Nvidia’s $12.93 Billion Hugging Face Deal Raises Questions for Chinese Open Models
TechWireAsia reported that Nvidia agreed to buy Hugging Face for $12.93 billion, raising governance questions as Chinese open-weight models lead major download rankings on the platform.
A $12.93 billion agreement to buy Hugging Face would put the main public hub for open-weight AI models inside Nvidia, TechWireAsia reported.
The deal matters beyond model hosting because Chinese model families lead many of the download rankings that developers use to choose what they run on their own infrastructure.
Nvidia founder and chief executive Jensen Huang announced the planned acquisition on September 3 in a company blog post.
Hugging Face hosts models whose parameters can be downloaded, modified and deployed without relying on a provider's application interface, and Nvidia listed more than 18 million developers, researchers and creators using the platform to share over three million models, 500,000 datasets and one million applications.
The transaction would rank as Nvidia's second-largest disclosed purchase, behind the roughly $20 billion Groq asset deal completed in December.
Hugging Face had been valued at $4.5 billion in a $235 million funding round in 2023.
Open-weight activity supports Nvidia's hardware business even when the models come from outside the United States.
Companies that download and run their own systems still need accelerators underneath them, and Nvidia has already made the platform part of its developer strategy by releasing more than 500 models and more than 250 open datasets there.
Chinese models are central to that developer surface.
Alibaba's Qwen family passed three billion downloads in six months, while Hugging Face's 2026 counts showed 418 million downloads for Google and 227 million for Meta over a different period.
Alibaba has open-sourced more than 460 Qwen models, with more than 300,000 derivatives built on them, and its cloud unit carries Qwen to enterprise customers in Southeast Asia and Africa.
Other Chinese labs also appear high in open model usage.
A March finding by the US-China Economic and Security Review Commission identified models from Alibaba, Moonshot and MiniMax as dominant in usage rankings on Hugging Face and OpenRouter, describing China's open-source model position as a compounding competitive edge despite restrictions on access to advanced chips.
Nvidia has promised to keep Hugging Face open to the wider AI ecosystem.
Huang's post promised continued support for model builders across open-source and open-weight releases, as well as deployments that use several clouds or accelerator suppliers.
Developers would not have to use Nvidia compute to create or serve work through Hugging Face.
The buyer also framed Hugging Face as an enterprise distribution layer.
Nvidia said more than 200,000 companies use the service to discover, evaluate, customize and deploy AI.
CNBC also compared the deal with Nvidia's almost $7 billion Mellanox purchase.
Hugging Face co-founder Clément Delangue approached Huang about the company's next phase, according to accounts the two executives gave CNBC.
The deal also follows a wider policy campaign around open weights.
Huang said he had co-authored an open letter urging U.S. policymakers not to impose early restrictions on open-weight models.
The letter drew signatures from 25 companies, including Microsoft, Meta, IBM, Dell, Palantir, Mistral AI, Hugging Face and Andreessen Horowitz.
That coalition is relevant to Nvidia's promise because the platform spans competing model builders, cloud providers and accelerator suppliers.
Keeping that range available would preserve Hugging Face's role as a neutral discovery and deployment layer; narrowing it would make the ownership change material even if existing model files remained downloadable.
The ownership question remains narrower than that promise.
The company has not specified who would make final decisions over hosted models or how a US owner would respond if Washington later pressed for changes to Chinese model availability.
For Asian developers and enterprise teams that begin fine-tuning work from Qwen, DeepSeek or related derivatives, the platform's neutrality will be measured by what remains downloadable after the acquisition closes.
