I open Hugging Face the way most developers open GitHub: without thinking about it. I need a model, I search the Hub, I check the downloads, I pull the weights, I get on with my day. It is infrastructure, and infrastructure is supposed to be boring. That is exactly why the news out of Washington last week stopped me in my tracks. The US Department of Justice is now asking whether Nvidia's planned $12.93 billion acquisition of Hugging Face could harm competition in AI — and the question at the heart of it is whether one company should be allowed to own both the chips that run AI and the place where most of the world's open models live.
Let me be precise, because headlines make this sound further along than it is. Nvidia agreed to buy Hugging Face on September 3, 2026, for $12.93 billion. Reuters reported on October 28, citing two people familiar with the matter, that the Justice Department has begun asking questions about the deal. No formal challenge, second request, or lawsuit had been publicly detailed as of late October. An inquiry is not a verdict — but it is a public sign regulators see something worth examining, before the expected first-half-2027 closing.
What the Deal Actually Is
The numbers first, because they are worth getting right. Multiple reports put the total value at $12.93 billion, made up of roughly $11.9 billion paid to Hugging Face stockholders plus up to $1 billion set aside for employee equity retention. That structure matters: the retention pool is a signal that Nvidia wants to keep the people who built the Hub, not just buy the brand and the domain name. Hugging Face is not a chip company or a model lab in the frontier sense. It is a platform — the largest public repository where developers host, share, test, and download open-source AI models, datasets, and code. If you have fine-tuned a model, evaluated one, or shipped a demo with a "try it on Hugging Face" button, you have used it.
Nvidia's graphics processors power most AI training and inference. Its chief executive, Jensen Huang, said at announcement that Hugging Face would remain open to third-party processors, frameworks, and cloud services. That promise is the sentence regulators are now testing: an open platform owned by a hardware giant is only open if the owner keeps it that way, and announcement promises are not binding post-closing commitments.
What the DOJ Is Asking
According to Reuters, the Justice Department's questions centre on a phrase worth quoting directly: whether the acquisition would let Nvidia "capture a fundamental distribution point for AI." That is regulator language for a simple idea. If almost every open model passes through one hub, and that hub is owned by the dominant chipmaker, then the chipmaker gains leverage over everyone else in the chain — rival chipmakers, rival cloud providers, rival model repositories, and ultimately the developers choosing where to host their work.
Enforcers are reportedly examining three risks: leverage in commercial negotiations; harm to competing chipmakers, whose hardware could quietly become second-class on the platform — slower to support, harder to deploy, less visible; and disadvantage to alternative model repositories competing against a hub tied to the biggest chipmaker. None are findings, but they map almost exactly onto developers' worries whenever a platform they depend on gets acquired.
This inquiry is not standalone. It sits inside a broader, multi-year Justice Department examination of Nvidia's conduct, begun under the previous administration and continued under the current one, covering its acquisitions, investments, sales practices, and whether access to scarce processors was tied to networking-gear purchases. The Hugging Face questions are one thread in a larger file.
The Wider File: Run:ai, SchedMD, Groq, and France
If you want to understand why regulators are paying attention, look at the other deals in that file. The department has examined Nvidia's acquisition of Run:ai, a workload-optimisation software company, and its December 2025 purchase of SchedMD, whose open-source software helps run supercomputers, including in settings relevant to national security. Both deals closed, but both drew scrutiny over whether open tools stay genuinely open once they belong to a dominant hardware vendor. That is the pattern regulators are now applying to Hugging Face, at much larger scale.
Then there is Groq. In 2025, Nvidia struck a technology-licensing arrangement with the AI chip startup Groq and hired some of its engineers, drawing scrutiny as a possible way to gain acquisition benefits without a full merger review. The facts are contested and the review is ongoing, but the episode explains why regulators now examine a deal's structure, not just its price.
Scrutiny is not only American: Nvidia also faces a French investigation tied to CUDA dependence and cloud-provider investments. Together, this is a sustained, multi-jurisdiction question about how much of the AI stack one company should control — chips, software, networking, and now the model hub.
To be fair, Nvidia has a real argument: owning the Hub lets it invest more heavily, integrate models more tightly with its hardware and software, and make open models easier to deploy at scale. Anyone who has wrestled with driver versions and CUDA compatibility knows the gap between "model exists" and "model runs well in production" is where projects die. The counter-argument is equally real: integration that helps everyone is a feature, but integration that only helps Nvidia's hardware is a moat — a difference often invisible at first and obvious only in hindsight, which is why regulators ask before the deal closes rather than after.
What This Means for Developers Like Us
So what should you actually do? Not panic, for a start. The deal has not closed, Hugging Face is still operating as an independent platform, and nothing about your model downloads changes tomorrow. But there are a few practical habits worth adopting now, while the outcome is still uncertain, because they cost almost nothing and they protect you under every scenario.
