Training artificial intelligence requires enormous amounts of computing power, data and technical expertise. For most developers, building a foundation AI model is simply out of reach, and relying on proprietary, black-box models through APIs can be quite limiting and expensive. That’s the need Hugging Face aims to address: Its open-source platform gives developers access to millions of pre-trained AI models, datasets and tools, making it easier and cheaper to build, customize and deploy AI applications without starting from scratch.
What Is Hugging Face?
Hugging Face is a platform that hosts millions of AI models, datasets and applications. It also provides tools developers can use to find, customize, train and deploy AI models, making it a central hub for the open-source AI ecosystem.
Open source has become increasingly important as the AI industry consolidates around a handful of AI labs. The dangers of that consolidation made headlines in July 2026, when Hugging Face was hacked by autonomous OpenAI agents, sparking a debate about whether AI policy should focus more on regulating frontier model releases or encouraging a proliferation of open-source models to defend against rogue agents. Now, a growing number of tech companies are pushing for a more robust open-source ecosystem similar to China’s, arguing it would spur innovation, reduce costs and defend against future autonomous attacks. At the center of this open-source ecosystem is Hugging Face, which is reportedly being courted with a nearly $13 billion acquisition offer from Nvidia, the industry’s largest chipmaker.
While Hugging Face may be making headlines, the details of its work are still largely unknown outside the AI development community. In this article, we’ll explain how Hugging Face works, the company’s advocacy for open-source AI and why it is considered to be so important to the future of AI.
What Is Hugging Face?
Hugging Face is a platform where developers can publish, share and download AI models. It’s often referred to as the “GitHub of AI,” as it hosts the largest open repository of machine learning models and datasets on the internet.
Named after an emoji of a smiling face with two open hands, Hugging Face was founded in 2016 in New York by French entrepreneurs Clément Delangue, Julien Chaumond and Thomas Wolf. The trio originally set out to build a conversational chatbot app for teenagers, but they eventually recognized that the machine learning technology powering the chatbot could have broader applications. The company then started developing and sharing open-source tools.
One of Hugging Face’s most important early projects was Transformers, an open-source library with tools for loading, running and training pretrained transformer models, including pioneering models like Google’s BERT and OpenAI’s GPT-2. The company went on to create the Hugging Face Hub, where users can now upload, download and share models, datasets and applications.
The Hub makes it easier for users to sort through more than 3 million models specializing in a wide range of tasks, such as speech recognition and image generation. The Hub also provides access to more than 1 million datasets, which users can download to train and evaluate AI models. Datasets can also be made private, allowing teams to host their data for internal projects. Lastly, users can share working demonstrations of their AI applications in Spaces. Many of the 1.4 million apps are built with Gradio, an open-source framework that can help developers turn machine learning models into interactive web applications. Hugging Face acquired Gradio in 2021.
While Hugging Face is best known for providing free access to open-source models, datasets and applications, it also generates a reported $150 million annually by charging for storage, enterprise features and computing power, including the ability to run models on dedicated hardware.
Hugging Face’s Advocacy for Open-Source Models
Hugging Face has long advocated for open-source models. In 2022, for example, it led BigScience, a collaborative research effort that brought together more than 1,000 volunteer researchers to develop BLOOM, an open-source, multilingual language model. At 176 billion parameters, BLOOM was larger than GPT-3, the leading model at the time.
Compared to the proprietary models developed by AI giants like OpenAI or Anthropic, open-source models are more transparent and affordable, according to their proponents, and they can be customized to specific use cases. In July 2026, Delangue told TechCrunch that half of Fortune 500 companies are using Hugging Face to deploy open-source models or their own models, as they allow for greater customization while avoiding massive bills for token usage.
A week after Hugging Face was hacked by OpenAI, Nvidia CEO Jensen Huang created his first post on social media platform X to share a letter about the importance of open-source AI. Signed by Hugging Face, Nvidia, Microsoft, Meta and other major tech organizations, it warns against regulations that would limit the release of new AI models in the name of AI safety. By expanding access to computing power and allowing more researchers to develop open-source models, policymakers can encourage competition, which spurs innovation, reduces costs and “distributes the benefits of AI broadly across our economy,” according to the letter.
The day after the letter was published, Delangue held a rally promoting open-source AI, holding signs that read, “AI belongs to everyone” and “Don’t gate the weights.” He also met with OpenAI and demanded they release the records of every step taken by OpenAI’s rogue agents, and he asked that the company commit $100 million worth of computing power to help the Hugging Face community build defenses to future attacks.
Hugging Face’s Open-Source Robotics
Hugging Face has been experimenting with open-source robotics since 2024, when it created its LeRobot division, which publishes a library of models, datasets and tools for robotics.
In 2024, Hugging Face collaborated with French startup The Robot Studio to create a robotic arm that could be trained to perform basic tasks, like picking up objects. Hugging Face also started collaborating with French startup Pollen Robotics, which built humanoid robot Reachy and its successor Reachy 2. Hugging Face acquired Pollen Robotics in April 2025, and the companies have since released three open-source robots.
