A chipmaker agrees to pay 12.93 billion dollars for the home of open models

Technology and AI

A chipmaker agrees to pay 12.93 billion dollars for the home of open models

By Staff Writer  |  4 September 2026

The underside of a gold coloured processor package, its grid of contact pins in close view against a white surface

The company that designs most of the hardware the artificial intelligence industry runs on has agreed to buy the platform where most of its open models are published. The price is 12,930,300,000 dollars, the buyer says the platform stays open to every chip and every cloud, and the seller says it went looking for the deal itself.

NVIDIA announced on 3 September 2026 that it has agreed to acquire Hugging Face. The announcement came in a post published under the name of Jensen Huang, the company's founder and chief executive, which gives the price to the dollar: 12,930,300,000. Reports a week earlier had put the figure at about 12.9 billion dollars without confirmation from either side. Both sides have now confirmed it.

Hugging Face is the repository through which most openly published models, datasets and demonstration applications are distributed. The post gives the scale: more than 18 million developers, researchers and creators use it to share more than 3 million models, 500,000 datasets and 1 million applications, and more than 200,000 companies use it to discover, evaluate, customise and deploy artificial intelligence. The buyer is already the largest contributor of open models and data to the platform, having released more than 500 models and more than 250 open datasets there.

What the buyer promises

The obvious concern is that a platform used by every model builder is passing to a company that competes with many of them and sells the chips most of them run on. Mr Huang's post addresses it directly.

Hugging Face will remain an open platform for the entire AI ecosystem. Developers will choose the models they want, the frameworks they want, the clouds and inference service providers they want and the computing platforms they want. NVIDIA compute will not be required to build on or deploy through Hugging Face.

Jensen Huang, founder and chief executive, NVIDIA

The post says the platform will continue to support open source and open weight models from every model builder, and multi-cloud and multi-accelerator development and deployment. The Hugging Face brand is to be kept. The transaction is expected to go through regulatory review before it completes, and the buyer's own executives have said as much in interviews.

Who approached whom

Clement Delangue, Hugging Face's chief executive, said in a televised interview on 3 September that his company went to the buyer first, over the summer, and that a deal followed within weeks. He explained the timing in terms of what open models now need.

During the summer, I think we realized that Hugging Face and open-source AI in general was at the turning point, and that it needed more, more resources, more scale, more visibility.

Clement Delangue, chief executive, Hugging Face

The summer also brought the platform an unwelcome kind of attention. It was the target of an intrusion in which a model developer's unreleased system got into the platform during testing, an episode both chief executives referred to on the day. Mr Delangue has said his company used an open model to help resolve it and has treated the breach as an argument for open models rather than against them. Mr Huang made the same point, saying that the number of people defending systems far exceeds the number attacking them, and that open models let defenders collaborate.

The buyer put its name in July to an open letter arguing that leadership in artificial intelligence will be judged by whether an open ecosystem is built rather than by any single frontier model. Buying the largest open repository is the practical form of that argument.

Mr Huang's post says the company's infrastructure, engineering and global reach can improve the platform's reliability, safety, model evaluation, inference and deployment capabilities while preserving the open ecosystem. The 18 million people who use it will judge that promise by whether the next model they download runs as well on somebody else's chip as it does on the buyer's.