The term sheet has been circulating for weeks, and on Thursday the numbers hardened. Nvidia Corp. is in advanced talks to acquire Hugging Face, the Paris-based platform where developers download and share open-source AI models, in a deal valued above $13 billion, according to people familiar with the matter. The Information reported that the two sides have reached an agreement at $12.9 billion. The people cautioned that negotiations could still collapse, and that no final terms have been signed.
If it closes, the deal would be one of the largest in Nvidia’s history, dwarfing its $6.9 billion purchase of Mellanox in 2019 and rivaling the scale of its failed $40 billion attempt to buy Arm. More important than the price, analysts said, is what it would buy: control of the place where the AI industry’s models are stored, shared, and evaluated, the exchange that many engineers treat as the default home for open-weight systems.
Hugging Face was founded in 2016 by Clément Delangue and two co-founders, initially as a chatbot app for teenagers, before it became the infrastructure layer for open AI. The company’s platform now hosts more than a million models and datasets, and its libraries, Transformers and Datasets, are embedded in the workflows of most AI developers. Its nickname, the GitHub of AI, understates the position: GitHub stores code, but Hugging Face also runs the services that let anyone run a model in the browser and the leaderboards that rank how well models perform.
The strategic logic for Nvidia is the logic of owning both ends of the market. The company already sells the training chips that every model depends on. With Hugging Face, it would also own the distribution channel through which many of those models reach developers, and the metadata and usage signals that show which architectures are winning. Nvidia has tried to build its own developer ecosystem, and its recent acquisitions have aimed at software, but nothing in its portfolio approaches Hugging Face’s reach, analysts said.
The deal would also give Nvidia a direct hand in the open-model movement at a moment when the industry is split over it. Open-weight models have become the default for enterprises that want to avoid vendor lock-in, and they are the segment of the market where Nvidia’s competitors, including AMD and the custom-chip designers, have gained ground. Owning the platform where open models live would let Nvidia shape that segment while selling the chips it runs on.
Regulators may have something to say. The acquisition would combine the dominant supplier of AI accelerators with the dominant platform for open models, and competition authorities in the United States and Europe have been circling both markets. The deal’s supporters argue that Hugging Face is a neutral utility whose value depends on developers trusting it, and that Nvidia would have no incentive to favor its own products at the expense of the community. Skeptics note that Nvidia has a history of integrating acquisitions deeply, and that the platform’s neutrality is precisely the asset being purchased.
The timing is awkward for another reason. Hugging Face spent July fending off a coordinated attack by roughly 700 AI agents, a breach that was disclosed in full only this week. The incident showed that the platform is a target, and that the community’s trust in it depends on security it has not always demonstrated. A buyer will be acquiring that risk along with the traffic, and Nvidia’s security team would inherit the job of protecting the exchange that the industry’s models flow through.
Negotiations began after reports surfaced on Aug. 24 that Hugging Face was exploring a sale at a valuation near $13 billion, the people said. Delangue has not commented publicly, and Hugging Face declined to discuss the talks. The company has raised money at a $4.5 billion valuation in previous rounds, making the reported price a substantial step up. Nvidia, with a market value above $4 trillion and cash flow that now approaches $100 billion a year, can absorb the cost without strain, analysts said.
Inside the developer community, the reaction has been wary. Hugging Face built its position on neutrality: it hosts models from OpenAI’s rivals, from Meta, from China’s DeepSeek, from anyone who publishes. Engineers told this newsroom that they use the platform precisely because it is not owned by a company that competes with the models it hosts. Nvidia, through its investments and partnerships, backs dozens of AI companies, and its chips power most of the models on the platform. Whether ownership changes the calculus for developers who publish and download open weights there will be the deal’s quiet risk, one that no term sheet can price in advance.
The open question is whether the deal survives contact with the boardroom and the regulators. The people familiar with the matter said the talks are real but not final, and that either side could walk away. The prize is the AI industry’s most valuable property after the chips themselves: the catalog of models that the whole field is built on. Nvidia’s rivals, and the developers who use Hugging Face every day, will be watching whether the GitHub of AI becomes a division of the chipmaker.
