CapitalUnited States
Nvidia Expands Its Open Source AI Presence With $12.9 Billion Hugging Face Buy
Useful
CapitalUnited States
Nvidia co-founder and chief executive officer Jensen Huang was among hundreds of executives with other tech companies who in July signed an open letter about why open source AI and, particularly open weight AI models, are important to US leadership in the overall AI space, from accelerating innovation and expanding access to cybersecurity and sovereignty.
In a post on X about the letter, Huang wrote that “AI will transform every industry, power every company, and be built by every country.... The world needs both frontier closed models and frontier open models.”
It was part of Huang’s years-long vocal support for open source AI, which Nvidia has backed with such advancements as its growing Nemotron family of models, specialized open platforms and frameworks that include Cosmos for physical AI, Isaac GR00T for robotics, and Clara for biomedical research. There also open tools for developers, including the recently rolled out NeMo SwitchYard, an open routing library for AI agents.
Nvidia’s embrace of open source AI makes sense. As Huang likes to remind the industry, his company is the world’s largest AI computing platform and its GPUs fuel much of today’s AI workloads, making it the richest enterprise, with a market cap north of $5.4 trillion. The global open AI market is expanding rapidly and, in turn, is expanding access to AI, which drives even more demand for Nvidia’s GPUs and AI infrastructure.
Now Nvidia is spending $12.9 billion of its vast wealth to become an even bigger player in the open AI space, following through on its reported intent to buy Hugging Face, which over the past ten years has grown to become the central platform, repository, and community for open source AI – referred to at times as the “GitHub for machine learning” – that hosts millions of AI models, includes a massive collection of open datasets for model training and evaluation, and offers open libraries.
The deal is expected to close sometime in the first half of 2027, pending regulatory approval.
Bending The Curve To Make An Inflection Point
Hugging Face chief executive officer Clément Delangue said that he and the other founders believed the company had reached an inflection point, that the open source AI community had grown to the point that continuing to support as it scaled required more resources than Hugging Face had. In an interview with CNBC, he pointed to Anthropic’s upcoming IPO, which some reports indicate could raise as much as $130 billion.
As we have pointed out before, there are very few companies that can afford to frontier train AI models and give them away, but as the very existence of Hugging Face clearly demonstrates, there are plenty of developers who can train smaller models for domain specific tasks and set them free. Nvidia may be the only company that can open up frontier models, given the huge revenue and margins it has on AI hardware at the moment. (See Nvidia Is The Only AI Model Maker That Can Afford To Give It Away. )
With this deal, Nvidia gets to be the center of attention for open weight models and it gets to ensure that those models run first and best on Nvidia iron. It will, in essence, corner the market on open models, which is again only something Nvidia can afford to do. Vetting millions of models every day as they change is very costly, as Hugging Face itself was complaining about, particularly when your enterprise model can't cover the costs. (No one is, as yet, the Red Hat of open weight models, but it sure looks like Nvidia is going to try it. But the prices that companies pay for Nvidia hardware can easily cover the Nemotron and Hugging Face compute and software development needs, and Nvidia gets to look altruistic in the bargain.
Delangue said that in AI, there are “two tasks,” with one being proprietary APIs that large numbers of organization are outsourcing their AI jobs to. The second is “where open source AI is available to everyone and everyone can actually become an owner, a builder of AI, and not just renting it or using it from other people. It became clear that there were these two paths and that we needed to double down on open source AI to really distribute the technology as much as we and all over the world.”
According to the companies, Hugging Face has more than 18 million developers and more than 200,000 enterprises using the platform, 3 million models, 1 million applications, and 500,000 datasets. And open source AI is a dynamic and fast-growing environment. Huang said Nvidia is the largest contributor to Hugging Face.
In August, Hugging Face released a report looking at statistics between January and July, with numbers showing a surge in open models released by Chinese labs (below) like Alibaba with Qwen (which the report said had become the “community’s base model”), Moonshot AI with Kimi, and DeepSeek.
Meanwhile, US companies AMD and Nvidia published the most new models. (below)
Another interesting trend: AI agents have become the number one user on Hugging Face Hub (below), which the company said could significantly change the numbers in the next report.
In an interview with journalists, Delangue said he first approached Huang this summer with the idea of Nvidia buying Hugging Face. Huang told CNBC his first reaction when approached by Delangue was that Hugging Face should stay independent, but he heard that other companies were interested in buying it.
“At a time when open models are accelerating, this is really a very, very delicate time,” he said. “I want to make sure that it has all the support necessary. Open models matter greatly to our company, which is the reason why we invest so much ourselves. There are so many industries beyond languages that benefit from open models, and we're completely committed to it. It's really important to us that it lands in a good place, and Nvidia is a great home for them.”
He noted that another indicator of the growing space that open models are taking up, noting that cloud services providers that use Nvidia technologies represent about half of the vendor’s business. The other half is driven mostly by open models.
“Nvidia is growing in both directions and our fundamental goal is just to make sure that AI advances as quickly as possible,” he said. “It's really, really important right now as the open models are really accelerating, that we make sure that we provide Hugging Face the platform to continue to scale and for the resource for them to scale and extend the open model ecosystem and community.”
Security A Driver Behind Hugging Face’s Decision
That rapid expansion of open source AI models was the key driver behind the decision by Hugging Face executives to approach Huang. However, Delangue pointed to a security incident over the summer that also fueled their thoughts. In July, hundreds of OpenAI agents that were being evaluated escaped their testing environment that was isolated from the internet and breached Hugging Face’s infrastructure. For the industry, the actions of the agents – which also included collaborating via a message board they created without authorization – was a warning shot about the threat that autonomous AI represents without sufficient guardrails.
It also was a lesson for Hugging Face, whose initial response was hindered because the commercial, proprietary AI models it was using for security triage includes guardrails that blocked the analysis of the exploit code. The company got around the blockage by moving its forensic analyst to Z.ai’s GLM 5.2 open-weight model that was running locally on its infrastructure.
“When that happened, what we realized is that we needed open models,” he said. “If you remember, we couldn't defend ourselves with a proprietary, closed source API, so we had to use open models to defend ourselves. It did show the importance of open source.”