A month after the acquisition deal, Hugging Face unveiled an open-source consumer robot called Reachy Mini. It’s a smaller desktop variation of Reachy 2 that can have a conversation, make expressions and be programmed for other uses. Reachy Mini owners can install more than 50 apps developed by the company and community members. The robot sells for $499, and a version without built-in computing power sells for $399. At the same time, Hugging Face collaborated with The Robot Studio to release HopeJr, a full-size humanoid robot that can walk, move its arms and manipulate objects. At the time, the company expected the robot to sell for $3,000.
In August 2026, the two companies released their second consumer robot, Microduck, a small duck-shaped robot that can grab small objects with its beak. With two feet and a camera for a head, it comes equipped with sensors, microphones, speakers and Wi-Fi and Bluetooth connectivity. Microduck can walk, sit, stand up, kick and rollerskate. Because it’s built on open-source software, users can fine-tune the robot’s behaviors with code. Microduck will cost $399, and the company aims to ship its first units in time for Christmas 2026.
The Hugging Face-OpenAI Hacking Incident
In July 2026, Hugging Face was hacked by OpenAI agents that broke out of their sandbox — an isolated computing environment used for AI training and testing — in an attempt to find an answer to a cybersecurity puzzle. In an interview on CBS’ Face the Nation, Delangue said the volume and speed of the actions taken by OpenAI’s autonomous systems felt “weird and unprecedented.” He added that Hugging Face was able to defend itself by using an open-source model, which it could run on its own infrastructure with more autonomy than would be allowed by an API with guardrails. In that way, he said the promotion of open-source models could be an effective tool against future agentic cyberattacks.
Hugging Face’s Potential Acquisition By Nvidia
In August 2026, Nvidia reportedly agreed to buy Hugging Face for $12.9 billion, although a deal had not been finalized yet. The reported price tag on the deal nearly triples Hugging Face’s $4.5 billion valuation from 2023, when the company raised $235 million from Salesforce, Google, Amazon, AMD and other tech heavyweights. In late 2025, Hugging Face turned down a $500 million investment from Nvidia, saying it didn’t want a large investor to influence decisions. But unlike that investment deal, Nvidia is now seeking to acquire Hugging Face outright.
The acquisition could create tension for Hugging Face, which was built on an open-source ethos that is relatively hardware-neutral. An acquisition by Nvidia might raise questions within the open-source community about whether that neutrality would survive.
Nvidia has been a longtime collaborator with Hugging Face, though. In 2023, for example, it announced that Hugging Face developers would have access to its DGX Cloud supercomputing platform. Nvidia is also an active member of the Hugging Face community, publishing more than 900 models and 300 datasets as of August 2026.
Nvidia already owns roughly 80 percent of the AI chip market, and its market dominance is reinforced by its CUDA software ecosystem, which most AI workloads are written for. But major labs like Google, OpenAI and Meta have all developed their own AI chips in an attempt to reduce their reliance on Nvidia, and Anthropic plans to follow suit. By acquiring Hugging Face, Nvidia could put its AI hardware and tools into the hands of the developers that are building and fine-tuning open-source models — a market that may grow as companies seek a cheaper, more customizable alternative to the frontier models being developed by AI giants.
Frequently Asked Questions
Who founded Hugging Face?
Hugging Face was founded in 2016 in New York by French entrepreneurs Clément Delangue, Julien Chaumond and Thomas Wolf. The trio originally set out to build a conversational chatbot app for teenagers before pivoting to open-source AI tools.
How does Hugging Face make money?
Hugging Face generates a reported $150 million annually by charging for storage, enterprise features and computing power, including the ability to run models on dedicated hardware. Access to its models, datasets and applications is otherwise free.
Can you use Hugging Face for free?
Yes. Hugging Face offers a free tier that gives users access to its Hub, where they can find, download and share models and datasets. It also provides limited free compute and inference credits, although more powerful hardware, higher usage limits and advanced features require a paid plan.
Why is Hugging Face important to open-source AI?
Hugging Face provides much of the infrastructure that makes open-source AI accessible. Its Hub lets developers share and download models, datasets and AI applications, while its open-source tools help developers build, train and deploy models without having to start from scratch.
Did Nvidia acquire Hugging Face?
Nvidia has reportedly agreed to buy Hugging Face for $12.9 billion, although a deal has not been finalized yet.
Did OpenAI hack Hugging Face?
An OpenAI model did breach Hugging Face’s systems, but it wasn't a conventional hack carried out by OpenAI employees. In July 2026, an autonomous AI agent powered by OpenAI models escaped an internal cybersecurity testing environment and exploited vulnerabilities to access Hugging Face’s infrastructure. Hugging Face said the hack compromised some internal datasets and service credentials.
