What is the Justice Department asking about Nvidia's Hugging Face deal?
The Justice Department has asked whether Nvidia's $12.93 billion Hugging Face purchase would give it control of a key AI distribution channel. Reuters reported the inquiries on 28 October 2026, citing two people familiar with the matter. Neither company has confirmed the scope of the questions, and no formal challenge to the transaction has been announced.
The questions are running alongside a long-standing Justice Department examination of Nvidia's business, the same people told Reuters. Officials have recently asked about the company's acquisitions and investments. They have also asked how it sells products, and whether it has put pressure on customers. That breadth matters because it connects the Hugging Face deal to Nvidia's wider conduct in the market.
The specific concern is distribution. Officials are examining whether owning Hugging Face would let Nvidia capture "a foundational point of distribution for AI", according to Reuters. Hugging Face is a public library where developers find, test and deploy open models. That makes it a gateway for the industry rather than a conventional software business.
Nvidia declined to comment, and Hugging Face did not respond to Reuters' request. A Justice Department spokesperson declined to discuss matters before the department, citing longstanding policy. The spokesperson said the agency would keep prioritising the "advancement of Super Intelligence".
An inquiry is not evidence of wrongdoing. Reuters said the review may not result in any action, and antitrust agencies routinely ask questions about deals they eventually clear. The deal was signed on 2 September and announced the following day. It is expected to close in the first half of 2027, subject to regulatory approvals. The reporting rests on two unnamed sources, so the existence of a formal investigation remains unconfirmed.
Why does a model hub matter to antitrust enforcers?
Hugging Face sits between the companies that build AI models and the developers who use them. The site hosts model weights, datasets, demonstration applications and documentation, and it has become the default place to compare open alternatives. That makes it closer to an app store or a search engine than to a typical software vendor.
Ownership of that choke point confers influence over discovery. A platform can decide how models are ranked, which formats are promoted, and which tools are easiest to try first. Those look like product decisions, but they also determine which model builders get attention and which fade. For a company selling the hardware that runs those models, that influence has obvious strategic value.
Naveen Chhabra, a principal analyst at Forrester, told the Guardian that ownership would give Nvidia unusual insight. "Nvidia gains visibility into customer's preferences and the AI models they use," he said. He cited "which models are trending, what datasets customers are downloading". That information arrives before public benchmarks and news coverage catch up.
The concern is not only advertising or subscription revenue. Early knowledge of which architectures and workloads are gaining ground can inform chip design, software optimisation and sales priorities. A platform owner can also make its own hardware the path of least resistance, without formally excluding rivals. Regulators tend to look for those softer forms of favouritism because they are hard to detect after the fact.
Open-weight models have become a serious alternative to closed APIs, especially for companies trying to control costs. Chinese labs such as DeepSeek, Moonshot and Z.ai have pushed that market, as the Guardian reported. A US-owned hub for those models therefore sits at the intersection of competition policy and AI policy. Nvidia says developers will keep choosing their own models, clouds and inference providers. It says its chips will not be required to use the platform.
How big is Hugging Face, and what is Nvidia actually buying?
Hugging Face was founded in 2016 by Clément Delangue, Julien Chaumond and Thomas Wolf, and is based in New York. Its backers include Intel, AMD, Qualcomm, Google and Amazon, according to reporting on its 2023 funding round. Nvidia itself was among those investors before it became the buyer, which gives the deal an awkward symmetry for rival chipmakers.
Nvidia puts the community at more than 18 million developers, researchers and creators. They share more than 3 million models, 500,000 datasets and 1 million applications. More than 200,000 companies use the platform to discover, evaluate, customise and deploy AI, according to Nvidia's announcement. Those numbers explain why the platform is treated as infrastructure rather than a single product.
The revenue is modest by Nvidia's standards. Reuters has reported annualised revenue of about $150 million, which implies a price of roughly 86 times sales on the announced figure. That multiple reflects strategic value rather than current earnings power. Nvidia is paying for position in the open-model ecosystem, not for a large profit stream today.
The consideration is split between cash for shareholders and retention for staff. Nvidia's regulatory filing describes approximately $11.9 billion payable to Hugging Face stockholders. It also provides an equity-based retention programme of up to about $1 billion for employees joining Nvidia. The filing also states the expected closing window and the customary conditions attached to it.
Reuters described the purchase as Nvidia's largest conventional acquisition, topping its roughly $6.9 billion purchase of Mellanox, which closed in 2020. The comparison excludes arrangements structured as licences rather than takeovers, which have become a separate focus for regulators. Hugging Face had also turned down a $500 million investment from Nvidia the year before, according to Observer. It had not raised outside money since a $235 million round led by Salesforce in 2023.
- •Price: $12,930,300,000, as announced by Nvidia.
- •Cash to stockholders: about $11.9 billion, plus up to about $1 billion in retention equity.
- •Scale: more than 18 million users, 3 million models and 500,000 datasets.
- •Timing: definitive agreement signed on 2 September 2026; closing expected in the first half of 2027.
What has Nvidia promised about keeping Hugging Face open?
Nvidia has made openness central to its public case. In the blog post announcing the deal, Jensen Huang made the promise plainly. He wrote that "Hugging Face will remain an open platform for the entire AI ecosystem". He said developers would choose their models, frameworks, clouds, inference providers and computing platforms. The post also commits the platform to supporting open and open-weight models from every model builder.
The same post states that "NVIDIA compute will not be required to build on or deploy through Hugging Face". It promises continued support for multi-cloud and multi-accelerator development, so builders can use the hardware that fits their work. Those are broad commitments, and they address the most obvious antitrust concern directly.
