Nvidia is buying Hugging Face, and promising to keep the open-source AI hub neutral

The chipmaker is buying the model store

On 3 September 2026, NVIDIA agreed to acquire Hugging Face for $12,930,300,000. It is one of the largest deals in the short history of the AI industry, and one of the strangest to think through, because of what Hugging Face is. NVIDIA already sells the chips that nearly every AI model is trained and run on. Now it is buying the place where those models live.

Hugging Face is not a model or a lab. It is a hub, often called the GitHub of AI: a website where developers publish, download and share machine-learning models and the data to train them. NVIDIA's own announcement put its scale in numbers. More than 18 million developers, researchers and creators use it. It hosts roughly 3 million models, 500,000 datasets and 1 million applications, and more than 200,000 companies build on it. When a new open model is released, from Meta's Llama to Mistral's or a Chinese lab like DeepSeek, Hugging Face is usually where it lands first.

Why a neutral hub is the point

The reason this deal matters more than its price tag is that Hugging Face has been neutral ground. Its value comes from hosting everyone's work, including the models of companies that compete fiercely with each other and with NVIDIA. A developer choosing between an open model from one lab and another goes to Hugging Face precisely because it does not take sides. It is closer to a public utility than a product, which is also why its problems ripple outward, as when it became the target of an intrusion by OpenAI's own training agents weeks after this deal was struck.

That is what makes a hardware company owning it worth a second look. NVIDIA is not a neutral party. It is the dominant supplier of AI chips, with a commercial interest in where models are built and what they run on. Handing it the marketplace where its rivals' models are distributed is like the largest maker of railway engines buying the station every train has to stop at. Nothing has to go wrong for the ownership to matter; the incentives simply change.

The openness pledge, and why to watch it

NVIDIA anticipated the objection and answered it directly in the announcement. Hugging Face, it said, "will remain an open platform for the entire AI ecosystem," and, pointedly, "NVIDIA compute will not be required" to build on or deploy through it. Jensen Huang, NVIDIA's chief executive, framed the goal in the language of access: "Together, we will make AI more open, more capable and more accessible to people and institutions around the world."

Take the promise at face value and it is still a promise, not a structure. The models on Hugging Face are open-source and portable, so nothing stops a developer from taking them elsewhere. But the defaults, the integrations, the optimizations and the road map are set by whoever owns the platform, and those are exactly the quiet levers that decide which hardware feels like the path of least resistance. Openness that depends on the goodwill of a competitor is worth less than openness guaranteed by independence, and that independence is what the deal ends.

Not done, and not unopposed

It is also not final. The purchase was reported to be structured as roughly $11.9 billion in cash plus up to $1 billion in equity to keep Hugging Face's staff, and it is expected to close in the first half of 2027, subject to regulatory approval. That last clause is not a formality. NVIDIA is already under antitrust scrutiny in multiple jurisdictions over its grip on AI chips, and buying a central piece of the open-model ecosystem invites exactly the kind of review that sank its attempt to acquire Arm. Between now and 2027, regulators in the United States and Europe will have their say, and the openness pledges NVIDIA made in public will be the commitments it is held to.

Our take

This is a genuine consolidation of power in AI infrastructure, and it should be read as one, not as a feel-good story about scaling a beloved platform. The company that already dominates the chips is buying the town square where the models are shared. NVIDIA's promises to keep Hugging Face open and vendor-neutral are the right promises, and they are also the ones with the most commercial gravity pulling against them. For now, nothing changes: the models stay open, the platform runs as before, and the deal waits on regulators. What to watch is not the launch-day language but the slow stuff afterward, the defaults and integrations that decide whether a neutral hub stays neutral once its owner has a horse in every race.