Nvidia is acquiring Hugging Face, the widely used repository and development platform for artificial intelligence models, for $12.9 billion, bringing a central piece of the open AI ecosystem under the control of the world’s dominant supplier of AI chips.
The transaction follows years in which Hugging Face resisted takeover approaches and offers to purchase stakes in the company. Its founders selected Nvidia in part because of the chipmaker’s support for open AI development, the company said. Nvidia already plays a critical role in the sector because developers and cloud providers rely heavily on its processors to train and run AI systems.
Hugging Face hosts more than two million models and has 18 million users. The service has become an important distribution point for open-source and open-weight technology, allowing developers to download, adapt and deploy models rather than access them solely through a provider-controlled interface. Its business generates revenue from storage, computing capacity and tools for managing AI systems.
That position also places Hugging Face near some of the most contested legal issues in generative AI. Model repositories can contain software, training artifacts, datasets and technical documentation released under different licenses and usage conditions. Developers and businesses adopting those resources must determine whether a model’s license permits modification, redistribution or commercial deployment, as well as whether separate restrictions apply to associated code and data.
Open-weight models are not necessarily open source under traditional software definitions. Some providers make model parameters available while imposing limits on particular uses or withholding training data and other components. The distinction can affect how companies assess intellectual property rights, compliance duties and their ability to build products on top of a model. Hugging Face’s role is primarily to provide hosting and development infrastructure; the availability of a model on the platform does not by itself resolve ownership, copyright or licensing questions.
The company began with a different product. Co-founders Clément Delangue, Julien Chaumond and Thomas Wolf initially sought to create a conversational AI service aimed largely at teenagers. They later shifted toward a business-focused platform centered on access to machine-learning models. Wolf, a quantum physics doctorate who also worked as an intellectual property lawyer, helped build the company alongside Delangue and Chaumond.
Although founded in New York and backed substantially by US investors, Hugging Face retains strong French roots. Its three founders are French, and the company participated in Station F, the Paris start-up campus. French investors joined some funding rounds, but larger investments came from US sources, including Lux Capital and basketball player Kevin Durant.
The sale has renewed concern about Europe’s ability to finance technology companies through later stages of growth. French Economy Minister Roland Lescure warned before the formal announcement that European start-ups would continue seeking capital elsewhere if investors in the region could not provide enough funding to scale them. He described the deal as a “wake-up call.”
Hugging Face expects to maintain its links to France after the acquisition. Delangue said a substantial portion of its workforce is French and expressed hope that employees would reinvest in Europe’s technology sector. He also said the transaction could encourage Nvidia to direct more investment toward France and the wider European market.
The acquisition gives Nvidia ownership of infrastructure used across an AI market that includes independent researchers, start-ups and large enterprises. It also connects Nvidia more directly to an open-model community positioned as an alternative to closed systems from OpenAI, Anthropic and Google. How Hugging Face will operate within Nvidia, including whether its platform policies or model-hosting practices will change, has not yet been detailed.
Sources: Generative AI intellectual property