Nvidia Acquires Hugging Face for $12.9 Billion and Promises Neutrality

The GPU manufacturer plans to acquire Hugging Face for $12.9 billion and assures that the platform will remain open and independent of hardware. The closure is targeted for the first half of 2027, pending regulatory approval, while Nvidia also sees it as a boost for its computing activities.

Cloud Contracts and Hosting: A New Outlet for Computing

At the end of July, Nvidia announced it had secured $36 billion in commitments from agreements with AI cloud partners. In these contracts, the share of commitments that Nvidia receives decreases when partners sell their capacity to third parties, and Nvidia partially covers its partners against unused capacity. Hugging Face, which distributes models and offers compute time rental for their execution, fits into this dynamic. Nvidia had previously reduced its DGX Cloud offering to avoid competing with its major rental clients. Its Lepton market now directs workloads to partners like CoreWeave, Lambda, and Nebius, and Hugging Face connects to it via a cluster service. The integration of a large developer hub would strengthen Nvidia's influence over a central showcase of competition around open models.

Financial Structure and Timeline Under Regulatory Control

The operation anticipates around $11.9 billion for the acquisition, supplemented by a stock program that could reach $1 billion aimed at retaining teams. According to a filing with the SEC, the finalization will not occur before the first half of 2027 and remains subject to regulatory approval. Nvidia thus plans to acquire Hugging Face for a total of approximately $12.9 billion. Jensen Huang officially announced the agreement on September 3, 2026, after information about the operation was released at the end of August.

Universal Access and Announced Hardware Neutrality

Jensen Huang assures that Hugging Face will remain open to all, without mandating the use of Nvidia hardware, and will continue to provide support for other clouds and types of chips. According to Nvidia, the platform has over 18 million developers sharing more than 3 million models, 500,000 datasets, and 1 million applications, while more than 200,000 companies use it. Nvidia claims to be the largest contributor of open models on Hugging Face, with over 500 models and more than 250 datasets. Hugging Face remains a central hub for open AI models.

Technical Context: In-House Accelerators and Open Models

Several major providers are now developing their own accelerators, including Google, Amazon, OpenAI, and Anthropic. Open models, on the other hand, run on numerous clouds, in enterprises, universities, and government agencies—clients that do not manufacture their own chips. Jensen Huang summarizes his strategy as follows: “free AI should be great for hardware.” Nvidia already offers open models with Nemotron, publishing the weights, a large portion of the training data, and associated recipes, and collaborates within the Nemotron Coalition with Mistral AI, Thinking Machines Lab, Perplexity, Black Forest Labs, Cursor, LangChain, Reflection AI, and Sarvam. The company also offers NVFP4 versions of Kimi K2.6 and GLM-5.1 optimized for its chips. It is noted that Nvidia is not leading against powerful Chinese models, even though its models are faster in some cases.

Who Initiated the Discussions? Divergent Accounts and Stated Goals

Accounts differ regarding who initiated the rapprochement: Jensen Huang claims that Clem Delangue reached out to him to consider the next step for the company, while according to a LinkedIn message from Thomas Wolf, Huang would have proposed to Delangue to transform the hub into an open, independent, and hardware-neutral platform. Thomas Wolf believes that Nvidia is the most relevant partner for the company's mission, which includes open weights, robotics, and science. He emphasizes that no changes impact users to date and considers that open AI has reached a phase where scale and computing power become decisive.