First, treat model hosting the way you already treat cloud hosting: never assume a single provider is permanent. If your product depends on a model that only exists on one hub, keep a local copy of the weights and the licence. Disk space is cheap. Re-downloading a model you can no longer access, or whose terms have changed, is expensive. This is basic engineering hygiene, and the Hugging Face news is a good excuse to finally do it.
Second, watch the terms, not just the technology. If the deal closes with conditions, those conditions will probably show up in platform policies, API terms, and partnership announcements — not in a press release that says "we changed the deal." Pay attention to how new models are featured, which hardware gets first-class deployment support, and whether competing chips and clouds keep getting equal treatment in practice. The health of an open platform is measured in its day-to-day behaviour, not its launch promises.
Third, keep your deployment path portable. If your inference pipeline only runs well on one vendor's stack, you are exposed to that vendor's roadmap decisions regardless of who owns Hugging Face. The same abstraction-layer advice I give for model providers applies to hardware: a thin layer between your code and your accelerator makes you resilient to pricing changes, availability crunches, and platform shifts. You do not need to switch hardware to benefit from being able to switch.
Fourth, do not read the inquiry as a verdict. Inquiries can end in no action, negotiated remedies, or litigation, and Reuters noted an inquiry is not evidence of wrongdoing. If regulators act, conditions — open access, fair treatment of rival hardware, compliance reporting — are more likely than a block. But conditions are only as good as their enforcement, and enforcement depends partly on developers noticing whether they are kept.
What I Am Watching Next
Three things will tell us where this goes. The first is whether the Justice Department escalates from questions to a formal process — a second request for information, or a public statement. So far there has been neither, and both Nvidia and the department have declined to comment on the inquiry, while Hugging Face has not responded to requests for comment. Silence from all sides is normal at this stage, but it will not last forever.
The second is what happens in the first half of 2027, the expected closing window. A closing that arrives with published commitments — open access, fair treatment of rival hardware, independent governance of the Hub — would look very different from a quiet closing with only the original announcement promise to rely on. Watch for the fine print, not the headline.
The third is the developer community's response. Hugging Face's value was built by millions of developers uploading models and datasets for free. If that community starts hedging — mirroring models elsewhere, building alternative hubs — both Nvidia and regulators will notice. Trust is the one asset in this deal that cannot be bought for $12.93 billion; it has to be earned after the papers are signed, in public, over time.
Conclusion
Nvidia buying Hugging Face for $12.93 billion would join the dominant AI chipmaker to the dominant open-model hub, and the Justice Department is asking whether that combination captures a fundamental distribution point for AI. That question is not an accusation, and the deal is not yet closed. But it is the right question, asked at the right time — before the combination happens, not after. For those of us who build on open models, the practical response is not to pick a side in an antitrust case we cannot influence. It is to do the unglamorous work that makes us resilient whatever the outcome: keep local copies of the models you depend on, keep your deployment portable, and watch what the platform actually does, not just what its owner promises. Open source has survived acquisitions before, and it will survive this one too — but only if the people who use it keep paying attention to whether "open" still means what it used to.
Frequently Asked Questions
How much is Nvidia paying for Hugging Face? The deal is valued at $12.93 billion, comprising about $11.9 billion to Hugging Face stockholders plus up to $1 billion set aside for employee equity retention, according to reports. It was announced on September 3, 2026, and is expected to close in the first half of 2027, pending regulatory approvals.
Has the deal closed? No. As of late October 2026, the acquisition was still proposed. The Justice Department's inquiry is part of the regulatory review process, and no formal challenge had been publicly detailed.
What exactly is the DOJ examining? Whether owning Hugging Face would let Nvidia capture a fundamental distribution point for AI, gain leverage in commercial negotiations, harm competing chipmakers, or disadvantage alternative model repositories, according to Reuters sources. The inquiry sits within a broader, multi-year review of Nvidia's acquisitions, investments, and sales practices.
Did Nvidia promise Hugging Face will stay open? Yes. Chief executive Jensen Huang said at announcement that the platform would remain open to third-party processors, frameworks, and cloud services. Regulators are now examining whether that openness will hold in practice after the deal closes.
Does this affect my models on Hugging Face today? No immediate change. The platform continues to operate independently while the deal is pending. Still, keeping local copies of models you depend on is sensible insurance against any future change in terms or access.
What happens if regulators block or condition the deal? Possible outcomes range from no action, to a closing with binding commitments on open access and fair treatment of rivals, to a formal legal challenge. The expected closing window in the first half of 2027 gives regulators time to decide, and developers time to prepare.
