Nvidia Buys Hugging Face for $12.9 Billion — the Open-Weight Town Square Gets a Landlord

The price is $12,930,300,000 — and yes, the digits are a joke. Nvidia announced Thursday that it has agreed to buy Hugging Face, the open-source platform that hosts the world's largest collection of open-weight AI models and datasets, for $12.9 billion. Hugging Face cofounder Thomas Wolf confirmed the exact figure and dared readers to find "the Hugging Face and Nvidia references hidden" in it (Business Insider). It is Nvidia's second-biggest acquisition ever, behind only the $20 billion purchase of Groq's assets last December, and roughly double the ~$7 billion Mellanox deal that was its record for years (CNBC).

The more important number in this deal is zero — the number of paywalls Nvidia is promising to add. In his announcement blog post, CEO Jensen Huang wrote that Hugging Face will "remain an open platform for the entire AI ecosystem" (CNBC). When the world's most valuable company buys the town square where open AI happens, that promise deserves to be read closely — and priced into every product that depends on the Hub.

What Actually Happened

• Nvidia agreed to acquire Hugging Face for $12.9 billion, announced Thursday morning — its second-largest deal on record after the $20 billion Groq asset purchase in December 2025 (CNBC).
• The platform stays open, per Huang: the blog post commits to keeping Hugging Face "an open platform for the entire AI ecosystem," writing that "together, we will scale Hugging Face's platform, strengthen its infrastructure and expand access to AI for developers and institutions worldwide" (CNBC). Nvidia has also said it will keep the platform open and consistent with the startup's existing practices (Mercury News).
• Hugging Face started it. CEO Clément Delangue told CNBC's Squawk Box that he approached Huang over the summer — "and a few weeks later, here we are" — because open-source AI was "at the turning point" and needed "more resources, more scale, more visibility." He called Nvidia "a perfect home."
• The deal was expected: The Information reported it last week, and the announcement landed Thursday morning (CNBC).
• The strategic read from the Wall Street Journal: the acquisition ramps up Nvidia's efforts to promote open-weight AI models — the same models the paper notes "pose a competitive threat to OpenAI and Anthropic."
• The deal lands three-plus weeks after a hacking incident at Hugging Face that raised concerns about the rapid evolution of AI-powered cybersecurity tools (CNBC). Delangue blamed engineering mistakes, said the company used an Nvidia version of a Chinese open model to resolve the attack, and framed the breach as proof that open source needed to double down.

Why Nvidia Wants the Town Square

Nvidia sells the silicon that every meaningful AI model — open or closed — trains and serves on. Hugging Face is where the open ones live. The Hub is the default place developers search, compare, and download models; MarketWatch described it as "one of the most important parcels of real estate in the AI market." Owning the real estate means owning the shelf placement.

The New York Times frames the deal as evidence of Nvidia's growing role as "Silicon Valley's central banker" — the company whose capital and now ownership stakes are woven through nearly every layer of the AI economy. That's the correct lens. Nvidia has already put billions into open-weight model companies (its $6 billion bet on Poolside being one recent example). The Groq purchase bought hardware; this one buys distribution for the software side of the same strategy.

The economics are lopsided in Nvidia's favor, and that's the point. Nvidia doesn't need Hugging Face to turn a profit. It needs open models to proliferate, because every open-weight deployment — whether it runs on a hyperscaler cluster or a laptop — is demand for GPU-hours. Every popular open model on the Hub is, functionally, marketing for the next data center order. Promoting open-weight alternatives to OpenAI and Anthropic (the WSJ's framing) isn't charity; it's a wedge that keeps inference workloads flowing through hardware Nvidia controls.

For Hugging Face, the sale is the honest endgame of a platform that couldn't monetize its position fast enough. Hosting millions of models and datasets is an infrastructure-arms-race business, and Delangue's own words — open-source AI needed "more resources, more scale, more visibility" — read as an admission that the platform alone couldn't fund that race. Selling to the company that benefits most from open AI's growth is a rational exit. It just concentrates the ecosystem's center of gravity in one place.

"Open" Under New Ownership — and the Security Backdrop

The promise to keep the platform open is the whole ballgame, so it's worth being precise about what's pledged and what isn't. Huang's blog post says Hugging Face remains "an open platform for the entire AI ecosystem." Nvidia told reporters it will keep the platform open and consistent with existing practices (Mercury News). Neither quote is a contract, and neither covers the questions that will actually matter: whether Nvidia-built models get preferential placement or bundled inference on the Hub, how the paid Pro and Enterprise tiers evolve, how datasets from labs that compete with Nvidia are treated, and who arbitrates takedown requests from governments.

There's precedent for optimism and for skepticism. Platform-acquisition openness pledges have held before — AWS committed to keeping DuckDB open source when it acquired Duck Labs, and so far has — but the AI version of this question carries more strategic weight. The Hub isn't just a code repository; it's the distribution chokepoint for a modeling ecosystem that Nvidia has every commercial reason to steer. Open platform and neutral platform are different promises. Only the first was made on Thursday.

The security backdrop sharpens the stakes. The August breach at Hugging Face — which Delangue attributed to engineering mistakes, and which the company resolved using an Nvidia version of a Chinese open model — is a reminder that the industry's shared commons is attack surface (CNBC). Huang's response is a real argument: there are "way more people who are protecting than there are people who are attacking," so transparent collaboration on open models gives defenders an "asymmetric advantage" (CNBC). That claim is testable, and now the company making it owns the lab where it gets tested. Regulators, eventually, will have views on the world's most valuable chipmaker owning the open AI commons. For now, both CEOs are pre-committing to openness — which is exactly what makes the next twelve months of Hub decisions worth watching.

What Builders Should Take From It

• Mirror your critical weights. "Remain open" is a commitment from executives, not a clause in your terms of service. If a model or dataset is load-bearing for your product, keep a pinned, verified copy somewhere you control. Downloads of public weights are cheap insurance.
• Watch the paid tiers. Hugging Face Pro and Enterprise are the canary for what "open" means in practice. Price changes, inference bundling, or Nvidia-only optimizations will tell you more about the deal's direction than any announcement post.
• Expect the free tier to get better. Delangue asked for "more resources, more scale, more visibility," and Nvidia has all three. Faster model hosting, better tooling, and deeper integration with the CUDA stack are the plausible upsides — at least while the openness pledge is fresh.
• Treat model supply chains as attack surface. The August breach was, per Delangue, an engineering-mistake story — but it happened to the industry's shared infrastructure. Pin versions, verify checksums, and audit what you pull from the Hub like you would any third-party dependency.
• Retire the assumption that the Hub is neutral ground. A platform can stay operationally open while structurally favoring its owner's stack. If your product strategy assumes the Hub will always be Switzerland, build a fallback now — self-hosting key artifacts or adding a second registry to your pipeline.

The open-weight ecosystem just got its biggest subsidy and its biggest governance question in the same announcement. Both come with the same price tag: $12,930,300,000, and one emoji's worth of fine print.

Developer312 covers the AI business signals builders actually need to act on. Get the weekday briefing at developer312.com.