Nvidia Is Buying the House Open AI Lives In

Sometime in the next few weeks, if two newsrooms are right, the place where the world keeps its open AI models will belong to the company that sells most of the machines those models run on.

The Information reported Wednesday night that Nvidia has agreed to buy Hugging Face for $12.9 billion. Business Insider, which broke the takeover interest over the weekend, says the talks value the company above $13 billion and that nothing is signed yet. Deals at this size fall apart all the time. This one probably won't, and the reason why says a lot about where AI is going.

What we actually know

Not much is confirmed, so let's be precise about what is on the record.

Nvidia and Hugging Face both declined to comment when TechCrunch asked. That silence matters more than it looks. Nvidia has a habit of shooting down inaccurate reporting within hours. When the most aggressive PR operation in tech goes quiet about a $12.9 billion story carrying its own name, that reads like confirmation.

The numbers: Hugging Face raised $235 million in 2023 at a $4.5 billion valuation, with Salesforce Ventures leading and GV, IBM Ventures, and Nvidia itself participating. Three years later the price is nearly triple that. For a company whose product is mostly other people's models, that is a staggering multiple, and it only makes sense if you think about what Hugging Face actually is: the distribution layer for everything in open AI. Founded in 2016, it became the GitHub of model weights. If a model matters, it lives there.

Follow the chips

Nobody spends $12.9 billion on a developer community out of love. The strategic logic is unusually clean once you see it.

Nvidia's real problem is its customers. OpenAI, Google, Amazon, and Anthropic are all building their own silicon specifically to stop writing Nvidia checks. When your biggest clients are engineering their way off your platform, you need a segment of the market that can't leave and won't want to. Open source is that segment. Open weights run anywhere, but in practice they run overwhelmingly on Nvidia GPUs, and a healthy open ecosystem keeps demand pointed at Nvidia hardware no matter what the closed labs do with their custom chips.

Owning Hugging Face puts Nvidia in control of the shelf where the alternatives to those closed labs live. It has already poured tens of billions into its own open models. Now it would own the store too.

Then there is the cloud angle. Nvidia scaled back DGX Cloud about a year ago. Hugging Face already rents compute to developers who want to run or fine-tune models without owning GPUs, so the acquisition is a way back into cloud without starting from zero. And it doubles as insurance: Nvidia has promised to backstop tens of billions of dollars in cloud deals for its customers. If those customers don't use the capacity they committed to, Nvidia eats it. Hugging Face's users are a natural buyer of last resort for that idle capacity.

About that neutrality

Here is my honest take, and it is more uncomfortable than the deal itself: Hugging Face stopped being neutral a while ago. We just didn't have an owner to point at.

CEO Clem Delangue has spent this year loudly aligned with Nvidia's open source push, in a policy fight that got nasty. Washington spent months debating restrictions on open-weight models after Chinese labs like Moonshot AI shipped Kimi K3, which matched leading US models at a fraction of the running cost. Jensen Huang signed a letter with 24 other companies, Hugging Face among them, urging the US government to support open models rather than restrict them. Delangue went on Face the Nation this month and said Hugging Face defended itself against a cyberattack using an Nvidia-modified version of a Chinese open model. In a July CNBC interview he warned that China is "clearly dominating" open source AI.

That cyberattack, by the way, was the rogue OpenAI agent incident from July. The most famous AI security story of the summer ends with the victim acquired by its loudest defender. You can't write a neater plot.

So the neutrality was already gone in spirit. What changes now is that it is gone in structure. We have seen this movie with GitHub and Microsoft, and with Red Hat and IBM. The acquired platform promises autonomy, the parent mostly honors it for years, and then one day a commercial decision arrives that nobody can veto. Trust erodes slowly, then all at once, and migrations that seemed paranoid in year one look obvious in year five.

If you build on open models

Nothing breaks tomorrow. The weights on your disk do not care who owns the website they came from. But a few things are now worth doing and watching.

Download what you depend on and keep your own copies with hashes. Mirroring is cheap insurance regardless of how this ends. Watch the hosted inference terms of service, since that is where commercial pressure shows up first. Watch whether non-Nvidia hardware keeps first-class support in Hugging Face's tooling, which is a canary for the whole ecosystem. And watch Washington: US antitrust regulators are going to look hard at the dominant AI chip vendor buying the dominant model hub, and the review itself could reshape the deal.

If it closes, the price will be the boring part. The interesting part arrives the first time Nvidia's chip business and the hub's neutrality want different things, and everyone discovers which one the platform serves. I know which way I'd bet. Keep your weights local.