Nvidia has agreed to acquire Hugging Face for $12.9 billion, according to The Information, a deal that would place the AI industry’s de facto neutral model repository, described by TechCrunch as “a kind of GitHub for the AI era”, inside the balance sheet of the company whose GPUs run most of the models hosted there. CNBC, Fortune, and Reuters have all matched the report. A source familiar with the matter told CNBC that “confirm acquisition [by Nvidia] has been part of ongoing and recent talks.” Bloomberg, citing Business Insider, reported that serious conversations have taken place in recent weeks without a signed agreement. Nvidia and Hugging Face didn’t respond to requests for comment, and Fortune notes the talks could still fall apart.
The price implies a remarkable rerating. Hugging Face last raised in 2023 at a $4.5 billion valuation and, per Fortune, generates roughly $150 million a year in revenue while nearing profitability. Last year the company turned down a $500 million Nvidia investment that would’ve valued it at $7 billion. Twelve months later, the number is nearly double.
What makes the deal structurally interesting isn’t the multiple. It’s the sequencing.
In roughly two weeks, three neutral layers of the open-weight stack have been spoken for. Stripe acquired OpenRouter, described by TechCrunch as “the top provider of open-weight models to businesses,” for more than $7 billion. Nvidia struck a $6 billion agreement with Poolside that’ll move most of its employees to the chipmaker. And now Hugging Face. The distribution hub, the routing layer, and a frontier open-weight lab, all folded into strategic acquirers inside a fortnight.
Nvidia’s logic is legible enough. As Fortune notes, downloads from Hugging Face usually run on Nvidia GPUs, and a strong open ecosystem “keeps more of the market tied to Nvidia’s hardware” at precisely the moment OpenAI, Google, Amazon, and Anthropic are building their own inference silicon. OpenAI announced capabilities for its Jalapeño inference chip this week. Nvidia’s countermove, evident in its reported LPU plans for China and the Etched-Jane Street rack deployment, is to own the substrate on which alternatives to its own hardware would otherwise incubate.
The demand-side numbers complicate the framing. Ramp data cited by TechCrunch puts open-weight usage at just 6% of companies; Jellyfish clocks engineer usage at 2%. Fireworks CEO Lin Qiao said the company processes 40 trillion tokens a day, more than Gemini’s or OpenAI’s APIs, and TechCrunch cites cheaper models from Chinese companies like Moonshot, DeepSeek, and Alibaba as drivers of that adoption at the margin. So the layer being consolidated is small in enterprise share but disproportionately load-bearing for the segment of developers and small operators who chose open weights precisely because the intermediaries were neutral.
That neutrality is the thing being priced. Whoever holds the hub sets the defaults.
Sources
- https://techcrunch.com/2026/08/28/open-weight-ai-companies-are-the-valleys-hottest-acquisition-targets/
- https://www.cnbc.com/2026/08/27/nvidia-hugging-face-acquisition.html
- https://www.bloomberg.com/news/articles/2026-08-27/nvidia-discussed-buying-ai-startup-hugging-face-insider-says
- https://fortune.com/2026/08/27/nvidia-hugging-face-billion-dollar-deal-open-source-ai/
- https://money.usnews.com/investing/news/articles/2026-08-26/nvidia-in-talks-to-acquire-hugging-face-in-13-billion-deal-business-insider-reports