Nvidia has agreed to purchase Hugging Face for $12.9 billion, placing one of the dominant forces in AI hardware in control of a platform that developers use to locate and deploy open models. The Information first disclosed the agreement Wednesday, according to someone with knowledge of the transaction, though neither company had publicly verified the deal as of publication. The acquisition raises questions about whether Nvidia will maintain the hardware neutrality that has made Hugging Face valuable to developers working across different chip ecosystems.
The purchase price represents roughly 86 times Hugging Face's annualized revenue, which the report pegs at approximately $150 million. Hugging Face CEO Clément Delangue told The Information in June that the company had doubled its paying subscriber count during the first half of 2026 and later stated the company was nearing profitability. Currently, Hugging Face's Inference Endpoints allow developers to deploy models from the Hub with the company managing infrastructure on AWS, Microsoft Azure, or Google Cloud, though most GPU options listed are Nvidia chips like the T4, L4, and A100. Developers get broader hardware choices through Hugging Face's open-source libraries than through its hosted services, and Nvidia's NIM already integrates with models hosted on Hugging Face, letting developers pull models directly from the Hub using an hf:// repository path.
The report notes that Hugging Face currently sits between the model and the chip, and doesn't steer developers toward any single chipmaker—a characteristic that makes this deal notable. According to the report, its Optimum libraries function with Nvidia's TensorRT-LLM but also accommodate hardware from AMD, Intel, and AWS, with projects like Optimum AMD and Optimum Intel enabling developers to run Transformers and Diffusers models on non-Nvidia hardware. The company influences what occurs after a developer selects a model, including how simply they can get it operational on their preferred hardware. Nvidia hasn't announced intentions to make NIM the default deployment option, and support for AMD and Intel could stay unchanged.
The report explains that this tension helps clarify why Hugging Face might be worth substantially more to Nvidia than its revenue alone would indicate. Nvidia has been building out its Nemotron family of open models while making significant investments throughout the AI ecosystem, yet some of its largest customers are attempting to lessen their reliance on Nvidia hardware—Google with TPUs, AWS with Trainium, Microsoft with Maia, and OpenAI and Anthropic developing their own AI server chips. A robust open-model ecosystem provides Nvidia a counterweight to that trend, the report argues, because open models are typically expected to operate in diverse environments, and Hugging Face enables developers to achieve that portability. Five European companies recently committed to buying AI compute built around non-Nvidia hardware that hasn't even been produced yet, while OpenAI this week released results from its Jalapeño accelerator showing 1.5 to 1.9 times more work per watt and latency reductions up to 3.6 times on large open-weight models, though the chip hasn't been deployed at Nvidia's scale.
The report concludes that the core challenge for Nvidia is that it's acquiring a platform whose worth depends on openness and hardware neutrality, but if the purchase leads to favoring Nvidia hardware, that worth could decline. Owning Hugging Face wouldn't grant Nvidia control over everything developers find there—libraries like transformers and diffusers are open source, models on the Hub remain subject to their own licenses, and openly licensed models can be hosted elsewhere while underlying libraries can be forked. The bigger question is what unfolds over time: Nvidia could offer earlier support for new models on its own hardware or simplify deployment, while AMD, Intel, and AWS may need to reconsider how much engineering effort they want to invest in integrations maintained within a competitor-owned platform. A competing chip doesn't need to vanish from Hugging Face to become less attractive if an Nvidia model deployment requires fewer steps, and much harder to recreate is the community Hugging Face has built, with developers already knowing where to look for models and having constructed workflows around the Hub and its integrations. The acquisition tests whether a marketplace built on neutrality can maintain its appeal under ownership by one of the vendors it serves, and whether developer loyalty to the platform will outlast any shifts in how easily different hardware options integrate.