Nvidia Agrees to Buy AI Model Hub Hugging Face for $12.9 Billion, The Information Reports

A $12.9 Billion Deal Nobody Has Confirmed

On Aug. 26, 2026, TechCrunch reported that Nvidia was closing in on a deal to buy Hugging Face, the site where developers share AI models, but that nothing had been signed yet^[3]. A day later, The Information said the deal was done: Nvidia had agreed to pay $12.9 billion, according to one person familiar with the matter^[2]. CNBC, Fortune, Forbes and Bloomberg all matched or advanced that figure the same day^{[1][4][5][14]}.

Here is the part that matters as much as the number itself: nobody who could confirm it has. Nvidia and Hugging Face both declined to comment when asked^[1]. As of Aug. 28, 2026, there is no signed agreement, no filing, no press release — just one anonymous source and a wave of outlets repeating the same figure^[1][3]. TechCrunch, the most cautious of the group, says the deal could still fall apart^[3].

That gap between "agrees to buy," the phrase several headlines used, and "one person familiar with the matter" is the story underneath the story. The Information is a subscription outlet whose business depends on scoops holding up, so its confident framing is also its product^[2]. Everyone else is choosing how hard to lean on borrowed sourcing.

What $12.9 Billion Would Actually Buy

Hugging Face calls itself the "GitHub of AI" — a website where developers upload, download and test AI models, largely for free. The company says its hub holds more than two million models, roughly 1.5 million data sets, and about 1.5 million small AI apps^[8][11]. Most of that is open-weight, meaning anyone can download a model file and run it on their own machine at no cost.

That is why the site matters more than its revenue suggests. It is the shelf almost the entire open-source AI world takes its software off of. Giving away that much storage and bandwidth is also expensive to run, and Hugging Face's own income is modest next to the bill^[8][11].

If the deal closes, it would be Nvidia's largest purchase ever, easily topping the networking company Mellanox, which Nvidia bought for $6.9 billion in 2019^[9]. Nvidia can afford it without blinking: on Aug. 26, 2026, it reported $96.2 billion in quarterly revenue and told investors to expect about $108 billion the next quarter^[10]. At $12.9 billion, this deal would cost Nvidia about 13% of one quarter's revenue^[10].

Nvidia had already tried a cheaper route. Reporting cited by the German outlet heise says Nvidia offered around $500 million in late 2025 at a $7 billion valuation, and Hugging Face turned it down because it did not want one investor holding that much sway^[6]. That was not long ago — Hugging Face's prior funding round, in 2023, had valued it at just $4.5 billion, with backers including Nvidia, Google, Amazon, Intel and AMD^[6]. A full buyout at $12.9 billion would nearly triple that most recent number.

The Software That Runs on Someone Else's Chips

Here is the mechanism the whole fight turns on. Hugging Face is useful precisely because it doesn't play favorites with hardware. Its Optimum software includes versions built specifically for AMD chips, Intel chips and Amazon's own AI chips — not just Nvidia's^[7]. A developer using Hugging Face can pick whichever hardware is cheapest or fastest for their job, and the site will still work.

Nvidia already dominates AI chips largely because of software, not just silicon. Its CUDA platform has spent 20 years becoming the thing AI code gets written for first, and that habit is Nvidia's real moat^[13]. Rival chipmakers in China, including Alibaba's T-Head unit and Huawei, are now building their own open-source software stacks aimed at loosening that grip^[13]. Owning the hub where the models themselves get published would be a defense of that same moat, one layer up the stack.

Critics — many of them rival chipmakers — argue that whoever owns Hugging Face controls small, easy-to-miss defaults: which hardware sample code assumes, which benchmark appears first, which Optimum build gets the most engineering attention^[7]. Those choices could tilt toward Nvidia's own chips over time even without any deliberate decision to break support for competitors. The Optimum builds for AMD, Intel and AWS exist today and need Hugging Face staff to keep maintaining them — staff who would now report to Nvidia^[7].

Nvidia's defenders have a real counterargument, not just a denial. Nvidia is already the single largest corporate contributor of open models on Hugging Face's own hub, which it points to as evidence it benefits from an open platform rather than a closed one^[11]. And a hub full of models that only run well on Nvidia hardware would be worth less to developers, who would have every reason to go somewhere else^[7].

Why Regulators, Not the Companies, May Decide This

A deal this size for infrastructure this central does not just need Nvidia and Hugging Face to agree. It needs antitrust regulators in the U.S., the EU and likely the U.K. to sign off, and that review is where the real argument will happen^[7]. Nvidia has been down this road before, on a much smaller scale.

In its purchase of the AI company Run:ai, Nvidia paid about $700 million — roughly 5% of the price tag on this Hugging Face deal^[7]. Even at that much smaller size, the deal drew an EU referral fight before it was ultimately cleared without conditions^[7]. A deal worth 18 times as much, for a platform this central to how AI software gets distributed, invites a longer and harder look.

The likely outcome, if regulators do intervene, probably isn't a flat yes or no. Binding conditions — commitments to keep supporting non-Nvidia hardware, for instance — are seen as more probable than either a clean approval or an outright block^[7].

A French Company, an American Buyer

Hugging Face was founded by French entrepreneurs, and European coverage has focused less on antitrust mechanics and more on control. If the main distribution point for open AI models ends up owned by an American chip company, European governments and firms would depend on that company's goodwill to keep using models they treat as public infrastructure^[6]. That framing treats the deal as something happening to European tech independence, whether or not that is how Nvidia or Hugging Face's owners see it.

Coverage in the U.S. right-leaning press has taken close to the opposite angle, framing the purchase as an American company buying the world's leading open-source AI hub and citing Nvidia's blowout earnings as proof it has earned the right to make that move^[10][12]. The Daily Caller spelled out the price as "$13,000,000,000" in its headline, a device meant to convey scale more than to add information, and largely left out the neutrality and antitrust questions^[12].

For Hugging Face's own leadership, the calculus looks different again. CEO Clement Delangue has argued that open models are a public good and that restricting them "would hurt first cyber security defenders, startups, small companies, researchers and everyone who's not a frontier lab." Selling to a company with deep enough pockets to permanently fund a platform that gives its core product away for free is one way to keep serving that mission, even if it costs the site its claim to being nobody's property.

What Would Actually Prove This Right or Wrong

Strip away the framing on every side, and what's left is thin. No signed agreement is public. Neither company has confirmed anything^[1][3]. Every dollar figure in this story traces back to one anonymous source at The Information, echoed rather than independently verified by everyone else^[2].

What can be checked holds up: Nvidia has the cash to pay for this many times over^[10]. Hugging Face rejected a smaller Nvidia investment less than a year ago specifically to avoid ceding control^[6]. The hub really does host more than two million models and is the default place developers go to get them^[8][11]. And its non-Nvidia software support is real, current, and requires ongoing work to maintain^[7].

Whatever headline any outlet chose this week, the concrete test of this deal sits two years out, not in this week's coverage. It is whether Hugging Face's AMD, Intel and AWS builds are still being shipped and supported once — if — Nvidia owns the place where they live.