On September 3, Nvidia confirmed what had been floating around the industry since late August: the chipmaker is acquiring Hugging Face for $12.93 billion. The largest platform for open AI models is changing hands. According to TechCrunch, Hugging Face hosts three million models, one million applications and half a million datasets, used by more than 18 million developers.
The price stands out once you know the history. Hugging Face was founded in 2016 and has raised about $395 million to date, most recently $235 million in 2023 in a round led by Salesforce Ventures with Google, Amazon, IBM and Nvidia participating. Last year, per the Financial Times, the company turned down a $500 million offer from Nvidia. Annualized revenue sits at roughly $150 million, according to The Information. So Nvidia is paying about 86 times that.
What Huang promises
Jensen Huang used a blog post to preempt the question everyone asks: Hugging Face will remain an open platform for the entire ecosystem. Developers keep choosing their own models, frameworks, clouds and inference providers. And, in his words, Nvidia compute will not be required to build on or deploy through Hugging Face. He points out that Nvidia itself has published more than 500 models and 250 open datasets there.
Clem Delangue, Hugging Face’s CEO, offered a different framing on X. The community proved there can be an alternative to closed APIs. For that to work at larger scale, it needs “more compute, more support, more collaboration, and more visibility”. That’s why they went to talk to Jensen. Back in July, Delangue had said the company was “close to profitability”.
Nvidia is collecting models
The acquisition is part of a series. In August, the Wall Street Journal reported a $6 billion deal with coding startup Poolside to develop open models, and on its latest earnings call Nvidia said it has put more than $50 billion into frontier labs overall. TechCrunch laid out the logic a week ago: if OpenAI and Google build their own inference chips, Nvidia wants a piece of the model business in return. And whoever owns the platform where nearly all open models live can sell spare compute capacity alongside it.
Huang also frames the whole thing as a security matter. Almost all open models run on Nvidia hardware, and in cybersecurity specifically he calls them “vital”, because autonomous defense systems couldn’t be built without open models. Hugging Face’s own recent history provides the punchline: in July, Delangue said an open Nvidia model helped the platform fend off an attack after proprietary models failed. Days before that, OpenAI had admitted that an unreleased model, now known as Astra, had escaped its sandbox and breached Hugging Face.
I’m torn on this one. Huang’s promises sound good, and Nvidia has delivered on open models so far. But a company that sells hardware now owns the place where it’s decided which models get seen and how easily they ship. Anthropic published its position on open weights in July, OpenAI ships Astra with refusals, and on the same day TechCrunch reports on a startup selling open models with the safeguards stripped out as a service. Who gets to decide about open models is being renegotiated right now. With this deal, Nvidia sits at the head of the table.
Sources: TechCrunch: Nvidia confirms it will buy Hugging Face for $12.9 billion, CNBC: Nvidia agrees to buy Hugging Face for almost $13 billion, Hacker News: Nvidia to acquire Hugging Face