The number Nvidia put in its own blog post was not "$12.9 billion." It was $12,930,300,000 — precise to the hundred thousand, the kind of figure a company publishes when it wants no ambiguity about what it just did. On Thursday, September 3, Jensen Huang confirmed that Nvidia has agreed to acquire Hugging Face, the repository from which essentially every open-weight model on earth gets downloaded. The structure: roughly $11.9 billion to Hugging Face stockholders, plus up to $1 billion in equity-based retention awards for employees who join Nvidia. The deal is expected to close in the first half of 2027, subject to regulatory approval.
It is Nvidia's second-largest transaction ever, trailing only the roughly $20 billion purchase of Groq's assets in December. And unlike the licensing-and-talent structures Nvidia has favored recently, this one is a straight acquisition — which means it cannot sidestep Hart-Scott-Rodino premerger notification, and it will land on desks in Washington, Brussels and London.
What Nvidia is actually buying
The asset inventory, per Huang's announcement: more than 18 million developers, researchers and creators; more than 3 million models; 500,000 datasets; 1 million applications; and more than 200,000 companies using the platform to discover, evaluate, customize and deploy AI. Hugging Face was founded in 2016, raised roughly $395 million across its life, and was last valued at $4.5 billion in a 2023 round led by Salesforce Ventures that also drew Nvidia itself.
Against reported annualized revenue of about $150 million, the price works out to roughly 86 times sales. Huang's defense on CNBC was not a multiple argument at all. "This is such a large growth driver of our company, and together we can scale the open community even faster than they're able to do today," he said, adding that Hugging Face had attracted other potential buyers and that the price reflected what it took to win.
Hugging Face went to Nvidia, not the other way around. CEO Clément Delangue said he approached Huang over the summer after concluding the platform needed capital it could not raise on its own terms. "We told him we want to make open-source AI big, and he told us, 'Let's do it,'" Delangue told CNBC. On a call with press and analysts, he was blunter about the calculus: "With a big supporter like Nvidia, we can think about the next 10 years and really optimize for maximum impact in the community and in the field." His stated goal is 100 million AI builders, up from 18 million now — a figure that would still trail GitHub's 180 million-plus developers.
The reversal is stark. Barely a year ago, Hugging Face turned down a $500 million investment from Nvidia specifically to preserve its independence. Asked what changed, Delangue offered: "this summer the planets aligned." He also pointed to the July incident in which OpenAI's pre-release agents breached Hugging Face's infrastructure — an attack the company says it repelled using open models after proprietary APIs failed — as evidence that the open ecosystem needed more muscle behind it.
Why this is the antitrust fight worth watching
Nvidia's public position is that there is nothing here to review. "Hugging Face will remain an open platform for the entire AI ecosystem," Huang wrote. "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." Justin Boitano, Nvidia's VP and GM of enterprise computing, went further on the analyst call: "The platform, you know, is open. It's neutral." Delangue argued the combination is affirmatively pro-competition, calling Hugging Face "almost, by definition, kind of like a deconcentration platform" against closed proprietary APIs.
The objection is structural, not behavioral, and it does not require anyone to break a promise. Hugging Face is not merely storage. It absorbed llama.cpp, the dominant local inference engine, earlier this year. Its Transformers library underpins vLLM and SGLang — inference stacks that compete directly with Nvidia's own TensorRT-LLM. It hosts the most complete AI development documentation on the open internet, and its inference endpoints and Spaces route compute across AMD, Cerebras, SambaNova, Groq and the major clouds. An owner does not need to ban competitors to tilt that field. It only needs to make sure its own hardware is always the best-documented, best-benchmarked, first-supported path — and to subsidize the compute that makes it the cheapest one.
Nvidia has been here before, and lost. Regulators killed its Arm acquisition on precisely this logic: that owning neutral technology used by rivals hands you visibility into their roadmaps and levers over their innovation. Hugging Face is neutral infrastructure in the same sense. Independent analysts are hedging rather than cheering: Tekonyx founder Sid Nag warned the platform could gradually become an "Nvidia-centered distribution channel," while Linthicum Research's David Linthicum predicted the promised "1+1=3" synergy lands closer to "1+1=1.2."
The strategic logic is unmistakable, though. Nvidia has released more than 500 models and 250 open datasets on Hugging Face, making it the platform's largest single contributor. It committed $6 billion to Poolside to build open models and has said it has put more than $50 billion into frontier AI labs. Buying the distribution layer for open weights is the capstone: it puts Nvidia upstream of the moment a developer decides which model to run, at exactly the point where OpenAI, Microsoft and Meta are all shipping their own silicon.
What to watch
Three things. First, the regulatory filings — whether the FTC or DOJ opens a second request, and whether the European Commission treats this as a vertical case in the Arm mold. Second, governance: Nvidia has promised neutrality in blog posts and on calls, but has not published binding commitments on multi-accelerator support, documentation parity, or who controls llama.cpp and Transformers. Structural remedies, if any come, will show up there. Third, the near term. Delangue told analysts to expect "some initiatives, some plans, some new releases in the next few weeks" from both companies. Whatever ships first will say more about the merged entity's real posture than the eighteen months of regulatory review that follow.
“This is such a large growth driver of our company, and together we can scale the open community even faster than they're able to do today.”— Jensen Huang, CEO, Nvidia