The regulatory filing repeats the theme in more formal language. It says Nvidia has committed to keep the platform open. The company also says it will let users upload and download models and datasets of their choosing, and support other silicon vendors. The wording tracks Hugging Face's existing practices, which makes the promise easier to keep in the short term.
The pledges are voluntary, not conditions imposed by an agency. In past technology mergers, regulators have sometimes turned similar assurances into binding commitments, including reporting obligations and independent monitors. That is one likely path if officials decide the transaction can proceed. It would also give competitors and developers a formal way to raise concerns.
Nvidia's existing role makes the promises harder to assess. The company says it is the largest contributor of open models and data to Hugging Face. It has published more than 500 models and more than 250 open datasets there. Being the biggest supplier and the owner of the marketplace is the combination officials tend to examine. Nvidia argues that open models expand access rather than concentrate it, and that its contributions show a long-standing commitment.
Does Nvidia's antitrust history shape this review?
Nvidia has fought a major merger battle before. It agreed in September 2020 to buy Arm from SoftBank. The FTC sued in December 2021 to block the $40 billion transaction, Reuters noted. The companies terminated the agreement in February 2022. Nvidia's announcement cited "significant regulatory challenges" as the reason. SoftBank kept a $1.25 billion prepayment, and Nvidia retained a 20-year Arm licence.
More recent transactions have drawn attention for their structure. Bloomberg reported in September that the Justice Department was examining Nvidia's $20 billion licensing agreement with Groq. Officials were looking at whether it was arranged to avoid premerger review. Nvidia obtained rights to Groq's technology under that arrangement. Licensing and hiring deals usually fall outside the Hart-Scott-Rodino rules, while an outright acquisition of this size requires a filing.
That filing gives the FTC and the Justice Department a formal window to review the transaction before it closes. Under those rules, the initial waiting period is usually 30 days, according to the Congressional Research Service. A second request for information extends the review, and most notified deals are cleared without one. The system is designed to catch problems before a merger is completed, not after.
The broader examination has also touched Nvidia's December 2025 purchase of SchedMD. That company develops the open-source Slurm scheduler, which is used on many of the world's fastest supercomputers, including systems with national-security relevance. Reuters reported that the deal raised concerns among AI specialists about continued access to the software. Nvidia pledged to keep it open and supported.
Together, those episodes describe a pattern that regulators care about. Nvidia has been buying companies that sit underneath the AI stack, from networking to scheduling to model distribution. Each purchase adds a layer that competitors and customers depend on. That pattern is why the Hugging Face questions matter beyond the price of one acquisition.
Who is affected if the deal clears or stalls?
Developers are the most immediate group affected. A change in how Hugging Face ranks models, prices enterprise features or supports rival accelerators would alter daily workflows. Teams that depend on it would have to adapt. Switching costs are real, even when alternatives exist, because documentation, tooling and community knowledge accumulate around one platform.
Chipmakers that backed Hugging Face have an obvious interest. Intel, AMD, Qualcomm, Google and Amazon were among its investors. Their hardware would compete with Nvidia's on a platform Nvidia would own. A neutral catalogue is more valuable to them than a curated one. Any formal commitment on interoperability would matter most to this group.
Enterprises use the hub to evaluate models before committing to a vendor. Procurement teams may want assurances that the catalogue will not quietly favour Nvidia-optimised builds or exclude competitors. Compliance teams will also watch how model licences, data and security disclosures are handled. For regulated industries, platform ownership is a supply-chain question, not just a price question.
Model builders, including Chinese labs, rely on the hub for distribution. Nvidia's filing warns that restrictions on models from any region, including China, could materially affect the platform. That risk connects the merger to a wider debate about open weights, export policy and AI sovereignty. It also ties the review to security policy, because restrictions on models from certain regions could reshape the catalogue.
For Nvidia, the deal is partly a hedge. Its largest customers are designing their own chips. Software and services could become a more durable source of revenue as hardware competition grows. Hugging Face gives it a direct relationship with millions of developers. For Hugging Face staff, the filing provides retention equity, and the founders have said they will join Nvidia after closing. Their influence over platform policy will depend on how much autonomy Nvidia grants.
What happens next, and what should you watch?
The first milestone is the premerger notification and the review that follows. A second request, a negotiated settlement or an unconditional clearance would each send a different signal about the deal's timetable. None has been announced yet. The Justice Department's questions may also end without a formal case; Reuters said the review may not lead to action.
Watch whether Nvidia offers formal commitments on model ranking, interoperability and support for rival chips. Voluntary blog promises are easy to make, while consent agreements are enforceable and public. Regulators often prefer the latter for platform deals because they give rivals a way to complain when behaviour changes. The scope of any remedy would reveal how much the agencies believe the platform matters.
Watch the closing date. Nvidia still guides to the first half of 2027. Any slip would suggest the review is taking longer than the companies expected. A delayed close can also change the commercial case as the AI market shifts.
Watch the wider policy debate. The Justice Department spokesperson's reference to "Super Intelligence" signals a framing. Enforcement priorities are being tied to leadership in advanced AI, not only consumer prices. That framing could apply to future deals as well. It also links antitrust review to export controls and the argument over open-weight models.
None of this means the transaction will be blocked. Reuters reported that the inquiry may end without action, and most mergers reviewed under the notification system close without a challenge. The questions are a signal of interest, not a verdict. But control of AI distribution is now part of the antitrust conversation. That shift has consequences beyond this one transaction, for platform owners, chipmakers and the developers in between.