US Justice Department questions Nvidia's $12.93 billion Hugging Face deal

When Nvidia announced in early September that it would buy Hugging Face, the reaction in the open source AI world was a mix of excitement and worry. Excitement, because the platform that hosts most of the world's open models would suddenly have the deepest pockets in the industry behind it. Worry, because the company that sells most of the chips those models run on would also own the place where developers find them. Now the US government is asking the same question.

According to an exclusive Reuters report published on October 28, 2026, the US Department of Justice has made inquiries about how Nvidia's $12.93 billion deal for Hugging Face would affect competition in artificial intelligence. Two people familiar with the matter told Reuters that officials asked whether the deal would let Nvidia "capture a fundamental distribution point for AI". CNBC later confirmed the inquiry with its own source.

Here is what is known, what the Justice Department is reportedly looking at, why Hugging Face is such a sensitive asset and what could happen next.

The deal in short

Nvidia signed a definitive agreement to acquire Hugging Face on September 2, 2026, and Jensen Huang announced it the next day in a post on the Nvidia blog. The headline number was precise: $12,930,300,000. According to Nvidia's filing with the SEC, about $11.9 billion goes to Hugging Face stockholders and up to about $1.0 billion is set aside as equity based retention for Hugging Face employees who join Nvidia. The transaction is expected to close in the first half of 2027, "subject to the satisfaction or waiver of customary closing conditions, including receipt of required regulatory approvals".

Huang described Hugging Face as "a vibrant home for the open model developer community". The numbers he cited explain why: more than 18 million developers, researchers and creators use the platform to share more than 3 million models, 500,000 datasets and 1 million applications, and more than 200,000 companies use it to find, test and deploy AI. Nvidia itself is the largest contributor of open models and data on the site, with more than 500 models and 250 datasets.

From the first day, Nvidia tried to answer the obvious concern. Huang wrote that Hugging Face "will remain an open platform for the entire AI ecosystem" and that "NVIDIA compute will not be required to build on or deploy through Hugging Face". The SEC filing goes a step further and describes a formal commitment to keep the platform open, consistent with its existing practices: model makers, developers and users can keep uploading and downloading the models and datasets they choose, and Hugging Face keeps supporting other silicon vendors.

What the Justice Department is asking

The Reuters report, written by Max A. Cherney, Jeffrey Dastin and Jody Godoy, describes questions on two levels. The first is about Hugging Face itself. Officials want to understand how Nvidia might use the platform to harm rivals in the AI market and whether the newly acquired business could give Nvidia leverage over others. They are also asking about the platforms that compete with Hugging Face, which is a normal step when an agency tries to define the market a deal touches.

The second level is much broader. The questions about Hugging Face sit inside a long running investigation into Nvidia's conduct that began under President Joe Biden and continued under President Donald Trump. According to the two sources, it covers Nvidia's acquisitions, investments and sales practices, and whether the company has put pressure on its customers. Specific threads named in the report include:

• The Groq deal. At the end of 2025 Nvidia paid about $20 billion to license technology from chip startup Groq and hire its engineers. Officials have explored whether Nvidia offered customers larger allocations of its scarce AI chips if they also bought its networking equipment and Groq processors. That would raise the classic antitrust question of whether a dominant product was used to push into a newer market.
• Run:ai. The DOJ has looked at Nvidia's purchase of Run:ai, an Israeli company whose software helps AI chips in a cluster run more efficiently.
• SchedMD. Nvidia bought SchedMD, the company behind the Slurm workload manager, in December 2025. Slurm is open source software that schedules jobs on many of the world's supercomputers, including systems relevant to national security, and the purchase made some AI specialists worry about future access. Nvidia has promised to keep it open and supported.

It is important to be precise about what this is and what it is not. The report is based on anonymous sources. Neither Nvidia nor the Justice Department confirmed details, Nvidia declined to comment and Hugging Face did not respond to Reuters. A DOJ spokesperson said the department does not comment on matters before it and added that "it would be inappropriate for the Department to conduct a merger investigation under the guise of a conduct investigation". As Reuters notes, government investigations are not evidence of wrongdoing and often end without any lawsuit.

Why Hugging Face is different from a normal acquisition

Nvidia has bought many companies, but Hugging Face is not a chip startup or a niche software tool. It is closer to what GitHub is for code. CNBC calls GitHub, which Microsoft bought for $7.5 billion in 2018, the closest comparable business. Hugging Face is where a new open model usually appears first, where developers compare model cards, download weights, try demos in Spaces and pick the version that fits their hardware.

That position matters in several quiet ways that do not depend on blocking anyone:

• Defaults. When a model page shows how to run a model, which hardware, inference engine or cloud gets listed first can steer millions of downloads.
• Formats and optimizations. Hugging Face decides which quantization formats, libraries and integrations get first class support. If Nvidia's formats always arrive on day one and a rival chip maker's support shows up weeks later, nobody has been excluded, but the playing field has tilted.
• Data. A platform at this scale sees which models are rising, which companies are testing what and where demand is heading. For a company that also sells the hardware and invests in many of the model makers, that is valuable insight into the market.
• Hosted services. Hugging Face runs Inference Endpoints and routes requests to outside inference providers. The way those services rank or recommend backends could favor one ecosystem.

