GPU giant Nvidia has agreed to buy Hugging Face, an AI model, data, collaboration and platform with more than 13 million registered users, for $12.9 billion, according to a report today in The Information. If the deal goes through, it would significantly bolster Nvidia’s AI model business.

Hugging Face was founded 10 years ago by Clément Delangue, Julien Chaumond, and Thomas Wolf to develop a chatbot for teens. The mobile app used natural language processing (NLP) models to interpret the text or graphical input, and generated a response designed to delight the user–and perhaps even use the “hugging face” emoji (hence the name).

Hugging Face Co-founder and CEO Clément Delangue

When the founders noticed more interest in the AI models behind the chatbot than the chatbot itself, they decided to open source the technology, posting it to GitHub, where the technology was widely accessed and used.

In 2018–a year after Google published the seminal paper on transformers, “Attention Is All You Need”–the New York City company its strategy. Instead of a chatbot, Hugging Face became a repository of open source AI and machine learning models, the data sets they were trained on, and other resources for AI engineers, data scientists, and machine learning specialists to work with the tools.

One of the first models they posted was an open source version of Google’s BERT, an early LLM. The PyTorch-based version was well-received by the community. In 2021–more than a year before OpenAI shocked the world with ChatGPT–Hugging Face launched a project called BigScience designed to help foster the development of LLMs and large datasets in the scientific realm.

Hugging Face was a popular repository for all sorts of AI models before ChatGPT triggered the tsunami of interest in LLMs.  But there’s no doubt the AI boom accelerated Hugging Face’s growth. It crossed the 1 million model threshold in 2024, 2 million models in 2025, and today has about 3 million AI models, as well as half a million data sets. Much of this technology and data is free; the company reported only $150 million in annual revenue.

The company has growth through venture capital funding, which has totaled $395 million over the years. In 2023, it raised $235 million in a Series D round (its last major funding round) at a $4.5 billion valuation. Nvidia was among those Series D investors, along with Google, Amazon, Intel, AMD, Qualcomm, and IBM.

It will be interesting to see how Hugging Face fits into Nvidia’s strategy or changes it. Nvidia and Hugging Face announced a partnership in 2023 that made it easier for Hugging Face users to get access to Nvidia GPUs for training. In late 2025, Nvidia launched its own line of open-weight models, dubbed Nemotron, which range in size from 3.5B to 550B parameters.

The popularity of Hugging Face among AI developers is its biggest asset; with just 2,000 paying customers and a mere $150 million in revenues, the revenue stream is clearly not it. Hugging Face has become the defacto standard place where a generation of AI engineers go to download and play around with AI models.

Open-weight models have surged in popularity in the U.S. this summer, following a surprising improvement in Chinese open-weight models like Alibaba Qwen, Alibaba’s Kimi, and DeepSeek, which now are nearly on par with the best proprietary models from the big American labs at Google, OpenAI, and Anthropic. Nvidia CEO Jensen Huang and Hugging Face CEO Clément Delangue were among the executives who signed the “Open Weights and American AI Leadership” letter last month.

If Nvidia sees a big figure in open-weight AI models, then it couldn’t do much better than acquire Hugging Face, which is where many of these models live.

Nvidia did not respond to a request for comment.