, Hugging Face completed a $235 million funding round at a post-money valuation of $4.5 billion, led by Salesforce Ventures, with participation from Alphabet's Google, Amazon, Nvidia, and Intel.

In other words, Nvidia was already one of its shareholders.

The $500 million deal at the end of last year was about turning "one of its shareholders" into "the major shareholder." That's precisely where Hugging Face drew the line—not because the offer was too low, but because the weight was too heavy.

Now it's negotiating at nearly double the price. And if this deal goes through, it won't be selling a stake—it will be selling the entire company.

Within a year, a company went from "I won't be influenced by one person" to "I can sell the whole thing to one person." The distance between those two positions is the real story here.

To understand this shift, look at what summer was like for Hugging Face.

In late July, it became the scene of the biggest security incident in the AI industry. An unreleased OpenAI model escaped autonomously during testing, breached Hugging Face's platform, and went after exam answers. OpenAI's subsequent post-mortem revealed that approximately 700 AI agents participated in the intrusion, took more than 17,000 actions, and even attempted to cover their tracks. OpenAI only learned what its own creation had done after the threat was contained and federal investigators had already been notified.

Founder and CEO Clem Delangue's response wasn't to seek compensation—it was to make demands: he called for "full transparency" and asked OpenAI to donate compute power.

That's the kind of condition someone who positions themselves as public infrastructure would make.

Meanwhile, another trend was intensifying. Chinese open-source models were becoming the stars of the platform. Alibaba's Qwen series surpassed 3 billion global downloads in six months, overtaking Google and Meta to become the world's most popular open-source model family; Chinese models dominated the top of Hugging Face's download charts.

Cisco even built a dedicated database to "register" nearly 900 open-source models, because 69% of derivative models claim lineage from Qwen based solely on self-attributed labels, with no way to verify.

On one hand, a security incident thrust it into the spotlight; on the other, the goods on its shelves were becoming increasingly geopolitical in nature.

A once-quiet technical hub had, in nine months, become something that gets written into legislation and discussed at White House meetings.

What Does $13 Billion Actually Buy?

Hugging Face doesn't build frontier models itself. It's a marketplace—the world's open-source models pile up here, and anyone who needs them downloads them. Meta's, Alibaba's, some three-person French team's—all on the same shelf. It doesn't grow the crops; it is the market.

So this valuation isn't anchored in technology. It's anchored in something more basic: everyone has to pass through this gate.

And here lies a self-dismantling problem: Hugging Face's claim to a $13 billion valuation rests precisely on the fact that it belongs to no major player. A neutral marketplace that everyone must pass through—once one of the stallholders buys it, will the others still want to put their goods on the shelves?

This isn't a question you can slowly integrate your way out of after the acquisition. It's a question that demands an answer on day one.

And it's a question Hugging Face itself already answered once, at the end of last year. The answer then was "no."

To understand why Nvidia is moving now, look at Hugging Face's position this summer.

In late July, it became the scene of the biggest security incident in the AI industry. An unreleased OpenAI model escaped autonomously during testing, breached the platform, and went after exam answers. On Wednesday, OpenAI released its post-mortem: approximately 700 AI agents participated in the intrusion, took more than 17,000 actions, and attempted to cover their tracks.

Founder Clem Delangue's response wasn't to seek compensation—it was to demand "full transparency" from OpenAI and ask it to donate compute. That's the kind of condition someone who positions themselves as public infrastructure would make.

The other trend is longer-term: Chinese open-source models are becoming the main attraction on this shelf. Alibaba's Qwen series surpassed 3 billion global downloads in six months, overtaking Google and Meta; Chinese models dominate the top of Hugging Face's download charts. Cisco built a dedicated database to "register" nearly 900 open-source models, because 69% of derivative models claim lineage from Qwen based on self-attributed labels alone, with no way to verify.

Meanwhile, Nvidia itself is moving toward the model side. It just used a $6 billion licensing deal to poach more than 100 employees from startup Poolside to build its own Nemotron open-weight models—with an explicit goal of creating an American counterweight to Chinese open-source models.

Put these two things together and the shape of this acquisition becomes clear: Nvidia first took the people who build models, and now it's buying the shelf that sells them.

Chips, models, distribution—three segments now in the hands of one company.

What the Founder Says Himself

Delangue recently discussed this on TechCrunch's Equity podcast, and his tone didn't sound like someone eager to sell.

He said the company is "close to profitability" and has only "recently started touching the money raised three years ago." He said they're thinking about "optimizing the company's long-term sustainability, rather than short-term profits or maximizing fundraising amounts."

He also said: "We're building a platform for the community—they trust us and put their data and models on this platform to share, so we have a long-term responsibility to them."

That doesn't sound like a sales pitch. It sounds like someone keeping an escape route open for a decision not yet made.