The Justice Department's phrase "a fundamental distribution point for AI" captures exactly this. The worry is less that Nvidia would suddenly ban AMD or Google chips, which Nvidia has explicitly ruled out, and more that ownership of the shelf where open models are displayed gives a powerful hardware vendor many small levers.

There is also a second angle. Hugging Face is one of the few places where rival chip makers, cloud providers and model labs all meet as partners. AMD, Intel, Google, AWS and Microsoft all work with it on optimizations and hosting. After the deal, those partners would be sharing roadmaps and early access with a subsidiary of their biggest competitor. Even with formal firewalls, some may decide to invest less, which could weaken the very openness Nvidia promises to protect.

Nvidia's side of the argument

Nvidia's case is straightforward, and it is not weak. Hugging Face hosts other companies' models, and Nvidia does not own those models. The platform has many alternatives for downloading weights, from GitHub and cloud model catalogs to sites like ModelScope, and any lab can publish on its own servers. Nvidia also argues that running a platform used by millions of people costs a lot of money and that its resources can improve reliability, safety, model evaluation and deployment tools.

Huang tied the deal to an open letter he coauthored on the value of open weights, arguing that open models "broaden access to AI and help ensure that AI leadership is distributed across companies, institutions and communities". From this angle, Nvidia is not trying to close the open ecosystem but to make sure it keeps growing, because more open models running on more servers means more demand for GPUs. The company also says Hugging Face founder Clem Delangue approached Huang about the deal, not the other way round.

The binding commitment in the SEC filing is a signal that Nvidia expected these questions. It is the kind of promise that regulators can later turn into formal remedies, for example a consent decree that requires equal treatment of hardware vendors or limits on how platform data can be shared with the rest of Nvidia.

The bigger picture: Nvidia's growing reach

This inquiry does not happen in a vacuum. Over the past two years Nvidia has moved far beyond selling GPUs. It has bought software companies, licensed technology from potential chip rivals, invested in leading model labs such as OpenAI and Anthropic and backed cloud providers that buy its chips. We have covered some of these moves on djcroman, from Nvidia's interest in Reflection AI to its late attempt to bid for OpenRouter, the model router that Stripe won in an 8 billion dollar deal.

Each deal on its own can look reasonable. Together they show a company trying to sit at every layer of the AI stack: hardware, networking, cluster software, scheduling, inference, model investment and now model distribution. That pattern is exactly what antitrust agencies tend to examine over years, not weeks.

The political context adds another layer. Jensen Huang has built a close relationship with President Trump and received the National Medal of Science from him, and Trump has disclosed that he owns Nvidia stock. Reuters writes that there is no indication Trump has any involvement in the probe. Staff led investigations at the DOJ often run for years and across administrations, and the department's spokesperson framed its mission around advancing "Super Intelligence for the betterment of the American people", which hints that the current leadership is wary of slowing down American AI companies.

Nvidia also faces questions in Europe. France's competition authority has spent years investigating the company over alleged anticompetitive practices, with concerns that include the industry's dependence on Nvidia's CUDA software and its stakes in cloud providers such as CoreWeave. Reuters reported in July 2026 that this investigation was nearing its end. Hugging Face was founded by French entrepreneurs and has a large team in Paris, which makes European interest in the deal likely as well.

What happens next

For now, there is no formal challenge and no public timeline. The likely paths look like this:

1. Quiet clearance. The DOJ collects information, Nvidia answers questions and the deal closes as planned in the first half of 2027. 2. Clearance with conditions. Regulators accept the deal but turn Nvidia's openness promises into enforceable rules, for example on hardware neutrality, equal access for inference providers and separation of platform data. 3. A longer fight. The DOJ opens a deeper review or sues to block the deal, which would push the closing date back and could make Nvidia walk away.

For developers, nothing changes today. Hugging Face still works the same way, models from every lab are still there and Nvidia hardware is still not required. What is worth watching is whether the conditions of the deal become public and whether rival chip makers and cloud providers speak up during the review.

Why this matters

Open models are one of the main counterweights to a few closed AI labs. Their value depends on a neutral place where anyone can publish, compare and download them. Hugging Face became that place partly because it was independent. If the most powerful company in AI hardware owns it, the openness has to be guaranteed by promises and regulators instead of by the structure of the market.

The Justice Department's questions do not mean the deal is in trouble. They do show that regulators see the same thing many developers saw in September: owning the shelf matters almost as much as owning the products on it.

Sources: Reuters via Investing.com, CNBC, The Decoder, Nvidia blog, Nvidia SEC filing.