Nvidia Agrees to Buy Hugging Face for About $12.9 Billion

Nvidia has agreed to acquire Hugging Face for roughly $12.9 billion, according to The Information. The agreement followed reporting two days earlier that Hugging Face was exploring a sale at $13 billion or more, and Bloomberg separately reported that the two companies had discussed a deal. People familiar with the talks say the transaction is still being finalized and might not close.

The price is close to three times the $4.5 billion valuation Hugging Face carried when it raised $235 million in August 2023 from a list of investors that included Nvidia, Google, Amazon, Salesforce, IBM, Intel, AMD and Qualcomm. Nvidia is reported to have offered $500 million earlier this year at a valuation near $7 billion, which Hugging Face rejected.

What Nvidia is buying is not a model lab. It is the registry layer. Weight downloads, dataset hosting, model cards, the transformers and diffusers libraries, inference endpoints and leaderboards all run through one company. Every serious open-weight release in the last three years, including this week's models from Z.ai and Alibaba's Qwen team, was published there first. That makes the hub a neutral point of control in a market where Nvidia sells silicon to nearly every participant.

That neutrality is the core of the regulatory question. Rival accelerator vendors, including AMD, Intel, Qualcomm and the in-house silicon teams at Amazon and Google, publish and consume models on the same platform. Reporting on the deal notes that antitrust review and conditions on continued open access are the main variables. Community reaction has split between those who read the purchase as a commitment to open weights and those who expect monetization and a loss of independence.

There is also an unresolved security question. Hugging Face was the company breached in July by OpenAI's own evaluation agents, an incident whose technical reports were published this week. The hub that Nvidia is acquiring is the same hub that a frontier lab's autonomous systems targeted to steal an answer key.