Executive Summary
Nvidia's agreement to buy Hugging Face for $12.9 billion survives the departure of Jensen Huang. Colette Kress, the chief financial officer who has taken over as interim CEO, said so on the November 17 earnings call. Huang had resigned with immediate effect five days earlier.
Hugging Face is the shared warehouse where 18 million developers and 200,000 companies upload and download models and datasets. Nvidia has promised to keep that platform open, and the promise sits in more than a press release: it is written into the Form 8-K the company filed with the Securities and Exchange Commission on September 2, down to a line about supporting other silicon vendors. What the filing never states is how much support that means, and the yardstick it sets, "Hugging Face's existing practices," has no public definition anywhere. The US Department of Justice is already examining the deal.
Sections 1 through 4 are what the announcements, the filings and the reporting say. Section 5 is this article's reading.
Key Figures
Four numbers give the shape of the deal: the price, the size of what is being bought, how far that price has moved in three years, and when the transaction actually closes.
Sources: Nvidia's Form 8-K dated September 2, 2026 and the September 3 acquisition coverage.
$12.9B
Total deal value
About $11.9 billion in cash to shareholders, plus up to $1 billion in equity to retain employees moving to Nvidia
18 million
Developers on Hugging Face
200,000 companies use it alongside them, and more than 3 million models are hosted there
2.9x
Price moved in three years
The 2023 round valued Hugging Face at $4.5 billion; this acquisition prices it at $12.9 billion
First half of 2027
Targeted close
Regulatory clearance is a condition. Until then the two companies operate separately
What the Interim CEO Confirmed, and What Went Unasked
Nvidia reported third-quarter results for fiscal 2027 on November 17. It was the first call held after Huang's departure, and the person leading it was Colette Kress, the chief financial officer who now doubles as interim chief executive.
Kress addressed the Hugging Face acquisition in prepared remarks. "At Nvidia, we are committed to fostering a vibrant open model ecosystem, so every country, company, and industry can build its own intelligence," Kress said, adding that "together, we will scale its platform and make intelligence more open, more capable, and more accessible." Harvey Jones, who has taken the newly created role of board chair, said the same day that there is no plan to bring Huang back as chief executive, and that the board has full confidence in Kress's leadership and capabilities. The search for a permanent CEO is being run by the executive search firm Heidrick & Struggles.
Then there is the part that was not confirmed. The acquisition never came up in the analyst question-and-answer session that day. Kress's sentence was a line written in advance, not an answer given under questioning. That is no reason to think Nvidia's stance has weakened, but it does mean the commitment has yet to face public scrutiny with Huang gone.
The terms themselves were documented back in September. Nvidia's board approved the definitive agreement on September 2 and announced it the following day. Of the $12.9 billion total, roughly $11.9 billion goes to Hugging Face shareholders in cash, and up to $1 billion is equity compensation meant to retain employees who move over to Nvidia. This is the second-largest acquisition in Nvidia's history, targeted to close in the first half of 2027, subject to customary closing conditions and regulatory clearances in multiple jurisdictions.
How quickly the two companies closed the distance stands out as well. Nvidia joined a $235 million funding round in 2023 alongside Google and Salesforce, when Hugging Face carried a $4.5 billion valuation. In 2025, according to the Financial Times, Hugging Face turned down a $500 million investment at a $7 billion valuation. The following September it signed an agreement to sell the whole company for $12.9 billion.
What Nvidia's Commitment Leaves Out
Announcing the acquisition on September 3, Huang, then still chief executive, made three statements about openness. Hugging Face "will remain an open platform for the entire AI ecosystem." Developers "will choose the models they want, the frameworks they want, the clouds and inference service providers they want and the computing platforms they want." And "Nvidia compute will not be required to build on or deploy through Hugging Face."
A document that goes further than the press release exists. In the Form 8-K dated September 2, Nvidia told the Securities and Exchange Commission that it "has committed to, among other things, keep Hugging Face's platform open, consistent with Hugging Face's existing practices." Under that commitment, the filing continues, Hugging Face "would continue to permit model makers, developers, and users to upload and download models and datasets of their choosing and to support other silicon vendors." The press release names something that will not be required. This sentence says competing chips will be supported, which is an affirmative undertaking rather than a negative one. The filing carries one signature: Colette M. Kress, then executive vice president and chief financial officer. That is the same person who reaffirmed the acquisition on November 17.
