The Information reported a $12.9 billion agreement on August 27. Reuters, CNBC, and TechCrunch carried that figure the same day. Neither company posted it. Jensen Huang put $12,930,300,000 on the NVIDIA blog today and said the company will acquire Hugging Face. NVIDIA’s Form 8-K, dated September 2 and filed September 3, splits the number: about $11.9 billion to Hugging Face stockholders and an equity retention program of up to $1.0 billion for employees who join NVIDIA. Close is expected in the first half of 2027 if regulators approve the deal (The Information, 2026; NVIDIA, 2026).
Huang’s post describes the catalog the way buyers already talk about it. More than 18 million developers, researchers, and creators share more than 3 million models, 500,000 datasets, and 1 million applications. More than 200,000 companies use the platform to find, test, customize, and deploy models (NVIDIA, 2026).
Print the pledge. Huang says Hugging Face remains an open platform, NVIDIA compute is not required to build or deploy through the hub, and other silicon vendors stay supported (NVIDIA, 2026).
NVIDIA Is Buying the Place Teams Pick Open Models
Boards still spend most of the AI budget argument on training clusters. A larger share of day-to-day work now happens after that, when an engineer picks a checkpoint, fine-tunes it, and pushes it into an application. For open weights, that work happens on Hugging Face.
Open models are also arriving faster from labs outside the United States. NVIDIA still needs those models to run on its racks. Owning the catalog does not stop those labs from shipping. It puts NVIDIA closer to which checkpoints get downloaded, which datasets get reused, and which inference paths get tried first.
I expect NVIDIA to keep posting its own open models at a high rate. The company already says it is the largest contributor of open models and data on Hugging Face, with more than 500 models and more than 250 datasets (NVIDIA, 2026). More open weights in circulation means more jobs that need accelerators, evaluation tools, and deployment software. NVIDIA already ships that stack under names such as Nemotron and NeMo. I wrote about the Nemotron 3.5 Lightning release and the open router that shipped with it on August 11.
A chip shipment report does not tell NVIDIA which open model a bank actually put behind an internal API last Tuesday. The hub logs do.
Chip Vendors Add Software. Software Vendors Add Compute.
Chip and infrastructure companies keep adding software that sits closer to the application. Application companies keep adding their own models, inference, and in some cases custom silicon so they are not stuck on one vendor’s release calendar. Hugging Face is one of the few public places where both of those moves are visible at once.
Huang already wrote the independence pledge. The purchase still gives NVIDIA a view of the work after a model leaves a lab: who forks it, who evaluates it, who puts it behind an internal API.
Huang also points back to the July open-weights letter he coauthored, which Hugging Face signed. I covered that letter when it posted. The cost case for open weights holds up. The verification case still has no accepted test. Read Nvidia's Open Weights Letter Settles the Cost Argument next to this deal. The letter asked Washington not to choke open models. The acquisition puts the main public shelf for those models on NVIDIA’s books.
The 8-K Pledge Is What You Test After Close
The 8-K says NVIDIA committed to keep the platform open in line with current practice, including uploads and downloads of models and datasets the user chooses, and support for other silicon vendors (NVIDIA, 2026). File that sentence. Then watch the product, not the brand line.
Check whether a new model card or inference endpoint defaults to an NVIDIA runtime. Check whether enterprise private hubs keep the same terms once the parent is a public chip company. Check whether NVIDIA’s own open models still appear next to Chinese and European checkpoints on the front page.
The deal still needs regulators. Until it closes, Hugging Face is the catalog you already use, and NVIDIA now has a reason to care about every download on it.
List the open models your teams pulled from Hugging Face last quarter and the chips those models run on. If that list includes accelerators NVIDIA does not sell, put a reminder on the calendar for the first quarter after close.
Huang, Jensen. "NVIDIA to Acquire Hugging Face." NVIDIA Blog, 3 Sept. 2026, blogs.nvidia.com/blog/nvidia-to-acquire-hugging-face/.
NVIDIA Corporation. "Form 8-K." U.S. Securities and Exchange Commission, 3 Sept. 2026, www.sec.gov.
Efrati, Amir. "Nvidia Agrees to Buy Hugging Face For $12.9 Billion." The Information, 27 Aug. 2026, www.theinformation.com.
Reuters. "Nvidia Agrees to Buy Hugging Face for $12.9 Billion, The Information Reports." Reuters, 27 Aug. 2026, www.reuters.com/technology/nvidia-talks-acquire-hugging-face-13-billion-deal-business-insider-reports-2026-08-27/.
Bellamkonda, Shashi. "Nvidia's Open Weights Letter Settles the Cost Argument. The Backdoor Question Still Needs an Answer." shashi.co, 25 July 2026, www.shashi.co/2026/07/nvidias-open-weights-letter-settles.html.
Bellamkonda, Shashi. "NVIDIA Ships a Router That Outlasts the Model Beside It." shashi.co, 11 Aug. 2026, www.shashi.co/2026/08/nvidia-ships-router-that-outlasts-model.html.
