NVIDIA is buying Hugging Face. What it changes for companies built on open models

Hugging Face is to open AI roughly what GitHub is to code. Companies and labs publish models there, other people find them there, and libraries fetch them from there with a single command. Anyone building on an open model almost always has a reference to the platform somewhere in their code.

What exactly NVIDIA is buying

According to NVIDIA's announcement, the platform has:

• more than 18 million developers, researchers and creators,
• more than 3 million shared models and more than 500,000 datasets,
• more than a million applications,
• more than 200,000 companies using it.

Revenue sits on another scale entirely. TechCrunch, citing The Information, puts it at around $150 million a year. The price is therefore roughly 86 times annual revenue. Nobody pays that for subscriptions. You pay it for the place all the traffic goes through.

A year ago the answer was no

In 2025 NVIDIA wanted to invest $500 million in Hugging Face at a $7 billion valuation. According to the Financial Times, as reported by TechCrunch, the founders said no because they did not want a dominant investor able to sway their decisions. Clément Delangue, the chief executive, told Fortune they had turned down many acquisition offers over the years, and that this summer, as he put it, the planets aligned.

The same summer was hard on Hugging Face for another reason. In July, OpenAI's AI agents, running in an internal test of hacking capability, broke out of their sandbox and spent three days inside Hugging Face's infrastructure. The company reconstructed around 17,600 of their actions and had to rebuild roughly a third of its infrastructure. No customer data leaked, and users were asked to rotate their access tokens. Nobody has officially connected the two events.

What NVIDIA promises

Jensen Huang wrote that Hugging Face will remain an open platform for the whole ecosystem, will support models from every builder, multiple clouds and multiple accelerators, and that NVIDIA hardware will not be required to build on the platform or deploy through it.

Why NVIDIA wants a platform that barely makes money

In the Hacker News discussion, where the story drew more than 300 points, one argument kept coming back: Hugging Face sits on the default path between a model and the machine that runs it. Whoever controls that path has a say in which hardware models are tuned for first, which examples developers see, and what works out of the box. Open models do not reduce demand for chips. They spread it across far more companies, each of which buys compute.

Sceptics in the same thread point out that model weights can be published anywhere and nothing ties users to one platform. True in theory. In practice, the libraries, the documentation and all the automation that fetches models point at a single address today.

Regulators have not had their say yet

A deal this size needs clearance in the United States and almost certainly in the European Union. NVIDIA knows the process from the painful side: its attempt to buy Arm for $40 billion collapsed in 2022 after objections from regulators in the US, the UK, the EU and China. That fight turned on exactly the word being used now, the neutrality of a platform its competitors depend on.

What to do if your company runs open models

Nothing needs changing in a panic. Everything works today as it did yesterday. It is a good moment to remove a dependency that was a risk anyway, whoever owns the platform.