Nvidia buys Hugging Face for $12.93 billion, and the DOJ is taking a close look

What's happening

Nvidia, the world's most valuable company by market cap, is buying Hugging Face. The price is exactly $12,930,300,000, or roughly $12.93 billion. Hugging Face is something like the GitHub of AI. Research labs, companies and hobbyists upload their models, datasets and small demo apps there so others can download and build on them. If you want to try an open model like Llama, Qwen, Mistral or DeepSeek locally today, you almost certainly get it from Hugging Face.

Nvidia CEO Jensen Huang announced the deal in a blog post. According to Bloomberg, the acquisition was first announced in September. Pulse2 reports that early details leaked in late August, and The Information had reported on the talks beforehand. Now the US Department of Justice (DOJ) has stepped in: according to Reuters, it is asking questions about how the acquisition will affect competition in the AI market.

The numbers behind the deal

According to Quiver Quantitative, about $11.9 billion of the purchase price goes to Hugging Face's existing investors. Ironically, those include Nvidia rivals Intel and AMD, along with cloud giant Amazon. Up to $1 billion is earmarked for stock-based retention awards: shares that employees receive gradually as long as they stay at Nvidia. The goal is to keep the team from walking out after the acquisition.

The Hindustan Times says this is Nvidia's second-largest transaction ever. The company only spent more in December 2025, when it paid $20 billion for assets of AI chip startup Groq. Before that, the record was its purchase of Israeli networking specialist Mellanox for just under $7 billion.

Nvidia itself has long been a heavy user of the platform. The company says it has published more than 500 models and over 250 open datasets there, plus libraries and developer tools.

What Nvidia is promising

The community's concern is obvious. If the biggest AI chipmaker owns the most important platform for open models, it could tailor that platform to its own hardware. Huang pushes back on this explicitly in his blog post:

"Hugging Face will remain an open platform for the entire AI ecosystem. Developers will choose the models, frameworks, clouds, inference providers and compute platforms they want. Nvidia compute will not be required to build on or deploy through Hugging Face."

Specifically, Nvidia commits to Hugging Face continuing to support open models from every vendor. That covers open-source models, where code and training details are public, as well as open-weight models, where only the finished model is available for download. Multi-cloud and multi-accelerator support is also meant to stay. In other words, models should keep running on different cloud providers and on chips from different vendors, such as AMD or Intel. Inference, meaning running a finished model, is supposed to remain possible anywhere.

Hugging Face's leadership and team are moving to Nvidia, and the Hugging Face brand will remain. Huang names co-founders Clément Delangue ("Clem"), Julien Chaumond and Thomas Wolf and praises them for building "a vibrant home for the open model community." He plans to use Nvidia's infrastructure and engineering teams to improve reliability, security, model evaluation, inference and deployment.

Why Hugging Face is selling

According to the Hindustan Times, citing a CNBC interview, the initiative came from Hugging Face itself. CEO Clément Delangue reached out to Huang over the summer, and the talks moved very quickly after that. Delangue said Hugging Face and open AI had reached a point where the company needed more resources, more scale and more visibility, and that Nvidia was the "perfect home" for it.

There's also a security incident. Hugging Face was recently hacked. Delangue blames the attack on mistakes in the company's own development process. To fend it off, Hugging Face used an Nvidia version of a Chinese open model. For Delangue, the incident shows why open models matter. Huang makes a similar argument: an open environment gives defenders an "asymmetric advantage," because there are potentially far more people protecting systems than attacking them. The sources don't say what data was affected by the hack.

Antitrust regulators take notice

According to Reuters (cited by Bloomberg), the DOJ is investigating whether the acquisition would give Nvidia control of a central distribution hub for AI. Separately, an investigation into Nvidia's business practices has been running for years. Nvidia declined to comment, and Hugging Face did not initially respond to requests. The DOJ doesn't comment on ongoing matters as a rule, but said that under President Trump it will continue to prioritize the development of "Super Intelligence" for the benefit of Americans. That's the administration's preferred term for AI.

The scrutiny didn't come out of nowhere. Last month, Bloomberg reported that the DOJ is examining the $20 billion licensing deal with Groq, specifically whether Nvidia deliberately structured it to avoid triggering an antitrust review. Groq formally remained independent, but its CEO and COO moved to Nvidia. Some US lawmakers called it a stealth acquisition. Bloomberg also reports that Nvidia is in talks with Reflection AI about a deal. The startup is already funded by Nvidia and released an open-weight model just this month.

Our take

If you work with open AI models, this is one of the biggest industry stories of the year. Until now, Hugging Face was neutral ground where Meta, Google, Alibaba, DeepSeek, Mistral and thousands of small teams offered their models side by side. Going forward, that ground belongs to the company that already supplies most of the world's AI hardware. Nvidia is extending its reach across the entire stack: chips, networking, its own models, and now the platform through which models are distributed.

The openness pledges are far-reaching, but they are promises from the buyer. It's not yet known whether there are oversight mechanisms or contractual guarantees, for example toward regulators. It's also notable that rivals like AMD and Intel profit from the sale as investors. For you as a user, nothing changes for now: you can still download models and run them on any hardware. The key question is whether Nvidia-optimized formats, tools and inference services will gradually get preferential treatment on the platform. That is likely exactly what the US antitrust review wants to find out. It's also unclear when the deal will close and whether regulators in other countries will review it too.