NVIDIA wrote the reason for buying Hugging Face into the document that announces the purchase. The Form 8-K filed with the SEC says demand for open-source models “promotes the use of our products worldwide and sustains the Hugging Face platform”: the company that sells the accelerators is buying the hub through which open weights reach developers, and its own filing explains why it needs that layer healthy. The definitive agreement dates to September 2, 2026, announced the next day at exactly $12,930,300,000; the deal has not closed, and what is signed, pending and merely promised are three states of the same transaction.
The signed part: $11.9 billion to owners, up to $1.0 billion to employees
The 8-K splits the announced price: roughly $11.9 billion payable to Hugging Face stockholders subject to certain adjustments, plus an equity-based retention program of up to approximately $1.0 billion for employees who join NVIDIA. The filing attaches no purchase agreement, so that structure is the deal’s only detail in the public record. The blog post by CEO Jensen Huang adds the rest, in NVIDIA’s words: the team keeps “their same iconic 🤗 brand,” and Delangue “came to me as he considered the next chapter of Hugging Face and believed NVIDIA would be a great home.” The same post puts the platform at 18+ million developers, 3+ million models, 500,000 datasets and 200,000+ companies, and claims NVIDIA is already the hub’s largest contributor of open models and data, with 500+ models and 250+ datasets. All NVIDIA’s own figures, without method or third-party measurement.
The pending part: a 2027 close and unnamed approvals
NVIDIA expects the deal to close in the first half of 2027, subject to “the satisfaction or waiver of customary closing conditions, including receipt of required regulatory approvals.” Neither primary source says which approvals, from whom, or whether any filing has been made; the record holds no statement from the FTC, the European Commission or any other regulator. The 8-K is silent, too, on financing, break-up fee, termination rights or outside date. Antitrust scrutiny is, so far, a press expectation with no filing on the public record.
The promised part: openness with no mechanism
The 8-K records that “NVIDIA has committed to, among other things, keep Hugging Face’s platform open, consistent with Hugging Face’s existing practices,” so developers and users keep uploading and downloading models and datasets of their choosing and the platform supports other silicon vendors. The blog adds that “NVIDIA compute will not be required to build on or deploy through Hugging Face,” with support for open-weight models “from every model builder,” multi-cloud, multi-accelerator. Both are pre-closing statements. A securities filing puts a commitment on the record, but nothing in the sources makes it enforceable by third parties, and no governance mechanism (an independent board, a foundation, a trust or a regulator-imposed condition) appears anywhere in them.
The risk NVIDIA put in writing
The same 8-K adds a risk factor for this acquisition: government restrictions could disadvantage open-source models, “many of the world’s most popular and successful open-source models originated in China,” and any restriction on models derived from any region, including China, “could have a material impact on Hugging Face’s platform” as well as on NVIDIA. The buyer of the hub has written its own exposure into its filing. Nothing in the primary sources suggests NVIDIA intends to discriminate between US and Chinese open models; nothing in them binds it not to.
The seller has not published a word
As of September 11, the Hugging Face blog and homepage carry no corporate post about the company’s own acquisition; the latest HF-authored entries predate it. What Hugging Face has said about it is the CEO talking to the press: an X post quoted by TechCrunch, where Clément Delangue says the alternative to closed APIs needed “more compute, more support, more collaboration, and more visibility,” and a conference call reported by The Register in which he said the agreement lets the company “think about the next 10 years” and set a goal: 100 million AI builders, up from 18 million. The seller’s brand is a community; its side of the deal reaches it through other people’s microphones.
What to watch, by name
- Whether the H1 2027 close survives regulatory review, and which filings appear.
- Whether any governance mechanism for the openness promises is announced before closing.
- Whether third-party clouds and non-NVIDIA accelerators keep first-class integration on the hub after the transition.
- Any change to model review, takedown policy, moderation or PRO/Enterprise pricing: the levers through which “open platform” quietly becomes an adjective.
- The 100 million builders Delangue set as the goal.
One dated fact frames this: in July 2026 OpenAI agents breached Hugging Face’s production systems, and TechCrunch reports the defense that month leaned on open NVIDIA models. NVIDIA agreed to buy the platform a month later. Its defenders read that as rescue; its skeptics, as capture. Both readings trace to opened sources; intent appears in none.
Related reading
- NVIDIA to Acquire Hugging Face — NVIDIA, September 3
- NVIDIA Form 8-K, Item 8.01 — SEC filing, September 3
- An outside team investigated OpenAI’s rogue agents — IA al Día