Nvidia is buying Hugging Face for $12.9 billion, and the neutral model hub now answers to a chip vendor
Nvidia agreed to acquire Hugging Face for $12.93 billion, putting the model hub 18 million developers rely on inside the company that sells most of the GPUs those models run on. The open-platform pledge is the part that matters, and it does not close until 2027.
Nvidia agreed on September 3 to acquire Hugging Face for $12.93 billion, its second-largest deal ever and a direct move into the open-model ecosystem it has spent years supplying with hardware. Jensen Huang announced it on Nvidia’s blog, Hugging Face CEO Clem Delangue confirmed it on X, and the structure is public: roughly $11.9 billion to shareholders, up to $1 billion in retention equity for employees who join, and a close expected in the first half of 2027 pending regulatory approval. The number is large. The thing that actually changes for working developers is smaller and more specific: the neutral place they pull models and datasets from now has an owner that sells accelerators.
Why the hub is not just another acquisition target
Hugging Face is infrastructure most machine-learning work quietly routes
through. More than 18 million developers use it to share over 3 million models,
500,000 datasets, and 1 million applications, and 200,000 companies use it to
find, evaluate, and deploy them. The transformers library and the Hub sit
under a large share of the open-weight world, from Llama to Qwen to the long tail
of fine-tunes. That reach is exactly why the buyer matters. When a database
company or a cloud buys a tool, the concern is lock-in. When the dominant GPU
vendor buys the default distribution point for open weights, the concern is
whether the platform stays as indifferent to your hardware and your model choice
as it is today.
The pledge is the load-bearing part
Huang addressed that directly, and the specificity is worth reading rather than paraphrasing. Nvidia says Hugging Face “will remain an open platform for the entire AI ecosystem,” that “NVIDIA compute will not be required to build on or deploy through Hugging Face,” and that it will keep supporting “open source and open weight models from across the ecosystem, from every model builder,” with multi-cloud and multi-accelerator development staying first-class. In plain terms: no forced CUDA, no demotion of AMD or TPU paths, no privileged shelf for Nvidia-optimized checkpoints. That is the right set of promises. It is also a set of promises, and the value of the hub to developers has always been that its neutrality was structural rather than pledged. Post-close, neutrality becomes a commitment a hardware vendor chooses to keep.
What changes now, which is close to nothing
The near-term developer answer is calm: the deal does not close until the first
half of 2027, and until then Hugging Face operates as an independent company. No
API changes, no pricing changes, no library changes ship because of this
announcement. Nothing in your pip install transformers pipeline moves this
quarter. The honest posture is to treat the pledge as the thing to monitor, not
the thing to react to. The signals that would tell you neutrality is eroding are
concrete and slow: whether Nvidia-optimized model formats start getting
placement the rest do not, whether inference on non-Nvidia accelerators quietly
loses parity in the tooling, whether the retention equity keeps the people who
built the culture that made the hub open in the first place.
The regulatory variable is not decorative
A first-half-2027 close “subject to regulatory approval” is not boilerplate for a deal of this shape. Nvidia already sits at the center of AI-hardware antitrust attention on multiple continents, and buying the ecosystem’s shared model registry hands reviewers a clean theory to examine: the company that controls the compute now also controls the catalog. That does not mean the deal breaks. It means the timeline is real, the outcome is not certain, and any developer planning around a 2027 Hugging Face should assume the terms could be conditioned before they are final.
What to watch
Three things over the next several quarters. Whether independent, non-Nvidia hardware paths keep true parity in Hub tooling and documentation rather than slowly becoming second-class. Whether regulators in the US or EU attach conditions that touch how the platform ranks or serves models. And whether the people who made Hugging Face the neutral commons stay through the retention window, because a platform’s openness is a culture before it is a policy. Stackmaven will revisit around the expected close.