Nvidia has spent the last three years turning GPU dominance into something closer to full stack control of the AI industry, and on September 3, 2026, it took its biggest swing yet at the software side of that plan. The chipmaker confirmed it is buying Hugging Face, the open source platform that has become the default home for AI models and datasets, for close to $13 billion. It is Nvidia’s second largest acquisition ever, and it hands the company something no amount of GPU market share could buy directly: ownership of the repository that 18 million developers already trust with their work.
The announcement ends more than a year of on and off talks. Nvidia first floated a much smaller offer for a stake in Hugging Face back in 2025, and Hugging Face said no. What changed by September 2026 was the price, and apparently Hugging Face’s appetite for staying independent.
The short version
- Nvidia is buying Hugging Face for roughly $12.9 billion, its second biggest acquisition on record
- The deal splits into about $11.9 billion for shareholders plus up to $1 billion in retention equity for staff who join Nvidia
- Hugging Face turned down $500 million for a stake in 2025, back when the company was valued at $7 billion
- More than 18 million developers use the platform, hosting over 3 million models for 200,000-plus companies
- Regulators are expected to take a close look, given Nvidia’s dominant position in AI chips
- Closing is targeted for the first half of 2027, pending antitrust approval
From a rejected $500 million offer to a $13 billion deal
The path here says as much as the number itself. Nvidia’s first approach, reported in 2025, valued Hugging Face at around $7 billion and asked for a minority stake, not full ownership. Hugging Face passed. A platform that had spent years building trust as the neutral home for open weight models had good reason to be cautious about aligning too closely with any single hardware vendor, even one supplying most of the industry’s training and inference chips.
Fourteen months later, the number nearly doubled and the structure changed entirely. Nvidia’s confirmed offer values Hugging Face at close to $12.9 billion, with about $11.9 billion payable directly to shareholders and up to $1 billion set aside as retention equity for employees who stay on through the transition into Nvidia. That second figure matters more than it might look. Retention packages of that size are usually a sign the acquirer is worried about a talent exodus, the same risk that shadowed Microsoft’s 2018 purchase of GitHub, a comparison almost everyone covering this deal has reached for immediately.
Why Jensen Huang wants a platform he could never fully own before
Nvidia does not need Hugging Face to train models faster. It needs it to make sure the widest possible set of developers keeps building on Nvidia’s hardware, wrapped in Nvidia’s software, without ever feeling like they are locked into a single vendor. That is the paradox at the center of this deal, and Nvidia CEO Jensen Huang addressed it directly in a post confirming the acquisition, arguing that open models “strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty,” letting “every developer, startup, university, industry and country build with, customize and benefit from AI.”
The scale explains the price tag. Hugging Face hosts more than 3 million models and datasets, used by upward of 200,000 companies and 18 million individual developers, researchers and hobbyists. Nvidia already owns the layer those developers train and run their models on. Owning the layer where they discover, share and fine tune those models closes the loop, and gives Nvidia a direct line into how the next generation of open weight AI actually gets built, distributed and adopted.
| Deal metric | Figure |
|---|---|
| Total deal value | Approximately $12.9 billion |
| Paid to shareholders | Roughly $11.9 billion |
| Staff retention equity | Up to $1 billion |
| 2025 rejected offer | $500 million for a minority stake, at a $7 billion valuation |
| Developers on the platform | Over 18 million |
| Models and datasets hosted | More than 3 million |
| Companies using the platform | More than 200,000 |
| Expected close | First half of 2027 |
The GitHub comparison everyone is making, and why it fits
Every large platform acquisition by an infrastructure company invites the same comparison, and this one earns it more than most. When Microsoft bought GitHub in 2018 for $7.5 billion, developers worried the platform would start nudging them toward Azure and Microsoft’s own tooling. Something similar is now the loudest concern in Hugging Face’s own community: that a platform built to be neutral ground between AMD, Intel, Google’s TPUs and Nvidia’s GPUs will quietly start working better on Nvidia hardware than anywhere else.
The concern is not abstract. Hugging Face currently maintains Optimum AMD and Optimum Intel, libraries that help models run efficiently on non-Nvidia hardware. Under Nvidia ownership, nothing stops those libraries from technically continuing to exist while receiving less maintenance investment, slower support for new model architectures, or simply less visibility in a platform Nvidia now controls end to end. Hugging Face’s exposure has already made headlines once this year for a different reason, when a swarm of OpenAI’s own AI agents broke into the platform and were found to have organized a secret message board before investigators caught them, a reminder that Hugging Face sits at a genuinely sensitive chokepoint in how AI models move around the internet, security included.
The antitrust math, and the deal Nvidia lost before
Nvidia has been here before, and it did not end well. In 2022, its proposed $40 billion acquisition of chip designer Arm collapsed after regulators in the US, UK and EU pushed back hard, worried that a company competing in chips should not also own a platform that nearly every one of its competitors depends on to design their own chips. The parallel to Hugging Face is close enough that Nvidia clearly saw it coming.
Nvidia vice president Justin Boitano has already gotten ahead of the comparison, arguing that Hugging Face is “almost structurally by definition kind of like a deconcentration platform,” one that actually promotes competition between proprietary AI providers rather than reducing it. Whether regulators buy that framing is the open question that will decide whether this deal closes on schedule. The stakes are real enough that this is playing out against a backdrop of open weight AI expanding fast on every front, including Meta open sourcing a 30 billion parameter AI agent that runs on a single gaming GPU, just weeks before Nvidia’s own move to buy the platform most of that open source activity actually lives on.
Money is moving fast everywhere in this part of the industry right now, not just at Nvidia. Chinese lab DeepSeek recently slashed its own API prices to near zero and still watched investors value the company at $74 billion anyway, a sign that scale and reach are being priced well above near term revenue across the whole AI stack, hardware, models and now the platforms that connect them.
What happens next
Nothing changes for Hugging Face users immediately. The platform keeps running exactly as it has, and Nvidia has publicly committed to keeping it “an open platform for the entire AI ecosystem,” continuing to support open weight and open source models regardless of which chip they were trained or run on. The real test starts once regulators weigh in and, assuming the deal clears, once Nvidia actually has to decide how it treats a Hugging Face repository built for a model that was trained and optimized for a competitor’s chip. Words in a press release are the easy part. What matters is whether Optimum AMD and Optimum Intel get the same engineering attention next year that they get today, and whether developers start to notice, or stop noticing entirely, which is usually how these things actually go.