So the question is not whether a promise was written down. The question is what the writing leaves out. The sentence in the 8-K sets no level of support. A line about supporting other silicon vendors says nothing about supporting them as well as Nvidia's own chips, and no term, no deadline and no remedy for a broken commitment appears anywhere near it. The yardstick is the largest gap of all: "Hugging Face's existing practices" is defined in no public document. Practices shift a little every year, and anything measured against them moves along with them. A promise not to lock the door is narrower than one to share the shelves inside.
The difference is not abstract, and it surfaces in very specific places: which chip the installation instructions on a model card assume, which hardware the first inference numbers are measured on, which runtime a new library supports on day one. A platform can be tilted through handles like these without ever being closed. That is why critics have asked for commitments in enforceable form, such as equal integration support for competing accelerators, rather than declarations.
And the person who wrote those sentences is no longer at the company. Nvidia's board accepted Huang's resignation as chief executive and director on November 11 and announced it publicly the next day. The company had earlier disclosed that it opened an inquiry related to reports that Huang had been targeted by a foreign intelligence service, and Huang apologized and stepped down after a run of reports about a personal relationship. Market capitalization fell sharply after the announcement, and shareholders filed a class action.
An individual's promise and a company's obligation do not carry the same weight. A Form 8-K is a disclosure to investors, not a contract a developer can take to court, and a commitment described in that filing is a different thing from a clause inside the merger agreement. The agreement itself was not attached to the 8-K. The only exhibit filed with it was a cover page, which leaves no way for anyone outside the company to confirm whether such a clause exists. A promise can outlive the person who made it, but what is left of it depends on who carries the duty to honor it, and in what form.
The Chip Seller Is Buying the Model Warehouse
Regulators are focused on the same spot. Reuters reported in October that the US Department of Justice had opened an antitrust inquiry into the Hugging Face acquisition. The issues are whether a chip company ends up holding the corridor through which AI gets distributed, and whether competing hardware from AMD and others keeps the treatment it has now.
That inquiry is part of a broader examination of Nvidia's other deals and sales practices, begun under the previous administration and still running. The broader file includes the company's arrangement with the inference chip maker Groq, its December 2025 acquisition of SchedMD, and allegations that networking equipment is bundled with chips. French authorities have a separate inquiry under way, and analysts note that the European Commission could examine the deal under competition law, given that Hugging Face began as a French company.
The precedent invoked most often for a platform bought by a larger company is Microsoft's 2018 acquisition of GitHub. Similar worries came from the open-source world at the time, and GitHub is generally credited with having stayed neutral. Put the two deals side by side, though, and one column reads differently.
| Axis | 2018: Microsoft → GitHub | 2026: Nvidia → Hugging Face |
|---|---|---|
| Core business of the buyer | Software and cloud | Chips for AI compute |
| What the platform hosts | Source code repositories | Models and datasets |
| Stake the core business has in the contents | Code hosted on GitHub did not compete directly with the core revenue, which is cited as the basis for its neutrality | Which models run well on which chips feeds straight into the core revenue |
▲ Only the contested axes from industry comparisons of the two deals
How closely the two layers already sit is something Nvidia's own announcement makes plain. Huang wrote in September that Nvidia has published more than 500 models and 250 open datasets on Hugging Face, and that nearly every open model runs on Nvidia hardware. The company that occupied the largest share of the shelves before the acquisition is now buying the warehouse.
The summary is short: Nvidia already holds the layer that runs the models, and this deal hands it the layer where those models are found and chosen. One company holding both layers is not illegal in itself. What draws an inquiry is that the seller of the lower layer decides how the upper one is arranged.
That same 8-K carries a warning Nvidia wrote for its investors in advance. Governments may impose new requirements on the development, training, deployment and use of open models, and such requirements "could restrict the models or datasets available through Hugging Face, require changes to Hugging Face's platform or practices, delay or restrict offerings, increase compliance costs or result in investigations or enforcement actions." A separate passage states that any restriction limiting the company's ability to support models derived from any region, including China, "could have a material impact on Hugging Face's platform." The company itself has put in writing that what stays in the warehouse is not decided by the owner alone.
