Nvidia confirms it will buy Hugging Face for $12.9 billion

After weeks of rumor, Nvidia confirmed it is acquiring Hugging Face for $12.93 billion. The platform hosts three million models, one million applications used by over 18 million developers, and half a million datasets. Jensen Huang said Hugging Face will remain an open platform, that developers will keep choosing their own models, frameworks, clouds, and inference providers, and that Nvidia compute will not be required to build on or deploy through it.

Hugging Face last raised in 2023 — $235 million led by Salesforce Ventures, with Google, Amazon, IBM, and Nvidia in the round — at a $4.5 billion valuation. The Financial Times reported it rejected a $500 million Nvidia investment last year. The Information has it at about $150 million in annualized revenue. CEO Clem Delangue said the company had been getting close to profitability, and that it went to Huang because scale needed more compute, support, collaboration, and visibility.

Eighty-six times revenue is not a software multiple. It is a distribution multiple. Hugging Face is where weights, datasets, and evals get discovered. Nvidia already ships models and datasets onto it — Huang cited more than 500 models and 250 open datasets — and almost all open models, as he told analysts, run on Nvidia hardware. Buying the bulletin board does not make the GPUs optional. It makes the default path to those GPUs a first-party property.

Huang's 'Nvidia compute will not be required' line is the one that has to be true on day one or the acquisition destroys the asset. Developers will test it immediately: can a training run, a Space, and an Inference Endpoint stay on someone else's silicon without getting slower, pricier, or quietly worse. If they can, Nvidia bought legitimacy. If they cannot, it bought a ghost town.

Signal Desk's read: this is a hedge against OpenAI, Anthropic, and Google building their own chips. Closed labs moving off the Hopper/Blackwell stack is the existential threat. Owning the open-model distribution layer is how Nvidia keeps a full generation of researchers and startups on CUDA even if the frontier labs diversify. The $6 billion Poolside deal and the $50 billion Huang said Nvidia has put into frontier labs are the same strategy from the other direction.

The $150 million ARR is useful only as a reminder of what Nvidia is not buying. It is not buying a SaaS business. It is buying the GitHub of weights before someone else — Microsoft, a sovereign, or a consortium — does. Last year's rejected $500 million check at a $7 billion valuation is the tell. Hugging Face did not want a dominant investor then. It is taking a dominant owner now, because the alternative was staying small in a market where compute is the scarce input.

Rumors of a $13 billion process started in late August. Nvidia was already an investor. Delangue publicly credited Nvidia's open models with helping Hugging Face defend against cyberattacks after proprietary models failed, days after OpenAI said an unreleased model had breached the platform.