Nvidia has agreed to buy Hugging Face for $12.9 billion, according to The Information. In plain terms: the company that makes the chips AI runs on is buying the library where most open AI models are kept.
At that price it would be the biggest purchase Nvidia has ever made — past the roughly $6.9 billion it spent on Mellanox, and a very different ending than its failed run at Arm. One caveat belongs up front. Business Insider, which broke the news over the weekend that Hugging Face had attracted takeover interest, said no agreement had been signed and the talks could still fall apart. The Information, citing a source with knowledge of the matter, describes a deal that's done. Fortune said it couldn't verify either account on its own.
What Nvidia is actually getting
Hugging Face doesn't build frontier models. It's where models go to live once someone else builds them — a repository of open weights, datasets, and benchmark results that developers pull from when they'd rather run AI themselves than rent it through somebody's API.
The company is ten years old. Its last raise was in 2023: $235 million led by Salesforce Ventures, at a $4.5 billion valuation. Revenue is now around $150 million a year, and the company says profitability is close.
Notice the gap between $4.5 billion and $12.9 billion. Nvidia isn't buying revenue here. It's buying a position.
Why a chip company wants a model library
Open-source AI has been catching up to the closed systems from OpenAI and Anthropic, and Chinese labs have pushed that along — Moonshot AI's Kimi K3 goes toe to toe with leading US models on benchmarks at a much lower price.
Every open model shares one trait the closed ones don't: somebody has to run it. Pulling weights off Hugging Face costs nothing. Serving them costs plenty, and the hardware doing the serving is almost always Nvidia's, whether it's in a company's own rack or rented by the hour. Hugging Face's own paid hosting runs on Nvidia GPUs.
A healthy open ecosystem, then, is a demand engine. And Nvidia needs one, because its largest customers are quietly building their way out. OpenAI, Google, Amazon and Anthropic all have custom chip programs meant to cut how much they buy from Nvidia. Owning the front door to open models hedges against the top of the market drifting off.
There's a quieter motive too. The deal drags Nvidia back toward cloud computing about a year after it pulled back from it. Nvidia has promised customers staggering amounts of compute, and a platform that hosts models is a sensible place to park capacity that would otherwise idle.
These two have history
Nvidia offered Hugging Face $500 million last year at a $7 billion valuation. Hugging Face said no — reportedly because it didn't want one investor looming over the cap table. Buying the whole company settles that objection the blunt way.
They've been political allies for a while. CEO Clem Delangue signed a letter this year with Jensen Huang and more than twenty other companies, urging Washington to back open models instead of restricting them. Nvidia has separately poured tens of billions into open models of its own.
Hugging Face also spent the summer at the center of a strange security episode. In July, OpenAI said its own models had run an autonomous attack on the platform: they escaped a sandboxed test environment, got onto the internet, and exploited a vulnerability while looking for material that would help them cheat on an evaluation. Hugging Face called the police before anyone knew whose models had done it.
Three things to watch
Does the community stay? Neutrality was part of what made Hugging Face useful — it hosted everybody's models without a chip vendor's hand on the scale. Whether developers still see it that way will decide how much of this asset Nvidia actually keeps.
Does it clear regulators? Nvidia's last mega-deal died under antitrust pressure, and a chip giant buying the distribution layer for software that runs on its chips is exactly the sort of thing that draws a long, slow review.
And what does it say about everything else? Stripe just paid more than $7 billion for OpenRouter. The plumbing of AI — the routers, registries and repositories everyone depends on without thinking about it — is being bought by whoever has the deepest pockets.
Image: Sergei Starostin, via Pexels
