NVIDIA buys Hugging Face in a 12.93 billion dollar deal
NVIDIA said on September 3, 2026 that it has agreed to acquire Hugging Face, the platform that hosts open AI models, applications and datasets, for 12.93 billion dollars. It is the company's second-largest acquisition, and closing is expected in the first half of 2027 subject to regulatory approval. NVIDIA also stated in its announcement that its own compute will not be a requirement for building on the platform.
A 12.93 billion dollar agreement, closing in 2027
NVIDIA announced the Hugging Face agreement on its own blog on September 3, 2026, and the news was carried by TechCrunch, Webrazzi and Tom's Hardware. The headline figure is 12.93 billion dollars, with Webrazzi giving the precise number as 12,930,300,000 dollars. The transaction is expected to close in the first half of 2027 once regulatory approvals are complete. It is the second-largest acquisition in NVIDIA's history: according to Reuters and Adweek, it sits behind only the roughly 20 billion dollar agreement in December to buy assets from chipmaker Groq. As notable as the deal itself is the openness commitment attached to it: NVIDIA says the platform will stay open to teams working across different clouds and different accelerators.
11.9 billion for shareholders, up to 1 billion for staff
The consideration comes in two parts. Adweek reports that approximately 11.9 billion dollars is payable to Hugging Face shareholders, with up to 1 billion dollars in retention equity set aside for employees joining NVIDIA. The announcement also put numbers on the platform's scale: by NVIDIA's own account, Hugging Face hosts 3 million models, 1 million applications and 500,000 datasets. The model count is where sources diverge, since Adweek instead describes upwards of 2 million models; the application and dataset figures are not contested. The platform is used by 18 million developers, researchers and creators, and by more than 200,000 companies. NVIDIA says it is the largest contributor of open models and data there, having published over 500 models and more than 250 open datasets. The commitment in the announcement is unusually direct: ‘NVIDIA compute will not be required to build on or deploy through Hugging Face'.
A hardware vendor is buying the distribution point
The price tag carries the headline, but the mechanism sits elsewhere. Hugging Face is where open models, datasets and applications sit together, so any team that pulls down an open model has a link of its supply chain running through this platform. That makes the durable question the openness commitment rather than the amount paid. Closing is not expected before the first half of 2027, which means nothing technically changes for developers this week. The sober read is this: once a single hardware vendor owns the main distribution point for open models, single-vendor dependency turns into a risk item that has to be planned against. NVIDIA's commitment is written and explicit today, and how it works in practice will be shown by post-closing product and pricing decisions. That is not an accusation, it is a line to keep watching.
Portability is the decision teams make in 2026
For agencies and companies in Türkiye the practical response is inventory, not alarm. If your stack uses open models, write down which models, which datasets and which libraries arrive through Hugging Face. The next step is keeping portability explicit at both the architecture and the contract level: model choice, framework choice and cloud choice should stay separable, and none of the three should be locked to the same vendor. If you run an assistant or an agent in production, holding your own copy of the model weights you depend on is cheap insurance against next year's uncertainty. For teams with enterprise clients there is one more item: name this platform in your KVKK documentation and vendor lists, because being able to show where a model came from is now an expected answer in an audit. Closing is in 2027, preparation happens in 2026.
The UNALSOFT take
From where we sit this is a supply chain story rather than an acquisition story. On projects built with open models, keeping model, framework and cloud choices independent is setup work, not cleanup work, and we put it on the table early in agentic AI engagements. Nothing changes technically before the first half of 2027, yet most of the architectural decisions are being made now. A system built portable today keeps its options whichever way the post-closing decisions go.
Running anything on open models?
We can review together how portable your model, framework and cloud choices actually are. A short conversation is enough to start.