Nvidia buys Hugging Face for $12.9 billion (September 3, 2026) — 86 times revenue
Nvidia confirmed on September 3, 2026 that it will acquire Hugging Face for 12.93 billion dollars, with the definitive agreement signed on September 2. Hugging Face, founded in 2016, is the public repository where open machine learning models are published and retrieved: more than three million models, one million applications, five hundred thousand datasets, and over eighteen million developers using it. Reported annualised revenue is roughly 150 million dollars, making the price about 86 times revenue against total funding raised of more than 395 million dollars. Nvidia is already one of the platform's largest contributors, with over 500 models and 250 open datasets published there. Chief executive Jensen Huang said the platform will remain open to the entire AI ecosystem and that Nvidia compute will not be required to build on or deploy through it. The transaction is targeted to close in the first half of 2027, subject to regulatory approval
The three lines
- Price — 12.93 billion dollars, roughly 86 times an annualised revenue of about 150 million
- Asset — three million models, half a million datasets and over eighteen million developers
- Terms — closing targeted for the first half of 2027, subject to regulatory approval; openness pledged
Key questions
- Why is a chip company buying a model repository?
- **Because it is the place where the output of its chips is published and distributed.** Hugging Face is the de facto standard repository for open AI models. When a research team releases a model, it generally appears there; when a developer wants to run one, that is where they fetch it. **Three million models, one million applications, five hundred thousand datasets and over eighteen million developers** is what creates that position. For Nvidia the platform provides three things. **First, control of the distribution path** — influence over what format a model appears in the moment it enters the world, and therefore which hardware it is optimised for by default. **Second, a channel for selling compute** — cloud capacity sold directly to the platform's users. **Third, demand telemetry** — which models are downloaded, and how often, is a leading indicator of next quarter's chip demand. Nvidia was already one of the platform's largest contributors, with over 500 models and 250 open datasets published there.
- Is 12.9 billion dollars a defensible price?
- **On revenue multiples it is extreme. On strategic position the arithmetic is different.** Reported annualised revenue is about **150 million dollars**, which makes the price roughly **86 times revenue**. In software acquisitions, 10 to 20 times revenue is considered high and 30 times very high. Eighty-six is outside that range. The justification is not the income statement but **irreplaceability**. Hugging Face has raised only about 395 million dollars in total, yet nothing in the open AI ecosystem occupies its position, and there is no obvious substitute. Had Nvidia not bought it, someone else might have, or a rival could have pushed a competing repository and split the standard. **The multiple is better read as the cost of foreclosing that scenario than as a valuation of current revenue.** For scale: Nvidia's fiscal Q2 2026 revenue was 96.2 billion dollars, so the price is about 13 percent of a single quarter's sales.
- Does the platform become Nvidia-only?
- **The company says no, in specific language — but the specific language matters.** Jensen Huang made two statements: Hugging Face will **remain an open platform for the entire AI ecosystem**, and **Nvidia compute will not be required** to build on or deploy through it. The second is the operative one. What is explicitly ruled out is **lock-in**, not preference. Publishing models for AMD or Google TPU will presumably remain permitted; making the Nvidia path faster, better documented or the default is not covered by the pledge. Founder Clem Delangue justified the sale by arguing that open models need **more compute, support, collaboration and visibility** to scale — that is, transferring ownership in order to defend openness. The transaction is targeted to close in the **first half of 2027** and requires **regulatory approval**. What actually happens to the platform's defaults in that period is the answer.
Nvidia confirmed on September 3, 2026 that it will acquire Hugging Face for 12.93 billion dollars. The definitive agreement was signed a day earlier, and closing is targeted for the first half of 2027, subject to regulatory approval.
Reports of the deal had circulated since August 26; September 3 made it official. One number captures its character: Hugging Face's reported annualised revenue is about 150 million dollars. The price is 86 times that.
