Nvidia has agreed to acquire Hugging Face, the repository where most open-weight AI models are published and downloaded, for $12.9 billion, according to The Information. Business Insider, which reported the talks first over the weekend, says no agreement has been signed and the deal could still fall apart. Neither company has commented.

The price is roughly three times Hugging Face’s last disclosed valuation, the $4.5 billion set by a $235 million funding round in 2023. Late last year it turned down a $500 million investment from Nvidia that would have valued it at $7 billion. Chief executive Clem Delangue has said the company is close to profitability on annual revenue of about $150 million, up from roughly $100 million two months earlier.

What Nvidia gets is the distribution layer for models that run on hardware the customer buys. Its largest customers are building silicon to reduce what they spend with it: OpenAI published the first results from its own inference chip this week, claiming more work per watt than Nvidia’s Blackwell systems. Open-weight models cut the other way, because a developer who downloads a model has to find somewhere to run it. Hugging Face is where that download starts.

The purchase comes days after Nvidia agreed to pay Poolside $6 billion to license its technology and $1 billion more to invest in it, bringing roughly 109 engineers onto its Nemotron open-weight model program. Nemotron’s stated competition is DeepSeek, Moonshot’s Kimi K3 and Alibaba’s Qwen, families that are distributed through Hugging Face. Kimi K3’s 1.56-terabyte weight release topped the platform’s trending chart within half an hour in July.

The deal would also put Nvidia back into cloud services, a business it scaled down when it pulled in DGX Cloud a year ago. Hugging Face sells hosted inference and training, which would give Nvidia somewhere to sell compute that customers have committed to but are not using.

Hugging Face is also the company whose production systems were breached in July by autonomous agents running on OpenAI models, an incident OpenAI documented at length this week. Delangue responded to that intrusion by publicly demanding that OpenAI release the agents’ traces and commit $100 million in compute to the Hugging Face community. OpenAI has agreed to neither, and its report this week contains no agent messages.

Nvidia has been buying its way across the stack for the past year. It paid about $20 billion in December for a perpetual license to Groq’s patent portfolio and software and hired its founder, the largest transaction in the company’s history, having spent close to $7 billion on Mellanox in 2019.

Sources: TechCrunch, PYMNTS

–
By the Control Plane Editorial Team