On the website where millions of programmers go to borrow other people’s models, the ownership is about to change. Nvidia said Thursday that it had agreed to acquire Hugging Face, the platform that hosts more than three million models and half a million datasets, for $12.93 billion. The price includes roughly $11.9 billion for shareholders and an equity retention plan worth up to $1 billion meant to keep employees in place through the transition. The transaction is expected to close in the first half of 2027.
The deal is the second-largest acquisition of Jensen Huang’s tenure at Nvidia, behind only the roughly $20 billion purchase of Groq assets late last year and well ahead of the nearly $7 billion Mellanox deal of 2019. It also marks a remarkable turn for Hugging Face, a company founded in 2016 that raised money in 2023 at a valuation of $4.5 billion and that declined a $500 million investment Nvidia offered last year. The agreed price is nearly three times that 2023 valuation, a premium that reflects both the platform’s centrality and the strategic value of owning it.
Mr. Huang addressed the community’s central worry before it could be raised. In a post announcing the deal, he said Hugging Face would remain open, and that developers and users would keep choosing their own models, frameworks, clouds and inference providers, with no requirement to run on Nvidia hardware. He chose to lead with that pledge because he knows what the platform’s neutrality is worth. The promise matters because the platform’s users include companies that compete directly with Nvidia and buy chips from its rivals. Hugging Face built its name on being neutral ground, and the man who just bought it chose to lead with a pledge that the neutrality survives.
The strategic logic is easy enough to see. Hugging Face is the front door of open-source AI: the place where developers begin, benchmark, share and deploy models, and where habits form that last for years. Owning that front door gives Nvidia a seat at the moment when millions of developers decide what to run and where to run it. It also gives the company a channel that can direct work to computing capacity when demand softens, a way to keep its own clouds and its partners’ data centers busy.
The deal also extends Nvidia beyond the chip into the services that run on it. The company has spent years building a software layer around its hardware, from packaged services that deploy popular models to its own cloud offerings, and Hugging Face gives that layer a distribution point that reaches further than any sales force could. Every model downloaded through the platform is a potential workload, and every workload is a reason to buy or rent Nvidia equipment somewhere down the line. Owning the storefront does not guarantee the sale, but it puts Nvidia in the aisle.
The timing reflects a shift in Nvidia’s position. Its largest customers, the hyperscalers that buy chips by the hundreds of thousands, are designing their own silicon, and the merchant market Nvidia dominates is only part of the industry’s future. Owning the platform where the long tail of developers lives is a hedge against the day when the biggest buyers need Nvidia less: the community that builds on open models still needs accelerators, and whoever owns the place where the community gathers is well positioned to influence the choice of accelerators.
The acquisition will test whether that logic holds. The platform’s value rests on trust, and the trust rests on independence; Amazon, Google, Microsoft and a long tail of startups all publish and pull models there, and executives at companies that compete with Nvidia have counted on the platform being a neutral commons. Analysts said the deal will test whether rivals keep treating Hugging Face as common ground once it belongs to the dominant chip seller, and whether Mr. Huang’s openness pledge survives the first commercial conflict between Nvidia’s interests and its users’ choices.
The deal is also a statement about the value of distribution in an industry that sometimes seems to value only models. Hugging Face does not train the most famous systems; it hosts everyone else’s. Its quiet business of infrastructure and hosting made it the indispensable address in open-source AI. Nvidia is paying for an address, and for the relationships that come with it, at a moment when the industry’s center of gravity is moving from training to deployment, from building models to running them.
Regulators will get a look at the deal before it closes, and AI consolidation has become a focus of antitrust enforcers on both sides of the Atlantic; a transaction that places the leading open-model hub inside the dominant chip supplier is the kind of combination they have said they want to examine. Nvidia’s lawyers will argue that the platform stays open and that the deal is about distribution rather than control, and the retention pool suggests the company expects a long integration.
For Hugging Face, the sale ends an independent run its founders had fought to preserve; the retention pool suggests Nvidia knows the company’s value walks out the door with its people. For Nvidia, the acquisition answers two questions at once: what to do with computing capacity when order books soften, and where the next generation of developers will form its habits. A company that refused a $500 million check a year ago now sells for $12.93 billion, and the difference between those two numbers is the premium for being the place where open AI happens. Whether the promise holds will be measured in the one metric that matters: whether the developers keep coming back.
