top of page

What Happens to Jensen Huang’s $13 Billion Bet on Open AI?

29 minutes ago
9 min read
A man in a black leather jacket stands beside a glowing amber circuit board holding a large smiling emoji face chip, with a gold OpenAI logo chip and a blank green processor chip linked by luminous traces, set against rows of green-lit data center racks

The Hugging Face deal gives Nvidia’s new leadership an early test. Can the world’s most powerful AI chipmaker own a platform that helps developers run models on somebody else’s chips?


Here's the part of the Jensen Huang story I keep coming back to.

On September 3, Nvidia agreed to buy Hugging Face for $12,930,300,000. And one of Huang's selling points was that you'd still be able to use it without buying a damn thing from Nvidia. You could pick someone else's chips, someone else's cloud, someone else's model.

The man selling the shovels wanted to buy the place that helps you figure out which shovel to use. He said you could pick one off somebody else's shelf.


Huang resigned on November 12. Colette Kress, his chief financial officer, is interim CEO. The deal is still pending, and the paperwork leaves Nvidia a lot of room to be flexible later about what "open" means in practice.

So the interesting question when Nvidia talks to investors on November 17 isn't whether the deal closes. It's whether anyone still running the company believes the reason he gave for doing it.


The Price Was Never $13 Billion

Start with what Nvidia actually filed, because it's thinner than the coverage implied.

The agreement was signed September 2 and disclosed the next day in an 8-K filed under Item 8.01, "Other Events". Not Item 1.01, "Entry into a Material Definitive Agreement." No merger agreement is attached to it. Which means the public record contains no outside date, no break fee, and no termination provision. There's no filed consequence if the thing drags into 2028.


The number is $12,930,300,000, and that's Huang's number, from the blog post he published to announce the deal. The 8-K never adds the pieces together. What it discloses is approximately $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 joining Nvidia.


That retention pool is contingent on the employees actually showing up and staying. So $12.93 billion is a ceiling, not a price. It's the maximum Nvidia pays if everything works out.

Also worth noticing: the filing never uses the word "cash." A lot of the reporting described the $11.9 billion as cash. The document says purchase price payable, subject to adjustments it doesn't define. Where the filing and the reporting disagree on terms, the filing governs, and the filing is silent about form of consideration.


For scale, consider a company that reported $96.221 billion in revenue for the quarter it announced on August 26. This deal is about 13.4% of one quarter's revenue. It's the second-largest acquisition in Nvidia's history, behind $20 billion for Groq's chip assets in December 2025, and ahead of Mellanox at roughly $7 billion in 2019. In that same quarter Nvidia returned $26.0 billion to shareholders, which is double what it's paying for the most consequential distribution asset in open AI.


What Nvidia Actually Bought

Hugging Face operates at a scale that's easy to underestimate if you don't live in it. More than 18 million developers, researchers and creators use the platform. It hosts more than 3 million models, half a million datasets and a million applications. More than 200,000 companies use it to find, evaluate, customize and deploy models.


Nvidia is the largest single contributor to the place: more than 500 models and more than 250 open datasets.


Now think about a small company that wants to build an AI tool for its own documents. It can send requests to a hosted model and pay for usage. Or it can download a model whose license permits what it wants to do, test it on its own material, and run it on infrastructure it controls. Hugging Face is the second path. It's where you find weights, check licenses, browse datasets, try demos, and see which hardware people are actually running things on. If you're the one who has to make the thing work, those details are the project.


Nvidia already sells the accelerators that train nearly every significant model. What it didn't own was the place where those models get found, evaluated and pulled down. That's the asset. Distribution is where defaults get set, and defaults decide which silicon a workload eventually runs on.


The Business Logic Held Up Fine

The strategic case was always decent. If a handful of labs run most of the world's AI, Nvidia has a handful of enormous customers, and every one of them can justify building its own silicon. If thousands of companies run their own models, Nvidia gets thousands of chances to sell hardware or win the cloud workload behind it.


Some of those workloads land on competing chips. That's fine if the overall market grows faster than Nvidia's share of any single layer shrinks.


Nvidia has been funding that thesis for a while, in public. On July 24, Huang published his first-ever post on X: an open-weights letter signed by 25 companies, including Microsoft, Meta, Hugging Face, IBM and Palantir, arguing that models anyone "can download, inspect, modify, and run on their own infrastructure" are what make AI reach factories, hospitals and small businesses. In August it agreed to put $6 billion into Poolside to build an open AI ecosystem in the United States. In October it added a 120-billion-parameter Nemotron model for enterprise agents, plus two lightweight Bolt models that run on PCs.


And the "paid either way" part is real, with one correction worth making. Nvidia isn't simply writing a check to Anthropic. It's anchoring Anthropic's IPO. The S-1 amendment filed November 10 lists seven companies buying $26.0 billion of Class N stock at the IPO price, and Nvidia and Alphabet are each in for $10 billion.


So Nvidia holds the front door for open models and an anchor position in the biggest closed lab. Whichever direction customers run, it owns a piece of the outcome.


The Person Now Holding the Pen

This is where it gets uncomfortable for the thesis.

Kress didn't just inherit a deal. She inherited the argument for it. And she came into the job having been publicly dressed down by the man who made it. The Wall Street Journal reported that in August, during a strategy meeting, Huang interrupted her presentation and called her "a deer in the headlights" and a "reporting CFO." The same reporting describes her as having questioned some of his dealmaking.


Three months later, she's the one who has to defend a $12.9 billion acquisition to a board that just retained Wachtell, Lipton to conduct an independent investigation of the circumstances around his exit.

