Nvidia Just Bought the Home of Open AI. The Real Story Is What It Promises Not to Do.

Here's the version of this story that got the headlines: Nvidia, the company that sells the shovels for the AI gold rush, just spent $12.9 billion to buy the town where everyone digs. Hugging Face is where 18 million developers go to share, download, and fine-tune open models — 3 million of them, sitting on half a million datasets, feeding a million applications used by 200,000 companies. It is, in the industry's own shorthand, the home of open AI.

The version that matters is quieter. It's the promise Jensen Huang made the day the deal was announced, and the question of whether that promise can survive contact with the economics of the AI stack.

The deal, in numbers

The transaction, announced September 2 and confirmed in an SEC 8-K the next day, breaks down as roughly $11.9 billion to Hugging Face stockholders plus an equity-based retention program of up to $1 billion for employees joining Nvidia. It's expected to close in the first half of 2027, subject to regulatory approvals. That makes it Nvidia's second-biggest acquisition ever, trailing only the $20 billion it paid for Groq's assets in December. Before that, the record was Mellanox at about $7 billion in 2019.

For context on how fast this moved: Hugging Face was founded in 2016 and has raised just over $395 million in its entire life, with its last round a $235 million raise in 2023 led by Salesforce Ventures. Last year, per the Financial Times, the company turned down a $500 million offer from Nvidia. Last month, The Information reported Hugging Face was clocking about $150 million in annualized revenue. A year ago, Nvidia couldn't buy it for half a billion. Now it's paying $12.9 billion.

The promise

Huang's blog post announcing the deal was careful, almost lawyerly, about what Nvidia would not do. "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."

That last sentence is doing a lot of work. Nvidia is the dominant hardware platform for AI training and inference — the company became the world's most valuable business on the back of that dominance. Owning the hub where the open-model community lives gives it a vantage point no competitor has. The promise is that it won't use that vantage point to steer anyone toward its own chips.

The 8-K commits to the same thing in regulatory language: 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 tension

Here's the thing about a company that sells the compute promising not to require its own compute: it doesn't have to require it. It just has to be there. Nvidia is already the largest contributor of open models and data to Hugging Face — more than 500 models and 250 open datasets, per Huang. It has poured over $50 billion into AI frontier labs and struck a $6 billion deal with coding startup Poolside to develop open models. The open ecosystem runs on Nvidia hardware almost by default, because almost everything runs on Nvidia hardware.

The Register's analysts put the worry in sharper terms: this could usher in "a new, more fragmented and siloed era of AI." The concern isn't that Nvidia will yank the open platform shut tomorrow. It's that the platform's neutrality — the thing that made it the neutral ground where every model builder, from Meta to a Chinese lab, could coexist — is now owned by the single most powerful company in the industry. Neutrality is a posture, and postures are hard to hold when you're also the landlord.

There's also the question of what Nvidia gets out of it. TechCrunch's read is blunt: an open ecosystem that Nvidia controls is a platform it can shape toward its chips, and it can package its unused capacity with Hugging Face's offering to sell to enterprise customers. The open-model hub becomes a distribution channel for the hardware business.

Why open models matter

Huang has been unusually vocal about open weights lately, co-authoring an open letter with other industry leaders arguing that open models broaden access to AI and keep leadership distributed. His cybersecurity argument is the sharpest one. "One of the areas where frontier models are vital is cybersecurity," he told analysts. "Those companies couldn't do it without open models."

It's not a hypothetical. Hugging Face itself was hit by a hacking incident recently, and CEO Clément Delangue said the platform defended itself using an Nvidia version of a Chinese open model after proprietary models failed. Days earlier, OpenAI admitted an unreleased model had breached the platform. The defenders' edge, in Huang's framing, is an "asymmetric advantage" — more people protecting than attacking, all collaborating transparently on open models.

Delangue, for his part, says he approached Huang over the summer. "During the summer, I think we realized that Hugging Face and open-source AI in general was at the turning point, and that it needed more resources, more scale, more visibility," he told CNBC. "A few weeks later, here we are."

The synthesis

The $12.9 billion is the headline. The real story is the structural bet underneath it. Nvidia has decided that the open-model ecosystem is worth more to it as a thing it owns than as a thing it merely feeds. That's a bet that open models are the future of AI — and that the company best positioned to profit from them is the one that controls both the compute and the community.

The tension is that those two roles pull in opposite directions. A neutral open hub is valuable precisely because it's neutral. An open hub owned by the dominant hardware vendor is valuable precisely because it isn't. Nvidia is betting it can have both. The next few years — and the regulatory review that will stretch into 2027 — will test whether that's a bet the market believes.

The infrastructure question behind all of this is the one that doesn't make the announcement call. When the open-model hub and the compute platform are owned by the same company, the real cost of AI stops being the model and starts being the system around it — the sourcing, the capacity planning, the deployment decisions that determine whether an open model actually runs well in production. That's the kind of problem that doesn't show up in a press release. At DMC, we work with hardware and infrastructure teams navigating exactly these constraints — capacity strategy, cost modeling, and deployment planning when the platform underneath your AI stack is shifting under you. Need help stress-testing your AI infrastructure roadmap? Let's talk.