The Information reported on August 27 that Nvidia has agreed to acquire Hugging Face for $12.9 billion, with CNBC, Fortune and TechCrunch following up shortly after. All three outlets included the same caveat: the agreement hasn't been signed yet, and the deal could still fall apart. Nvidia and Hugging Face did not respond to requests for comment.
Three days earlier, the company was still just "testing the waters" — it had asked banks to gauge buyer interest, with market chatter putting the valuation around $13 billion. The speed with which a buyer's name landed suggests the list of interested parties was never very long to begin with.
What $12.9 billion buys
Hugging Face doesn't have a flagship model of its own. It's a repository: it hosts open-source model weights, benchmark leaderboards and training datasets, and it's where developers go to find, test and download models. Annual revenue is around $150 million — up from roughly $100 million just two months ago — and the company is close to break-even.
Run $12.9 billion against $150 million and the price-to-sales ratio comes out to roughly 86x. That multiple is absurd for a software company, but this money clearly wasn't spent chasing cash flow. Every open-source model downloaded from Hugging Face needs a GPU to run, and at this point the default deployment path for the vast majority of open models leads straight to Nvidia's chips. Tie the point of model selection to the hardware, and the math starts to make sense.
The valuation trajectory is worth laying out. In the 2023 round, Hugging Face raised $235 million at a $4.5 billion post-money valuation, led by Salesforce Ventures, with GV, IBM Ventures and Nvidia itself also on the cap table. In late 2025, Nvidia proposed investing $500 million at a $7 billion valuation; Hugging Face turned it down, citing concern that a single investor would gain too much influence. Less than a year later, the same buyer has nearly doubled that valuation and switched from a minority stake to a full buyout.
Who Nvidia is guarding against
The obvious rival is custom silicon. OpenAI, Google, Amazon and Anthropic are all pushing their own accelerators, and if inference workloads at the top labs gradually migrate away, Nvidia's share on the closed-model side will keep eroding. The open-source ecosystem is territory it can still hold: as long as developers keep pulling models from Hugging Face and keep defaulting to CUDA, chip demand stays intact. Nvidia has already poured billions into training its own open models; the acquisition rounds out that strategy.
There's a cloud angle too. Hugging Face already sells compute-rental services for running models, and Nvidia scaled back its own cloud business about a year ago. If the deal closes, Nvidia effectively gets a ready-made outlet back — a way to resell capacity that customer commitments haven't yet absorbed to the same pool of developers.
The neutrality question nobody has answered
That's also where the trouble lies. Hugging Face became public infrastructure precisely because it didn't belong to any single model maker or cloud provider — everyone's weights could sit on it on equal footing. Once the owner is a hardware giant, details like which deployment method gets recommended on a model page, how leaderboards are ranked, and whose chip documentation gets the most attention all become choices with a built-in tilt.
One point of contrast: Hugging Face CEO Clem Delangue and Nvidia's Jensen Huang recently co-signed an open letter arguing that governments should support, not restrict, open-source models. The two do stand on the same side on open-source policy — but a shared stance and shared ownership are two different things. The trust that let the community hand over its data and weights was built on the former kind of relationship.
There's no public antitrust statement yet. Given Nvidia's current position in AI hardware, whether this deal clears review is itself a variable worth watching over the next few months.
Sources: The Information, TechCrunch, CocoLoop, Fortune, CNBC; the deal price, revenue figures and past funding valuations were cross-checked across multiple reports, and the price-to-sales multiple is a rough calculation based on publicly reported revenue.