Business & Markets · · Team LATENT
NVIDIA agrees to acquire Hugging Face in a deal worth about $12.9 billion
Hugging Face, which supports publishing and using AI models, is set to join the chipmaker. Support for other chips and clouds is to continue. We explain the deal's components and implications for developers.
NVIDIA, the AI chipmaker, has agreed to acquire Hugging Face, the service for sharing models and datasets. What changes when a chip supplier owns a major developer platform? We examine the rationale and the commitments that matter to users.
Key takeaways
- The company behind services for finding, downloading and running models is set to become part of NVIDIA.
- Support for other chips and clouds is to continue, and using NVIDIA computing infrastructure will not be mandatory.
- The acquisition has been agreed, but closing still requires regulatory approvals and other conditions.
What happened
- On September 3, NVIDIA announced an agreement to acquire Hugging Face for about $12.9 billion. The agreement was signed on September 2, US time.
- According to its SEC filing, consideration paid to shareholders is approximately $11.9 billion, with up to approximately $1 billion in additional employee equity awards.
- Closing is expected in the first half of 2027, subject to regulatory approvals and other conditions.
Primary sources: NVIDIA announcement, SEC filing
Hugging Face supports everything from publishing models to running them
Hugging Face operates a service for publishing AI modelsNote 1 and training datasets and sharing them with other developers. Its central Hub lets users find a model, read its documentation and download the necessary files. It also hosts applications that developers make available to try.
Its role extends beyond distributing files. In January 2025, it launched Inference Providers, allowing users to select among multiple execution providers from a model's page. Users can have an external service run the model without provisioning their own server. It connects discovering a model with starting to use it.
According to NVIDIA's announcement, Hugging Face serves more than 18 million developers, researchers and other users, and over 200,000 companies use it to select or run models. The acquisition includes both this service and the team behind it.
The roughly $12.9 billion figure includes employee equity awards
NVIDIA CEO Jensen Huang announced a deal size of $12.9303 billion. The US Securities and Exchange Commission filingNote 2 separates approximately $11.9 billion in shareholder consideration from up to approximately $1 billion in equity awards intended to retain employees joining NVIDIA.
Shareholder consideration is subject to certain adjustments, and the employee awards are a maximum amount. Together, they account for the approximately $12.9 billion announced deal value.
The acquisition agreement was signed on September 2, US time. Closing is expected in the first half of 2027, subject to regulatory approvals and other conditions. The acquisition was not complete at the September 3 announcement.
Units: US$100 million. Employee equity awards are shown at their maximum amount.
The announced approximately $12.9 billion includes employee awards as well as shareholder consideration.
SourceNVIDIA Form 8-K. Bar lengths use the approximate figures in the filing.
Support for other chips and clouds is to continue
A direct concern for users is whether they can still choose where to run models. In its announcement, NVIDIA explicitly says that developing on Hugging Face or serving models will not require NVIDIA infrastructure. Support for other chips and multiple clouds is also to continue.
This policy also includes users' choice of models and development software. It is a commitment to preserve uses such as finding a model on Hugging Face and running it on company-owned infrastructure or another provider's service.
For developers already running models on their own hardware or another cloud, the issue is whether continued use of Hugging Face will require changing environments. NVIDIA says the choice will remain with users after the acquisition.
Wider adoption of open models can drive demand for NVIDIA products
In its SEC filing, NVIDIA explains that demand for open foundation models and applications built on them encourages use of its products. Broader model adoption increases the computing needed for training and inference. As an AI chip supplier, it has a commercial reason to support that adoption.
The companies have worked together before. Their August 2023 partnership announcement outlined plans for businesses to access NVIDIA's DGX Cloud computing infrastructureNote 3 through Hugging Face to train and adapt models on their own data.
If the acquisition closes, NVIDIA will directly operate a platform where developers choose models and execution providers, as well as supply computing infrastructure. It can introduce its infrastructure while users are still looking for models. This additional point of contact is another plausible commercial benefit of the deal.
Alongside the promise to support other chips, it will therefore matter which execution services are presented and how easy they are to choose. The announcement commits to preserving choices but does not specify the future interface or service structure.
NVIDIA identifies regulation of open models as a deal risk
The SEC filing also notes that government regulation could limit which models and datasets Hugging Face can host. New requirements on publishing, distributing or using models could require service changes or raise costs.
NVIDIA specifically notes that models developed in China are used by developers worldwide, including in the US. Restrictions that prevent support for models from particular regions could materially affect Hugging Face and NVIDIA's businesses.
This is not an announcement of new regulation or the removal of any particular model. It expresses NVIDIA's view that realizing the acquisition's benefits depends on an environment in which open models can be distributed and used internationally.
No user pricing changes or migration requirements have been announced
The September 3 announcement and SEC filing do not announce pricing revisions, account migration or required changes to existing APIs. The acquisition agreement does not itself require users to change anything immediately.
Users incorporating models into their work still need to check each model's license. Hugging Face's official documentation likewise directs users to follow the individual licenses of published code and data. Commercial use and redistribution depend on the chosen model's terms.
Timeline
Dates follow the announcing organizations' own dates.
NVIDIA and Hugging Face announce a partnership and plans to use DGX Cloud for model training and adaptation.
Hugging Face announces Inference Providers, letting users choose among execution providers from a model page.
NVIDIA signs the agreement to acquire Hugging Face.
The acquisition agreement is announced. Closing is expected in the first half of 2027, following regulatory approvals and other conditions.
Sources and references
- NVIDIA acquisition announcement NVIDIA / September 3, 2026 / Archived version
- NVIDIA Form 8-K Filed September 3, 2026 / Acquisition terms and risks
- Inference Providers announcement Hugging Face / January 28, 2025 / Archived version
- Licenses for published models and datasets Hugging Face / Archived version
- NVIDIA and Hugging Face partnership announcement NVIDIA / August 8, 2023
Update history
- Published.