1. What is being bought
Hugging Face, founded in 2016, is the de facto standard repository for open machine learning models. The comparison used everywhere is that GitHub is to code what Hugging Face is to models.
| Item | Scale |
|---|---|
| Hosted models | 3,000,000+ |
| Applications | 1,000,000 |
| Datasets | 500,000 |
| Developers using it | 18,000,000+ |
| Total funding raised | $395m+ |
| Annualised revenue | ~$150m |
The gap between the last two rows and everything above them is the whole story. Reach and influence are dominant; revenue is not remotely proportional. That is the standard shape of a free public repository, and it is why 12.9 billion dollars cannot be justified from an income statement.
2. Nvidia is not buying revenue
The platform gives Nvidia three things.
A standardised distribution path. Owning the point where new models first enter the world confers influence over the form in which they are packaged, and therefore over which hardware they run best on by default. Notably, the buyer was already inside: Nvidia has published over 500 models and 250 open datasets on the platform, making it one of its largest contributors.
A channel for selling compute. Cloud capacity sold directly to eighteen million developers.
Demand telemetry. Download patterns are a leading indicator of chip demand.
Then there is the cost of not buying. Nothing in the open AI ecosystem substitutes for Hugging Face's position. Had a rival acquired it, or pushed a competing repository hard enough to split the standard, Nvidia would have lost that vantage point. Eighty-six times revenue is more coherently read as the price of foreclosing that scenario than as a valuation.
For scale: Nvidia's fiscal Q2 2026 revenue was 96.2 billion dollars (see "Nvidia Q2 revenue of $96.2 billion"). The purchase price is roughly 13 percent of one quarter's sales.
3. How far does the openness pledge extend
The obvious question is what happens when a neutral repository becomes the property of a chip company.
Jensen Huang's answer has two parts.
- Hugging Face will remain an open platform for the entire AI ecosystem.
- Nvidia compute will not be required to build on or deploy through it.
The second is the substantive one. What it explicitly forecloses is lock-in. Publishing models targeting competing hardware will not be blocked. What it does not address is preference — which formats are supported first, which paths are best documented, what the defaults are set to.
Founder Clem Delangue justified the sale on the grounds that open models need more compute, support, collaboration and visibility to compete with closed ones. The structure of the argument is that ownership was transferred in order to defend openness.
4. Reading it against the last few weeks
This acquisition is not an isolated event. It sits alongside a pattern that has been accumulating.
- August 26 — Marvell fell 8.2 percent despite a reported 120 billion dollar custom chip arrangement with Google, because revenue recognition begins in 2029 (see "What custom AI chips (ASICs) are"). Hyperscalers are moving toward their own silicon.
- August 27 — OpenAI unveiled its own chip, Jalapeño, claiming a 700W part outperformed a 1,200W Nvidia one.
- September 2 — Broadcom guided to 230 billion dollars of AI semiconductor revenue in fiscal 2028 (see "Broadcom's fiscal Q3 2026 results").
As its largest customers move down the stack into their own silicon, Nvidia moved up it — to the layer where models are built and distributed. When hardware competition narrows, position in the ecosystem becomes the defensible line.
5. What is unresolved
- The deal structure has not been published — the cash and stock split, or retention terms for staff.
- Regulatory review remains. The first-half-2027 target assumes US and EU approval. A dominant chip supplier acquiring the distribution point for open models invites conditions, but no authority's position is established.
- The 150 million dollar revenue figure is a reported number for a private company and cannot be independently confirmed.
- The August security incident. Roughly 700 OpenAI agents attacked Hugging Face, reported on August 27 (see "OpenAI agents attacked Hugging Face"). Whether it influenced the decision to sell is not known.
- One thing to watch: after closing, whether publishing, discovering and deploying models for non-Nvidia hardware actually becomes any harder. Huang ruled out lock-in, not preference.
Sources
- TechCrunch — Nvidia confirms it will buy Hugging Face for $12.9 billion
- The Register — Nvidia buys Hugging Face for $12.9B, promises not to squeeze too hard
- Forbes — Nvidia Is Acquiring Hugging Face For Almost $13 Billion
- US Securities and Exchange Commission — NVIDIA CORP Form 8-K (2026-09-02)
- Variety — AI Giant Nvidia to Buy Hugging Face for $12.9 Billion
- CNBC — Nvidia agrees to buy Hugging Face for $12.9 billion, report says