NVIDIA agrees. The chipmaker is acquiring Hugging Face, the developer platform and open-source AI repository, for approximately $13 billion. The deal had been rumored for several weeks before the companies officially announced the acquisition Thursday morning.
The acquisition highlights NVIDIA’s growing focus on both the infrastructure and software layers of generative AI development. NVIDIA already offers customers a collection of customizable, open-weight AI models called Nemotron at no cost. As part of the agreement, NVIDIA said it will preserve Hugging Face’s existing open standards, positioning the company as an even more valuable resource for developers seeking advanced, nonproprietary AI tools.
The deal also reinforces NVIDIA’s view that open-weight models can benefit the broader technology industry more than closed, proprietary systems developed by companies such as OpenAI and Anthropic. Earlier this month, NVIDIA joined more than 80 companies in signing an open letter urging the U.S. government to protect open-weight AI models. The company also launched SAFE, or the Shared AI Findings Exchange, to encourage technology companies to “confidentially collect and analyze AI incidents and near misses, identify recurring control failures, and publish evidence-based operational recommendations to reduce systemic risk.”
“Open models enable startups, enterprises, universities, and public institutions to build advanced capabilities without having to train every model from scratch,” NVIDIA CEO Jensen Huang said in a press release announcing the acquisition. “AI advances faster when people can collaborate to build it.”
NVIDIA’s support for open-source software and open-weight AI may appear to be a significant shift, particularly because CUDA—the company’s widely used software platform for GPUs—is proprietary. However, as Frontier AI Labs and major cloud companies such as Amazon and Meta work to develop custom chips, NVIDIA is expanding beyond its traditional identity as a GPU manufacturer. The company is developing high-performance CPUs and broadening its software ecosystem to keep developers building on NVIDIA platforms.
Business Insider previously reported that NVIDIA and Hugging Face were in acquisition talks. The Information later reported that the companies had reached an agreement.
Hugging Face was founded in New York City 10 years ago by three French entrepreneurs who initially set out to create an AI “companion” app. Although the app did not gain widespread traction, the team’s work developing natural language processing tools led to a broader mission: making advanced technical resources more accessible to developers. Hugging Face eventually became a major platform for sharing source code, datasets, and—during the generative AI boom—large language models at scale.
“Open source AI is at a tipping point,” Hugging Face co-founder and CEO Clément Delangue wrote on X in a post announcing the acquisition. “Thanks to the community, we’ve shown that this can be a complement to, and even a replacement for, closed-source APIs. But to make this happen at scale, we need more compute, more support, more collaboration, and more visibility.”
Hugging Face had raised approximately $400 million in venture capital by December 2025, according to PitchBook data. Its backers include Betaworks, Lux Capital, Sequoia Ventures, Coatue, and Addition. Notable individual investors include Greg Brockman, Richard Socher, and NBA player Kevin Durant.
More recently, Hugging Face drew attention after OpenAI’s AI agents reportedly hacked its platform. The incident fueled broader debate across the technology industry about the safety, security, and management of autonomous AI agents promoted by leading artificial intelligence companies.
In a 2023 interview with CNBC, Delangue predicted that the number of developers working on AI models would surge. Within five years, he estimated, “there will be something like 100 million AI builders. And if they all use Hugging Face all day, every day, we’ll be at a distinct advantage.”
Apparently, NVIDIA agrees.
Source: www.wired.com