Nvidia agreed to acquire Hugging Face, the open-model hub used by more than 18 million developers, for $12,930,300,000, a deal Nvidia confirmed on September 3, 2026 and expects to close in the first half of 2027. The acquisition matters because it is the first major AI software transaction Nvidia cannot structure as a technology license to sidestep mandatory merger review, and because it puts the world's dominant chip supplier in charge of the distribution layer for open-weight models. If the deal closes, one company will own both the accelerators that train and serve most models and the catalog where those models are discovered, downloaded and deployed.
What Nvidia actually bought
Hugging Face, founded in 2016 by Clément Delangue, Julien Chaumond and Thomas Wolf, is often described as the GitHub of AI. In Jensen Huang's confirmation post, Nvidia put concrete numbers on the platform: more than 18 million developers, researchers and creators; over 3 million models; 500,000 datasets; 1 million applications; and more than 200,000 companies using it to discover, evaluate, customize and deploy AI. The BBC reported the same figures and noted that the deal would bring one of the world's largest AI developer communities in-house.
The strategic asset is not the revenue. Hugging Face was clocking roughly $150 million in annualized revenue, according to The Information, and had raised just over $395 million in total funding, with its last round in 2023 valuing it at $4.5 billion. Nvidia is paying about 86 times annualized revenue, a multiple that only makes sense if you price what the platform controls rather than what it earns.
The chokepoint is from_pretrained()
What Hugging Face truly owns is a workflow convention embedded in production code across the industry. The Transformers library's from_pretrained() call pulls model weights from the Hub into pipelines, and it sits inside hundreds of thousands of codebases. That makes the Hub hard to leave in a way that a generic API router is not: developers already know where to look, and their tooling assumes it.
The New Stack's Amanda Caswell framed the risk precisely: buying Hugging Face would not give Nvidia control over everything on the platform, since libraries such as Transformers and Diffusers are open source and models stay under their own licenses, but it would mean buying "a place developers already turn to when they want to work with open models, including models that don't have to run on Nvidia hardware." The community, that piece argued, resists easy forking even when the code can be cloned.
From chip vendor to full-stack owner
The Hugging Face deal is the latest step in Nvidia's transformation from a component supplier into a platform company. CNBC called it Nvidia's second-biggest acquisition on record, behind the roughly $20 billion purchase of assets from chipmaker Groq in December, and ahead of the nearly $7 billion acquisition of Israeli chipmaker Mellanox in 2019.
The pattern around it is consistent. Nvidia has committed up to $100 billion to OpenAI, taken a $5 billion stake in Intel, and struck multi-billion-dollar arrangements with coding startup Poolside and networking silicon firm Enfabrica. TechCrunch noted a second motive beyond developer mindshare: Nvidia can package unused compute capacity to enterprise customers alongside Hugging Face's offering. Owning the catalog also lets Nvidia tune a platform around its own chips, even as it promises not to require them.
The open-weights paradox
Huang's post insists Hugging Face "will remain an open platform for the entire AI ecosystem," that developers will choose their models, frameworks, clouds and compute platforms, and that "NVIDIA compute will not be required to build on or deploy through Hugging Face." Nvidia also positions itself as the largest single contributor of open models and datasets to the Hub, with more than 500 models and 250 open datasets released there.
The paradox is structural. Nvidia argues open weights broaden access to AI, and Huang told CNBC that open models give defenders an "asymmetric advantage" because "there are way more people who are protecting than there are people who are attacking." Delangue told CNBC he approached Huang over the summer, reasoning that open-source AI "needed more, more resources, more scale, more visibility." Yet the same deal concentrates the discovery layer for open models under the vendor whose hardware monetizes their deployment. The New Stack put the tension bluntly: the announcement is the moment an open ecosystem becomes another Nvidia platform.
The antitrust review Nvidia cannot structure around
Here the deal differs from Nvidia's recent AI moves. TechTimes reported that Nvidia spent roughly $27 billion across three transactions in nine months, each styled as a technology license plus talent transfer rather than an outright acquisition: the $20 billion Groq license, an approximately $900 million arrangement with Enfabrica, and a $7 billion license and equity stake in Poolside. That structure matters legally, because characterizing a deal as a license and investment can avoid triggering Hart-Scott-Rodino premerger notification, which applies above a threshold near $119 million.
A direct $12.9 billion acquisition has no such escape hatch. It requires HSR filing with the FTC and the Department of Justice and a waiting period before closing. The earlier license structures had already drawn political heat: Senators Elizabeth Warren and Richard Blumenthal wrote to Huang in March 2026 questioning whether the Groq deal was "structured to evade scrutiny by antitrust regulators," and FTC Chair Andrew Ferguson said in January 2026 the agency was examining whether acqui-hires were "being constructed to try to escape Hart-Scott-Rodino review."
Regulators will likely focus on a vertical relationship: the leading accelerator supplier owning the leading open-model distribution layer. TechTimes flagged three specific mechanisms. First, self-preferencing, meaning Nvidia could advantage CUDA-optimized or Nemotron-family models in Hub search rankings or featured placements. Second, investment redirection, meaning Nvidia could deprioritize maintenance of the Optimum AMD and Optimum Intel libraries that today give developers non-Nvidia paths to run Transformers and Diffusers models on AMD ROCm hardware and Intel Gaudi accelerators. Third, informational advantage, meaning Nvidia could see which models are gaining adoption before AMD, Intel and rival clouds do, and act on that knowledge in GPU product and pricing decisions. In the EU, a $12.9 billion deal almost certainly clears turnover thresholds, triggering a standard review rather than the Article 22(3) referral route Nvidia contested after its 2024 Run:ai acquisition.
What it means for developers and the open-model market
For builders, the practical advice is to assess exposure before the deal closes. Auditing which production models are available only through the Hub, checking self-hosting licenses, and documenting where from_pretrained() calls would need rewrites are all cheap now and expensive later. Migrating a large pipeline off the Hub is a project measured in months, not days.
The larger ecosystem question is whether a neutral alternative emerges. Openly licensed models can still be hosted elsewhere and the underlying libraries can be forked, but no other platform matches Hugging Face's combination of community, dataset hosting and library integration in one place. That is exactly why neutrality commitments matter: there is no obvious second Hugging Face for the industry to flee to if trust in the first erodes.
Timeline and what to watch
Nvidia and Hugging Face signed the agreement on September 2 and confirmed it on September 3, 2026. Nvidia will pay about $11.9 billion to Hugging Face investors and offer up to $1 billion in stock-based incentives to employees who join the company, per the BBC. The deal is expected to close in the first half of 2027, pending regulatory approval. MLex reported that Nvidia had not yet settled whether it must file in Europe, and that the platform's native France may not be among the jurisdictions with review authority.
Three things to watch. Whether regulators accept behavioral neutrality commitments or demand structural conditions for the Optimum libraries and Hub ranking. Whether Nvidia's open-weights pledges survive the first incentive to favor its own models and silicon on the platform. And whether Hugging Face's developer community treats the acquisition as funding for open AI or as the closing of the open-model era's most neutral discovery layer. The answer will shape how the next generation of open-weight models reaches production.