Two people familiar say Justice Department officials asked whether the $12.93bn Hugging Face deal lets Nvidia capture a fundamental distribution point for AI. Nvidia has promised to keep the platform open; the department has already examined its December 2025 purchase of SchedMD.
Nvidia's $12.93bn purchase of Hugging Face now faces a long antitrust review, and the likely price of approval is binding terms on access, not cash. Justice Department officials have asked whether the deal would let the chip designer capture a fundamental distribution point for AI.
Approval could restrict Nvidia's freedom to operate the platform; the purchase is small relative to its finances. The $12.93bn is 13% of revenue in the quarter to 26 July, which was $96.2bn. Rivals such as Broadcom, Google and AMD have the most to gain or lose from the outcome.
Reuters reported the questions on 28 October, citing two people familiar. Nvidia expects the deal to close in the first half of 2027. It says Hugging Face would keep supporting a range of chip, cloud and AI providers.
The Hugging Face questions sit inside a wider inquiry into Nvidia's conduct. The two people said recent questioning covers its acquisitions, investments, sales practices and whether it has pressured customers. That examination began under President Biden and has continued under President Trump.
Reuters notes that investigations are not evidence of wrongdoing and may not end in any action. Nvidia declined to comment, and Hugging Face did not respond.
The question concerns control over model distribution. Hugging Face is where developers find and fetch open AI models. The issue is whether a hardware seller can tilt the path to them.
Nvidia has answered with promises. The department has already examined its December 2025 purchase of SchedMD, which carried a similar promise.
What Nvidia would control
The platform is large. Nvidia's announcement post says more than 18 million developers, researchers and creators use it. They share more than 3 million models, 500,000 datasets and 1 million applications. More than 200,000 companies use the platform.
Nvidia already benefits from the site. It calls itself the largest contributor of open models and data, with more than 500 models and 250 open datasets. The purchase moves ownership of the access layer.
The commitments sit in the company's 8-K filing on the 2 September agreement. Nvidia will keep the platform open. Model makers, developers and users can upload and download models and datasets of their choosing. Hugging Face will support other silicon vendors.
The announcement post adds that Nvidia compute will not be required to build on or deploy through Hugging Face. A promise in an 8-K binds no one but Nvidia until a regulator writes it into an order.
Earlier acquisitions under scrutiny
The department has been here before. It has examined Nvidia's December 2025 purchase of SchedMD, according to one of the people. SchedMD develops Slurm, the job scheduler used in more than half of the top 10 and top 100 systems in the TOP500 list of supercomputers.
Nvidia's announcement promised to keep Slurm open source and vendor neutral. The report says the purchase worried AI specialists about their access to it.
Investigators have also looked at Nvidia's 2025 Groq licence and engineer hiring deal, and at its Run:ai acquisition. One question was whether Nvidia offered larger chip allocations to customers that bought its networking equipment and Groq processors. Officials are also asking about Hugging Face's competitors.
The questions may be routine, because the department seeks company input on proposed mergers and staff led investigations often run for years. A department spokesperson said it would be "inappropriate for the Department to conduct a merger investigation under the guise of a conduct investigation".
The Hugging Face questions are specific, though. They ask how Nvidia could use the platform to harm rivals and whether it could gain leverage from it. Those are merger questions, and the 8-K makes receipt of required regulatory approvals a closing condition.
Rivals have the most at stake. The report names two groups of challengers to Nvidia's lead in AI chips. Broadcom and Google design processors together. AMD has struck billion dollar deals with AI labs and cloud operators.
Each sells hardware that runs models developers fetch from sites like Hugging Face. Ranking, defaults and support are where an owner could favour its own chips.
The cost of approval
Nvidia can pay the price without strain. The $12.93bn is 17% of first half operating cash flow ($74.4bn) and 23% of cash and marketable debt securities ($56.6bn).
The total is about $11.9bn payable to Hugging Face stockholders plus up to $1.0bn of equity based retention pay for employees joining Nvidia.
A remedy would restrict how Nvidia operates the platform rather than require a cash payment. A single platform cannot be divided the way a product line can, so the remedy is one of two things. It is either binding terms on access, defaults and support for other chips, or abandonment.
Nvidia has met that choice before. The FTC sued to block its $40bn Arm purchase in December 2021. Nvidia and SoftBank ended the deal on 7 February 2022, citing "significant regulatory challenges".
Nvidia can afford the first outcome, and it has already offered the substance of it as a promise.
Other risks and the closing timetable
Beijing can also decide what Hugging Face hosts. The same 8-K warns that many of the most popular open models originated in China. It adds that any restriction on supporting models from any region could materially hurt the platform.
Terms invite supervision after closing, as Nvidia knows from Mellanox. China's antitrust regulators published a preliminary finding on 15 September 2025, according to the 10-Q. It said Nvidia had violated the terms of their approval of that acquisition. Binding terms on Hugging Face would invite the same kind of continuing scrutiny.
The best objection is that open weights are portable. Models can be mirrored, and Nvidia says its compute is optional. Copying a file is easy.
Replacing the place where a developer expects to find 3 million models, 500,000 datasets and 1 million applications is not.
Nvidia's own timetable already prices a long review. It signed on 2 September and told investors the deal would close in the first half of 2027, up to ten months later. The 30 day Hart-Scott-Rodino waiting period alone is far shorter.
Nvidia reports fiscal third quarter results on 17 November at 5:00 pm Eastern. It is the company's first report since the Justice Department's questions became public. Expect it to hold the first half of 2027 closing date, because approval, not price, now sets the pace.