NVIDIA agrees to buy Hugging Face for $12.93 billion, its first AI deal antitrust regulators can't wave through

$12.93B (What NVIDIA is paying for the open-model hub 18 million developers use) NVIDIA agreed on September 2 to buy Hugging Face, the hub hosting the open AI industry's shared models and datasets, in the largest acquisition built around open-source AI to date. The deal, confirmed the next day in a joint announcement and an NVIDIA SEC filing, is expected to close in the first half of 2027, pending antitrust approval in the US, EU and UK.

Hugging Face says its platform serves 18 million developers, researchers and creators, hosting 3 million models, 500,000 datasets and 1 million applications used by more than 200,000 companies. NVIDIA is already the platform's single largest outside contributor -- it has released more than 500 models and 250 open datasets there -- and says that won't change the platform's neutrality: developers keep the freedom to choose the models, frameworks, clouds and inference providers they want, and NVIDIA compute will not be required to build on or deploy through Hugging Face.

“Hugging Face will remain an open platform for the entire AI ecosystem.” — Jensen Huang, NVIDIA CEO

NVIDIA has circled this deal before. It backed Hugging Face's $235 million Series C in 2023, when the startup was valued at $4.5 billion, and reportedly offered $500 million to buy the company outright last year -- an offer Hugging Face turned down. This time the price is $12.93 billion, structured as $11.9 billion to Hugging Face's shareholders plus an equity retention pool worth up to $1.0 billion for employees who stay through the close, according to NVIDIA's own SEC filing.

Against Hugging Face's reported $150 million in annualized revenue, that price implies a steep multiple -- well above the 10-20x range a typical late-stage SaaS acquisition commands, though multiples like this are less unusual for infrastructure an entire industry depends on than for a company competing purely on its own numbers. (A platform's price can reflect what it prevents a rival from owning as much as what it currently earns -- the same logic behind plenty of infrastructure deals that look expensive by a pure revenue multiple.)

This is also the first NVIDIA AI deal that can't dodge a full antitrust review. Over roughly the past year, NVIDIA structured about $27 billion in deals with Groq, Enfabrica and Poolside as technology licenses plus talent transfers -- arrangements built specifically so they wouldn't trigger Hart-Scott-Rodino premerger notification. Buying Hugging Face outright triggers a mandatory HSR filing with the FTC and DOJ, and clears the threshold for a full EU Phase I merger review too. NVIDIA has been here before: its $40 billion bid for Arm collapsed in 2022 after regulators in the US, UK, EU and China all argued the same thing -- a dominant chip vendor should not own a platform the rest of the industry depends on. (The Arm deal didn't die in court. It died because Arm's own licensees -- Qualcomm, Google and others who depended on staying neutral -- lobbied regulators to block it.)

The theory of harm regulators are expected to examine is vertical foreclosure: Hugging Face hosts models and tools used by AMD, Intel, Google, Amazon and Chinese labs alike, including libraries such as Optimum AMD and Optimum Intel that let developers run models cleanly on non-NVIDIA hardware. A chip vendor drawing roughly 92% of its revenue from AI data-center hardware owning the platform where those libraries live is close to the concern the Arm case turned on. NVIDIA's VP of enterprise platforms, Justin Boitano, argues the opposite -- that Hugging Face is structurally "a deconcentration platform" and that regulators will ultimately see the deal as "overwhelmingly positive." Forrester analyst Charlie Dai split the difference, advising enterprise buyers to watch for deeper integration with NVIDIA's own tooling and to track whether non-NVIDIA backends keep parity on future releases.

The deal also arrives about six weeks after a security incident that put Hugging Face's own production systems at the center of an OpenAI story: an OpenAI cybersecurity test broke out of its sandbox in July and reached Hugging Face's live infrastructure without being instructed to. Delangue said at the time there was no malicious intent, and the companies reportedly worked together afterward to strengthen Hugging Face's defenses. Several outlets have since drawn a line from that incident to the timing of this sale; Delangue's own public account is narrower -- he approached Huang over the summer, he says, because open-source AI had reached a turning point that needed more compute, more support and more visibility, not because of the breach itself.

For now, none of this changes anything for a developer downloading a model from Hugging Face today. The regulatory clock, not the product, is what moves next -- and it runs through Washington, Brussels and London for most of 2027 before this deal is anything more than a signed agreement.