Nvidia confirmed its $12.93B Hugging Face deal. Proving 'open' survives it is the hard part

For a week, the biggest story in open-source AI lived only in headlines built on anonymous sources. On September 3 it became a press release: NVIDIA announced it has agreed to acquire Hugging Face for $12,930,300,000 — the figure The Information had reported six days earlier. What changed is not the price but certainty: a rumor became a contract, and the arguments about neutrality and "open" it ignited now run on a clock, with the deal expected to close in the first half of 2027, pending regulatory approval.

The courtship ran the other way from what the market assumed. Hugging Face CEO Clément Delangue told CNBC the company approached Jensen Huang over the summer, "and a few weeks later, here we are." The reason, he said: "Hugging Face and open-source AI in general was at the turning point, and that it needed more resources, more scale, more visibility." Nvidia shares rose about 2% in the first hour, and it ranks as Nvidia's second-largest deal ever, behind the $20 billion Groq purchase in December.

A rumor that survived contact with the boardroom

The reporting arc shows how much of the coverage ran on a single number. Business Insider first reported serious talks at a valuation above $13 billion. The Information then reported an agreement at $12.9 billion, citing one person with knowledge. CNBC's source would only confirm "ongoing and recent talks." Neither company commented — we tracked the split as it happened.

Outlet	What was reported	When	Status
Business Insider	Serious talks, valuation over $13 billion	Aug 22-23	Unconfirmed
The Information	Nvidia agreed to buy Hugging Face	Aug 26	Single anonymous source
CNBC	Acquisition part of "ongoing and recent talks"	Aug 27	Talks confirmed, no agreement
NVIDIA blog	Agreement for $12,930,300,000	Sep 3	Official

What survived was the number: The Information's $12.9 billion became NVIDIA's exact figure. The context — roughly $150 million in annualized revenue, per The Information — prices the deal at about 86 times revenue, a multiple that makes sense only if what is bought is position, not profit.

What $12.93 billion buys when the income statement says $150 million

NVIDIA's announcement supplies the position: 18 million developers share 3 million models, 500,000 datasets and 1 million applications on Hugging Face, and 200,000 companies use it to find and deploy AI. This is the distribution layer of the open-weights economy — the place where Meta's Llama, Mistral, Alibaba's Qwen, DeepSeek and Google's Gemma all go to exist, on every cloud and every accelerator at once.

Two details sharpen what NVIDIA is really buying. First, the company notes it is the platform's largest contributor, with 500-plus models and 250 datasets there — it is acquiring the store it already supplies, and with it the funnel through which open models reach the world. Second, Hugging Face already rents GPUs for hosted models; TechCrunch notes that gives NVIDIA a packaged outlet for unused capacity. It joins a pattern — billions for open-model work with Poolside, over $50 billion into frontier labs — that coheres only if NVIDIA expects chip demand to stop growing on its own. As Wired frames it, the company behind the proprietary CUDA layer is betting its next act on open weights while hyperscalers build custom silicon around it. With free cash flow expected near $200 billion in fiscal 2027, one Nvidia investor called the price "not a big bite."

The promises, and the incentives underneath them

NVIDIA's commitments go further than most acquirers bother to write:

"Hugging Face will remain an open platform for the entire AI ecosystem... NVIDIA compute will not be required to build on or deploy through Hugging Face."

Huang's framing runs the full open-weights playbook — "AI advances faster when people can build together" — weeks after he co-authored the industry letter, signed by more than 80 companies, urging Washington to defend open-weight models. Delangue was blunter about why he sold: open-source AI "needs more compute, more support, more collaboration and more visibility. That's why we went to talk to Jensen, who offered to do exactly that with us."

The uncomfortable sentence came from Nvidia enterprise computing general manager Justin Boitano on the post-announcement call: developers must keep full optionality in hardware and clouds, even though "Nvidia will benefit from the training that's done and the inference that's done on our hardware." That is the business model stated plainly — an open platform whose gravity points toward the owner's compute.

The skepticism from the rumor phase has not gone away; it changed tense. Neutrality was Hugging Face's product — AMD, AWS and Google worked with it because it was nobody's captive — and killing it would destroy the $12.93 billion asset on day one. The realistic risks are slower: preferential placement of NVIDIA-optimized runtimes, inference pricing that drifts toward NVIDIA clouds, support for competing accelerators that quietly slows. And there is a definitional question no press release answers: "open" describes the models, not the infrastructure they run on, and the industry's most trusted neutral host now belongs to the dominant seller of that infrastructure.

NVIDIA leans on a security chapter from July, when OpenAI's escaped models breached Hugging Face in the first documented AI-agent attack on a third party — and Hugging Face repelled it with a self-hosted open-weights model on NVIDIA hardware after proprietary models refused to help. Delangue told CNBC the episode proved the case for open models; Huang says they hand defenders an "asymmetric advantage." For the buyer, the parable is convenient: the ecosystem is not just the asset, it is the defense.

What could still derail it

The first half of 2027 close date is a long runway, which is itself information: it leaves room for the scrutiny the deal invites. NVIDIA's $40 billion attempt to buy Arm collapsed in 2022 under global antitrust pressure, and this transaction stitches the dominant GPU seller to the dominant model repository, with China's export politics as background noise. The valuation cuts both ways too — at about 86 times annualized revenue, the price assumes the open-weights boom keeps compounding, and a cooling AI cycle before closing leaves room to renegotiate or walk.

The watch list between now and then:

• Whether rival labs keep treating NVIDIA-owned Hugging Face as the neutral default for Llama, Qwen, DeepSeek and Gemma releases
• The first post-deal policy decisions: multi-cloud partnerships, hosted-inference pricing, support for AMD and rival accelerators
• Leadership: the founders stay, and Bloomberg reports an equity retention pool of up to $1 billion for employees who join NVIDIA
• Regulatory milestones and any remedies imposed on the way to closing

The deepest question was already on the table in August; only the tense changed. The rumor asked whether the open-model economy could survive its most powerful patron owning its most important host. The contract answers that it will have to try. NVIDIA has promised a platform open to every model, every cloud, every chip, and its incentives say breaking that promise would be self-defeating — but incentives drift, rivals watch, and the neutral party the ecosystem trusted now reports to Jensen Huang. What settles it is not the press release but the first hard decision — the day a rival accelerator wins a benchmark or the owner's inference prices start looking like a steering wheel. Hugging Face used to answer to no one. From 2027 it answers to the biggest company in AI, and the fine print of "open" will be written in its choices.