skip to content

🫂 Nvidia Takes Hugging Face Into its Embrace with Acquisition Deal

General-Globe-5 Green-2

This article originally appeared in Galaxy Research's weekly newsletter. Subscribe to get timely insights delivered to your inbox every Friday morning.

On Thursday, Nvidia announced it had agreed to acquire Hugging Face, often described as the “GitHub for AI,” for $12.93 billion. CEO Jensen Huang said Nvidia would “scale Hugging Face’s platform, strengthen its infrastructure and expand access to AI,” and pledged that “Hugging Face will remain an open platform for the entire AI ecosystem.” The deal is expected to close in the first half of 2027 pending regulatory approval.

Nvidia participated in Hugging Face’s 2023 Series D at a $4.5 billion valuation. Last year, however, Hugging Face rejected a further $500 million Nvidia investment at a $7 billion valuation, saying it did not want to compromise the platform’s neutrality. That stance flipped this summer after Hugging Face entertained bids amid interest from other parties, reportedly including Salesforce and Microsoft.

Hugging Face CEO Clément Delangue said open-source AI has reached an inflection point and needs more compute, support, and visibility. The founders and the team would stay on after the acquisition and remain on the Hugging Face brand.

Our take

Nvidia is evolving from the world’s leading GPU producer into a company with offerings across nearly every layer of its self-described AI “5-Layer Cake” that includes energy, chips, infrastructure, models, and applications. Hugging Face sits at the model and application level, but Nvidia recently made forays into the infrastructure level with its push to make “AI factories” repeatable and financeable, including efforts to mobilize more than $500 billion capital for compute infrastructure.

The acquisition is both defensive and offensive. OpenAI, Anthropic, Google, Meta, and Microsoft are all building or buying custom accelerator chips to reduce reliance on Nvidia GPUs. Owning the default home where open models are published helps mitigate the fallout from accelerators’ adoption. Even if frontier training moves off Nvidia chips, a large share of fine-tunes, agents, forks, and apps will still route through Hugging Face.

The offensive case is about optimization rather than lock-in. Nvidia's position no longer rests on its CUDA platform being hard to leave, because models handle much of that porting work now and Nvidia does performance engineering directly alongside the frontier labs. Hugging Face adds distribution. Nvidia can optimize a popular open model for its own runtimes the week it's published and make that the default path a developer sees, without ever requiring its chips.

The acquisition lands at a sensitive moment for open-weight adoption as U.S. policy makers consider how to regulate them. Open models now account for a larger share of AI token use on leading routing platforms like OpenRouter, with Chinese labs dominating adoption. Nvidia has been a consistent contributor and advocate of open-source AI and stepped up its efforts significantly in the past year. It is already one of the largest publishers on Hugging Face, where it shipped Nemotron and other open model families, and in August acquired a non-exclusive license to Poolside’s Model Factory. On the policy front, in July Nvidia co-signed an open letter, “Open Weights and American AI Leadership,” urging policymakers not to restrict open-weight development and use. Buying Hugging Face lets Nvidia argue that American open infrastructure, not a ban, is the right answer to Chinese dominance. It also gives Nvidia a tighter loop for improving its own open-weight stack to better compete with today’s leaders.

The biggest criticism of the deal so far is the threat to neutrality. In its 8-K filing announcing the acquisition, Nvidia’s says it will keep the platform open, “consistent with existing practices,” continue to allow model makers and users to upload and download models and datasets of their choosing, and continue to support other hardware vendors. Nvidia compute will not be required to build on or deploy through Hugging Face. Delangue’s announcement of the acquisition reinforced this view, saying Nvidia committed to keeping the platform “open, independent and compute agnostic.” Still, bringing the hub in-house still improves Nvidia’s ability to shape search, evals, recommended runtimes, and enterprise defaults in ways that benefit Nvidia offerings. Even if the terms of service never change, the commercial logic is straightforward. More accessible open models mean more fine-tuning and inference, which in turn means more demand for Nvidia chips and the software stack that makes those chips usable. -Lucas Tcheyan

You are leaving Galaxy.com

You are leaving the Galaxy website and being directed to an external third-party website that we think might be of interest to you. Third-party websites are not under the control of Galaxy, and Galaxy is not responsible for the accuracy or completeness of the contents or the proper operation of any linked site. Please note the security and privacy policies on third-party websites differ from Galaxy policies, please read third-party privacy and security policies closely. If you do not wish to continue to the third-party site, click “Cancel”. The inclusion of any linked website does not imply Galaxy’s endorsement or adoption of the statements therein and is only provided for your convenience.