Explained: Why Nvidia wants Hugging Face—and what Jensen Huang sees in open AI models

Nvidia is reportedly in advanced talks to acquire Hugging Face for $12.9 billion, with the deal potentially reaching about $14 billion including a $1 billion employee-retention package. No final agreement has been reached, according to the Bloomberg report. The proposed acquisition would give Nvidia control of one of the most important platforms for developers to discover, share and build AI models.

Fortune India explains what this acquisition is and what it means for Nvidia.

What is Hugging Face?

Hugging Face is essentially a global hub for the AI development community which allows developers and researchers to share and use AI models, datasets and applications. The Hugging Face Hub currently hosts more than 2 million models, 1.5 million datasets and 1.5 million AI applications, according to the company.

The company was founded in 2016 and was valued at $4.5 billion in a funding round three years ago. It is backed by Nvidia, Google's parent Alphabet, Amazon. Intel and Salesforce.

The important distinction is that Hugging Face does not primarily make one giant AI model like OpenAI or Anthropic. Instead, it provides the platform on which thousands of developers and organisations can build, share, test and deploy models. Models can be downloaded, modified and run on a user’s own infrastructure, depending on their licence and level of openness.

Why does Nvidia want to buy it?

The answer lies in how Nvidia’s business has expanded. Nvidia remains the dominant supplier of AI accelerators, but it has increasingly built software, cloud and infrastructure products around those chips. Its DGX Cloud Lepton platform, for example, connects developers with Nvidia GPU capacity and integrates Nvidia’s NIM and NeMo software for developing and deploying AI applications.

Hugging Face fits into that strategy because the model developers choose ultimately create demand for computing. Nvidia already has a relationship with the platform: Hugging Face integrated DGX Cloud Lepton into its Training Cluster as a Service in 2025, giving researchers access to scalable Nvidia GPU computing for model training. Nvidia’s deployment tools also support models hosted on Hugging Face.

Nvidia also says it already has more than 650 open models and over 1,000 repositories on Hugging Face.

So, the strategic logic is not simply that Nvidia wants to become a model company. It wants to be present wherever AI developers are building and using models. Owning Hugging Face could give Nvidia a much deeper relationship with that developer ecosystem—and potentially help drive more AI workloads onto Nvidia’s computing infrastructure.

There is another reason. Nvidia’s biggest customers are also developing their own AI chips. The Bloomberg report says Huang is supporting open models partly to prevent AI from being dominated by a handful of large companies that currently account for much of Nvidia’s revenue but are simultaneously pursuing their own chips.

What exactly is Jensen Huang’s idea about open-source models?

Huang’s position is more nuanced than simply saying “open source is better.”

He believes the AI world needs both closed and open models. At Nvidia’s GTC conference in March 2026, he said, “Proprietary versus open is not a thing. It’s proprietary and open.” His argument is that open models are important because they allow startups, researchers, universities, enterprises and countries to build on AI rather than simply consume it through a service controlled by a handful of companies.

Huang has made this argument repeatedly. In a 2025 discussion, he compared open-source models to foundational open technologies such as Linux, Kubernetes and PyTorch, arguing that without open technologies, startups and researchers would have much less ability to advance.

Nvidia has subsequently put that philosophy into practice. It has released open models across areas including reasoning, robotics, physical AI and biomedical AI, and says open models and platforms are intended to broaden access to AI and allow enterprises to integrate and deploy the technology with greater control.

Huang’s argument is also about AI diffusion. If powerful models are available openly, more developers and countries can adapt them for their own needs instead of depending entirely on a handful of proprietary providers. In July 2026, Huang used his first-ever X post to promote an industry letter arguing that open models “strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.”

What does Hugging Face have to do with Nvidia’s bigger AI bet?

This is where the potential acquisition becomes significant. Huang describes AI as a five-layer cake, spanning energy, chips, infrastructure, models and applications. Nvidia began with the chip layer but has steadily expanded into the surrounding infrastructure and software.

Nvidia provides the computing through its chips, developers build and train models, where Hugging Face helps distribute and develop those models, and in return, Nvidia provides more of the infrastructure needed to run them.

The more open models spread, the more AI applications can be built. And the more those applications are trained and run at scale, the greater the requirement for computing infrastructure.