The transition announcement is careful, and almost entirely about continuity. Harvey Jones, the first-ever chair of the board, says he expects "the company's operations, customer commitments and product roadmap to continue without interruption." Kress says she's "committed to building on the extraordinary foundation Jensen created over the last 33 years." Neither of them mentions Hugging Face. TechCrunch asked whether Hugging Face had spoken with Kress, and whether the deal's status had changed. No answer came back.


Then there's Clément Delangue, who sold the company and is now watching all of this from inside it. On September 18 he posted that "concentration of power in a few labs is the biggest risk in AI in my opinion." On November 12, the day Huang resigned, he wrote: "This whole thing sucks, and I wish none of it had happened. It's a really sad day! But I want to make one thing clear: we'll keep fighting for open-source AI, harder than ever."


That's the founder of a company being acquired, publicly promising to keep fighting for the exact thing his acquirer promised not to restrict.


The Promise That's Actually in Writing

Here's where the deal gets more interesting than the commentary around it.

Huang's blog post made a commitment that's unusually specific for an acquisition announcement. "Hugging Face will remain an open platform for the entire AI ecosystem," he wrote. "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. NVIDIA compute will not be required to build on or deploy through Hugging Face."

Blog posts aren't filed, though. The 8-K repeats the pledge in language a regulator can hold onto. Nvidia "has committed to, among other things, keep Hugging Face's platform open, consistent with Hugging Face's existing practices." Under that commitment, 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."


Read those verbs again. Committed to, among other things, keep open. Would continue. Consistent with existing practices. That's not a covenant to do anything new. It's a promise to keep doing what the platform already does, surrounded by qualifiers that a good lawyer can work with.


It's still a behavioral remedy written before a regulator asked for one, and companies don't usually volunteer that in an SEC filing unless they expect the question.


They were right to expect it. On October 28, Reuters reported that the Justice Department had begun asking how the acquisition would affect competition in AI.


The same filing also added a brand-new risk factor, which is the part almost nobody read. Nvidia warned that governments may restrict open-source models, that "other parties are actively lobbying the U.S. Government and other stakeholders worldwide to adopt legislative or regulatory measures that would restrict or disadvantage open-source models," and that such rules "could restrict the models or datasets available through Hugging Face."

So the company told the SEC, in writing, two things at once. That open models drive demand for its products. And that the political risk to open models is now material enough to disclose.


The Reading From the Right

David Sacks picked this up on All-In and got to the political version of the argument fast. His framing was who benefits, and his answer was China. Pull the most visible champion of open weights out of the most important company in AI, and the people who want tighter control over model distribution get what they wanted without having to win the argument.


That's a theory. The podcast also spent time speculating about who wanted Huang gone, which is a question the public record doesn't answer. The board retained outside counsel and hasn't published findings. What's actually documented is the resignation, the investigation, and the reporting about two extramarital affairs, one of them with a woman suspected of Chinese intelligence work.


I don't have anything to add to the speculation, and neither does the evidence. What I have is better than a theory: a specific public commitment, in a filing, that can be compared against how the platform actually works. That's rare. Most of the time when a company buys a neutral platform, the neutrality promise is a press release and a shrug.


How the Promise Dies Without Anyone Breaking It

Suppose Nvidia keeps every model downloadable, keeps every license visible, and never blocks a competitor's chip. It can still tilt the field.

Which hardware gets tested on release day? Which setup gets the best documentation? Which provider is pre-selected when you click Deploy? Whose tutorial sits at the top of the model card?

Each choice is defensible on its own. Taken together over a couple of years, they decide what a developer does. Nobody on a deadline reads the license comparison chart. They take the path that worked on the first try.

That's how much influence a platform can have without touching a single model license. And it's exactly the kind of influence that never shows up in a press release about openness.


What to Listen For on November 17

Nvidia reports fiscal third-quarter results on November 17, with numbers around 4:20 p.m. Eastern and the call at 5. Guidance was $108.0 billion, plus or minus 2%.


Three things matter more than the revenue number.

First, does management still expect the deal to close in the first half of 2027. That's the date in the filing, and there's no filed consequence if it slips, because there's no outside date to breach.

Second, does anyone on that call repeat the commitment to support other silicon vendors. The commitment is sitting in a document. Saying it out loud under questioning from analysts is a different act of will than leaving it in a filing.


Third, and this one matters for the whole story: who's actually answering for this acquisition now. An interim CFO turned interim CEO. A board chair in his first quarter. A permanent CEO search running in the background. Deals survive leadership changes all the time. But the rationale for this one was personal. Huang wanted it, Delangue went to him first, and when asked who else had bid, Huang said: "It doesn't matter who the other bidders were. It only matters who wins."

That's not a strategic framework. That's a founder.


The Simple Test

If the acquisition closes, the test isn't a policy paper. It's an afternoon.

Pick a model on Hugging Face. Pick a chip that isn't Nvidia. Try to get the thing running the way the platform suggests you should. Count the steps. Count the places where the documentation assumed something you don't have.


Then do it again in a year.


That's the difference between a platform that stays open and a platform that technically stays open while quietly becoming the front door to one vendor's hardware. Both are defensible. Only one of them is what Huang told the world he was buying.


The money was never really about owning Hugging Face. It was about whether the most powerful company in AI infrastructure can hold a neutral platform and resist tilting it, especially after the person who promised not to tilt it is gone.


November 17 is the first honest read on that. But the answer won't come from anything said on the call. It'll come from what the platform looks like in eighteen months.


Rich Washburn is a technologist, strategist, and Founder & Chief AI Architect of ARIA AI Labs, working at the intersection of AI, infrastructure, communications, and capital. He also serves as Managing Partner and Chief AI Officer at Eliakim Capital.

Comments


Animated coffee.gif
cup2 trans.fw.png

© 2018 Rich Washburn

bottom of page