Nvidia agrees to buy Hugging Face for $12.9B

Nvidia has agreed to acquire Hugging Face, the default distribution hub for open-weight models, for $12.93 billion. Closing is expected in the first half of 2027, pending regulatory approval — a year in which the neutrality of the open-model ecosystem's central hub is now an open antitrust question.

Key facts

• Nvidia will pay $12,930,300,000 for Hugging Face, per CEO Jensen Huang's announcement on September 3, 2026.
• Hugging Face hosts more than 3 million models, 500,000 datasets, and 1 million applications, used by over 18 million developers and 200,000 companies.
• The deal is expected to close in H1 2027 subject to regulatory approval, per CNBC's reporting on the agreement.
• Huang says Hugging Face will remain an open, multi-cloud, multi-accelerator platform and that "NVIDIA compute will not be required to build on or deploy through Hugging Face."
• Nvidia describes itself as the largest contributor of open models and data to Hugging Face, with more than 500 models and 250 open datasets published.
• The Register called the deal one of the most antitrust-worthy AI mergers to date, arguing Hugging Face is too important to the ecosystem to sit inside the dominant GPU vendor.

What happened

Nvidia announced on September 3 that it has agreed to acquire Hugging Face for $12.93 billion, confirming reporting by The Information and Reuters from late August. Huang's announcement, published under his own byline on the NVIDIA Blog, frames the acquisition as a bet on open weights: he notes he recently co-authored an open letter on the importance of open weights to the AI economy, and commits that Hugging Face "will remain an open platform for the entire AI ecosystem."

The platform numbers in the announcement show what is actually changing hands. Hugging Face is not a model lab; it is the distribution and evaluation layer for more than 3 million models and 500,000 datasets, the place where nearly every open-weight release this publication covers — Qwen, DeepSeek, GLM, Kimi, Llama — lands first. The New York Times described it as "a library of open artificial intelligence models."

Two details from the surrounding reporting add texture. Huang told InfoQ he had wanted Hugging Face to remain independent, but other bidders emerged — an unusual public admission that this was partly a defensive move. And Hugging Face CEO Clément Delangue, who rejected a $500 million Nvidia investment roughly a year ago, is now backing the full acquisition, telling press the "planets aligned."

Why it matters

The open-weight ecosystem runs on an assumption so basic it rarely gets stated: the hub where weights are published does not favor any hardware stack, framework, or lab. That neutrality is what makes Hugging Face useful as shared infrastructure — the same reason builders treat it as a given in guides like what is decentralized AI inference and our map of decentralized AI inference networks.

Putting that layer inside Nvidia creates a structural conflict regardless of anyone's intentions. Nvidia sells the GPUs those models run on, competes with other accelerator vendors the hub must serve equally, and builds its own models that compete with every other publisher on the platform. The company's own announcement concedes the point by addressing it directly: the promise that "NVIDIA compute will not be required" only needs saying because the incentive to require it now exists. Commitments made at announcement time are the easy kind; what matters is what the merged entity's terms of service, API rate limits, and promotion algorithms look like in 2028.

For builders, the practical exposure is dependency concentration. If your pipeline pulls weights, datasets, or eval tooling from a single hub, the next 12 months are the window to know your mirrors — model mirrors on other hosts, alternate hubs, and the aggregators we track in our OpenRouter alternatives coverage. This is not a prediction that Hugging Face degrades; it is the standard observation that single points of failure get more expensive to exit the longer you wait.

Background

Nvidia's position in open weights has been building for years. The company says it has released more than 500 models and 250 open datasets on Hugging Face, making it the platform's largest contributor of open models — a fact that cuts both ways. It demonstrates real commitment to the ecosystem, and it also means the largest hardware vendor was already the ecosystem's biggest supplier before it bought the shelf space.

The regulatory backdrop is hostile to deals like this. The Register's coverage argues the acquisition concentrates control over both AI hardware and model distribution in one company, the vertical-integration pattern antitrust authorities in the US, UK, and EU have been circling since the Microsoft–OpenAI and Nvidia–Run:ai reviews. A $12.9 billion price tag on a company whose revenue is a fraction of that figure is itself an argument regulators will read as buying strategic control rather than cash flow.

There is also a recent-history echo worth naming. Microsoft bought GitHub in 2018 amid near-identical fears about a platform vendor owning neutral developer infrastructure; GitHub stayed largely neutral, but the reassurance campaign then looked exactly like the one now. The difference is that code hosting had viable rivals at scale. Open-weight model distribution, today, effectively does not.

What's next

Watch the H1 2027 close. Regulatory review in the US and EU is the gate, and remedy demands — data-access firewalls, non-discrimination conditions, even structural separation of the hub — are all live possibilities. Delangue's role and retention package, the formation of any independence oversight board, and the first post-announcement changes to Hugging Face's terms or promotion policies are the concrete signals. If regulators attach real neutrality conditions, this deal could end up strengthening the open ecosystem's plumbing; if it sails through untouched, builders should treat distribution diversification as unfinished work.