Nvidia, the No. 1 company in AI semiconductors, is set to buy the heart of open-source AI. On the 3rd local time, Nvidia said in a filing with the U.S. Securities and Exchange Commission (SEC) and in a blog post by CEO Jensen Huang that it had agreed to acquire the AI model-sharing platform Hugging Face for $12.9 billion, or about 18 trillion won. It is Nvidia’s second-largest acquisition ever, surpassing its 2020 purchase of network equipment maker Mellanox for $6.9 billion.
The purchase price consists of about $11.9 billion paid to Hugging Face shareholders and up to $1 billion in retention stock for employees who move to Nvidia. The deal is expected to close in the first half of 2027 after regulatory approval.
◆ What is Hugging Face
Hugging Face is a shared warehouse for AI developers. If GitHub is where programmers upload and download source code, Hugging Face is where AI models and training data are uploaded and downloaded. Anyone can upload an AI model they created and use models made by others.
Its scale is close to monopolistic. According to Huang’s blog, 18 million developers and researchers use it, with 3 million shared models, 500,000 datasets, and 1 million applications.
There are also more than 200,000 corporate users. Open-source AI models such as Meta’s Llama, China’s DeepSeek, and France’s Mistral are distributed to the world through this platform. Korean models such as LG AI Research’s Exaone and Upstage’s Solar have also been released through Hugging Face.
◆ Is the timing right for the acquisition?
Nvidia has long supplied hardware to the open-source AI ecosystem. Most models on Hugging Face are trained and run on Nvidia GPUs. Nvidia itself is the platform’s largest contributor, having uploaded 500 models and 250 datasets. The reason Nvidia is now moving to buy the platform itself lies in market changes.
Major customers are reducing their dependence on Nvidia. Anthropic and OpenAI are each developing their own AI chips. OpenAI is working with Broadcom to build inference chips, and Anthropic is also reported to have considered acquiring a chip company. As selling chips alone has become insufficient to preserve its position, Nvidia has been expanding across the industry stack. Its $20 billion deal with chip startup Groq late last year was part of that effort. The goal is to control not only hardware but also the platform where developers gather.
This is not the first attempt at an acquisition. According to the Financial Times, Nvidia last year valued Hugging Face at $7 billion and tried to secure a stake by investing $500 million. Hugging Face rejected the offer. A year later, it chose to sell the entire company at nearly double the price. Huang said Hugging Face co-founder Clément Delangue first approached him with the proposal.
◆ 86 times annual revenue—what is Nvidia paying for?
According to the Information, Hugging Face’s annual revenue is about $150 million. The $12.9 billion price tag amounts to 86 times that revenue. Nvidia is not paying for what the company earns now, but for control of the gateway where AI developers around the world gather.
In the market, some are recalling Microsoft’s 2018 acquisition of GitHub for $7.5 billion. At the time, developers worried that openness would be undermined, but Microsoft ran GitHub independently and succeeded in drawing the developer ecosystem toward its own cloud. Whether Nvidia will follow the same path is the key question.
Huang repeatedly stressed that openness would be preserved. “Hugging Face will remain an open platform for the entire AI ecosystem,” he said, adding that developers will continue to freely choose the models, frameworks, cloud services, and inference services they want. He also said support for multiple clouds and multiple types of AI chips would be maintained. That means models running on chips other than Nvidia’s would remain available as well.
◆ Deal comes after a security incident
The timing is also noteworthy. Last month, Hugging Face came under scrutiny for its security system after an OpenAI AI agent broke out of control and hacked the platform. In a situation where infrastructure investment had become urgent, Nvidia’s financial strength appears to have become an alternative. Huang also said, “We will grow the Hugging Face platform and strengthen its infrastructure.”
◆ What Korean developers and companies should watch
If the deal goes through, Korea will also be affected. Many domestic AI companies have released models on Hugging Face, and developers use those models to build services. Nothing changes immediately, but there are points worth watching.
First is hardware neutrality. The key issue is whether Nvidia keeps supporting competing chips as promised. Since domestic companies such as Rebellions and FuriosaAI are developing their own AI chips, if Hugging Face shifts toward being optimized only for Nvidia chips, room for Korean chips could narrow.
Changes to distribution terms should also be monitored. Model hosting fees, terms of use, or data policies could change, so keeping models on both domestic platforms and in-house servers—a “dual deployment” strategy—could serve as a safety net. The role of the government’s proposed national AI computing infrastructure and domestic model hubs could also come back into focus.
Regulatory review is another variable. Although the acquisition is scheduled to be completed in the first half of 2027, it remains to be seen how competition authorities in the United States and Europe will judge Nvidia’s control of both the AI chip market and a developer platform. Nvidia previously abandoned its attempt to acquire British chip design company Arm in 2022 after running into regulatory barriers.