Even if the Founder Stays, the Buyer's Board Decides
Clément Delangue, co-founder and chief executive of Hugging Face, posted on X within hours of the news about Huang. The post opened with "this whole thing sucks, and I wish none of it had happened. It's a really sad day!" and went on to say that "this is too important to slow down now," closing with a line about open source: "we'll keep fighting for open-source AI, harder than ever."
That pledge has a backstory. Delangue has said that Hugging Face approached Nvidia first. Open source needed more compute, more support and more reach, by that account, and Nvidia was the company that had them. This acquisition reads less like a forced sale than like a road the founder chose.
A pledge and the authority to keep it are separate things. Once the deal closes, final say over how the platform is run passes to the board of the acquiring company. A founder who stays and keeps speaking up may still find that whether that voice can overturn a decision is settled by equity and contract. Nor has either side confirmed whether Delangue and Kress have discussed the direction of open source from here. Both companies declined to answer press inquiries.
Three sets of words are on the public record: Huang's September announcement and the commitment in that month's 8-K, Kress's sentence in November, and Delangue's pledge. All three may well be honored. Even the sturdiest of them, the 8-K, sets no level and no term, and not one of the three says what happens if it goes unhonored.
Why Pebblous Is Watching This Deal
From a distance this is news about one large company buying another. Up close it shakes an assumption sitting under our daily work. For several years now, using a model has meant, in practice, downloading it from Hugging Face. Pretrained models came from there, so did evaluation datasets, so did tokenizers, and so did the license text. That single place is changing hands.
Nothing is about to be cut off. The deal closes in the first half of 2027 at the earliest, and the platform will run as it does now for some time yet. What changes is the kind of question an organization that leans on this distribution channel has to ask. Until now the question was which model is best. Alongside it now belongs another: in what form does this model remain on our side of the line?
Four places are worth checking. Each of them asks the same thing: if the platform changes tomorrow, what stays with us?
- Do the models and datasets we use exist as copies in our own storage, or only as a line of address?
- Can the license and provenance record for a model be reconstructed outside the platform? Are the retrieval date and version still recorded?
- How tightly are the training and evaluation pipelines bound to one platform's API? How many days would it take to strip those calls out?
- Can a past experiment be rerun today? Are the weights and the data used back then still in hand, exactly as they were?
When we call data ready for AI, we mean more than tidy formatting. We mean that the provenance, the rights and the reproducibility stay on our side. Anything parked on an outside platform is convenient, and the terms of that convenience are set by someone else. The moment the party setting those terms changes, what an organization actually holds becomes visible.
There is no reason to doubt what Kress said on November 17. Still, how long such words last is determined by the ownership structure of a company rather than by the person who said them. These two weeks have shown that the person making the case most forcefully can be gone within five days. Whoever owns the platform, the one thing that does not move is what has been copied into our own hands.
References
Primary Sources
- 1.NVIDIA Corporation. (2026-09-02). Form 8-K — Current Report (Hugging Face acquisition agreement). U.S. Securities and Exchange Commission (SEC EDGAR).
- 2.NVIDIA Corporation. (2026-11-12). Form 8-K — Current Report (CEO transition). U.S. Securities and Exchange Commission (SEC EDGAR).
Industry & Press Sources
- 3.TechCrunch. (2026-09-03). Nvidia confirms it will buy Hugging Face for $12.9 billion.
- 4.TechCrunch. (2026-11-17). Nvidia says its open-source AI strategy remains intact after Jensen Huang's exit.
- 5.Technology.org. (2026-11-18). Nvidia's open-source AI strategy, Hugging Face, Kress, and the Huang exit.
- 6.PYMNTS. (2026-10). DOJ Probes Nvidia's Hugging Face Deal Over AI Competition Concerns.
- 7.Fortune. (2026-11-18). Nvidia's Q3 earnings: interim CEO Colette Kress on leadership and the Hugging Face deal.
- 8.CryptoBriefing. (2026-11). Hugging Face CEO vows to uphold open-source AI commitment amid Nvidia CEO resignation.