Nvidia Paid 86 Times Revenue for Hugging Face

Nvidia just spend $12.93 billion on Hugging Face, a company producing roughly $150 million in annualized revenue. The multiple is about 86 times revenue.

It shifts depending on which number counts. Nvidia's SEC filing puts the price paid to Hugging Face shareholders at about $11.9 billion and reserves another $1 billion for equity-based employee retention, which puts the shareholder multiple closer to 79x. Neither version comes anywhere near what $150 million of revenue supports.

The software model is the wrong lens here. The deal makes sense from the other side of Nvidia's ledger, where the operative question is what happens to the Data Center business if AI software closes up.

Hugging Face sits between more than 18 million developers and three million models, 500,000 datasets and a million applications. Nvidia is buying the place where a large share of open AI gets built and handed around, and Nvidia sells the hardware underneath all of it.

The Hugging Face tax

Two developers build the same AI product.

The first one calls a closed model API. The provider decides where inference runs, which accelerators power it, how the infrastructure gets tuned, and eventually whether Nvidia hardware belongs in the answer at all. One company makes that call. Nvidia has to win it against a customer that has both the capital and the chip designers to build its own alternative.

The second downloads an open model, fine-tunes it, and runs it somewhere. That somewhere might be AWS, Azure, Google Cloud, CoreWeave, or a rack in the company's own building. It might run on AMD or on custom silicon, since Hugging Face stays multi-cloud and multi-accelerator and Nvidia has said outright that its hardware will not be required to use the site. Even so, Nvidia has a far better chance of being in that second transaction than the first.

Closed AI concentrates infrastructure decisions inside a handful of model companies, and those are the same companies with enough capital and silicon talent to design their own accelerators. Open models scatter the same decisions across thousands of companies, clouds and individual developers. Nvidia would rather compete for thousands of small decisions than five enormous ones, and Hugging Face keeps them scattered.

What $13 billion has to protect

Nvidia reported $89 billion in Data Center revenue last quarter, up 117% from a year earlier, at a 75% gross margin. Annualized, that run rate is roughly $356 billion.

If owning Hugging Face preserves or creates demand equal to 5% of that run rate, the arithmetic comes out to:

• $4.45 billion in Data Center revenue per quarter
• about $3.34 billion in gross profit at a 75% margin
• about $13.35 billion in gross profit over four quarters

Which covers the purchase price.

Nvidia is not going to book 5% more Data Center revenue because it owns a model repository, and no public evidence suggests otherwise. The arithmetic earns its place for a different reason: it shows how low the bar sits next to the business Nvidia is protecting. A 1% effect on $356 billion is $3.56 billion a year.

On Hugging Face's income statement, $13 billion is absurd. On Nvidia's, it is an insurance premium.

The $10 billion above the software

Hand Hugging Face a generous 20x revenue multiple, the kind reserved for fast-growing software companies with better numbers than these. That gets to $3 billion.

Nvidia committed almost $13 billion. Even under the flattering assumption, the software business explains about a quarter of the price, and the remaining $10 billion buys the repository, the developer relationships, the distribution layer for models, and the defaults a developer reaches for at the start of a project. Call that $10 billion gap the Hugging Face tax. None of it lands in ARR, and all of it matters to the company selling the accelerators those models run on.

GitHub and Red Hat were cheaper

Microsoft paid $7.5 billion for GitHub in 2018, when GitHub was producing somewhere between $200 million and $300 million in annual recurring revenue. Call it 25x to 37x. Microsoft explained the purchase as a way to reach developers and put its tools and services in front of them.

IBM paid about $34 billion for Red Hat in 2019 against more than $2.9 billion of fiscal 2018 revenue, roughly 12x. IBM's pitch was hybrid cloud, and it promised to leave Red Hat independent and neutral.

Both buyers were paying up for a position among developers, and Nvidia is paying more than twice GitHub's multiple and about seven times Red Hat's. Microsoft was after the developers themselves. IBM was after the enterprise open-source layer. Nvidia needs something larger and stranger: open AI has to stay enormous, because the moment it shrinks, infrastructure decisions collapse back into five companies that build their own chips.

Commoditize the complement

Joel Spolsky wrote the rule in 2002: "Smart companies try to commoditize their products' complements." Make the things used alongside a product cheaper and more plentiful, and demand for that product climbs.

Nvidia sells expensive accelerated computing. Models, frameworks and AI software are the complements. Open models are cheap to download, modify and deploy, so thousands of companies can compete at the software layer without first spending a billion dollars to train a frontier model. More companies building means more inference running, and a share of that inference runs on Nvidia hardware.

Nvidia has worked this way for years. CUDA makes its chips easier to program, its own open models push people to experiment, and it publishes those models and datasets on Hugging Face. Buying the site puts it inside one of the largest distribution systems feeding the loop.

The strategy depends on Hugging Face staying neutral, which Nvidia has promised: other accelerators supported, other silicon vendors welcome. An Nvidia-only store would destroy the reason developers show up. The neutrality is the asset.

Nvidia's customers are the competition

The easy mistake is to assume Nvidia's main risk is a rival GPU vendor with a faster part. The structural risk is the five customers who would prefer not to buy general-purpose accelerators at all. Google has TPUs, Amazon has Trainium, Microsoft has Maia, and Meta has been building its own. Each runs workloads large enough that custom silicon pays for itself, which gives all of them a standing economic reason to design Nvidia out.

Reuters described the deal as a bet on open models at a moment when Nvidia's largest customers are building chips to depend on it less, which is about as close as a press cycle gets to naming the real threat.

Five giant proprietary platforms can vertically integrate, each one owning its model, its API, its cloud and its silicon. Millions of developers running thousands of competing open models cannot consolidate that way, and a fragmented buyer base is the one Nvidia sells into best.

Hugging Face does not have to become a $13 billion software business

Judging the deal on Hugging Face's standalone P&L will produce the wrong answer. Subscriptions, enterprise products, inference services and hosted infrastructure can all grow, and Nvidia has plenty of ways to make the business bigger than it is. But Nvidia's Data Center segment generated $193.7 billion in fiscal 2026, and six months later a single quarter reached $89 billion. Hugging Face cannot move numbers of that size and does not need to.

It needs to shift probabilities: a few more companies building on open models, self-hosting a little less painful, model distribution spread across clouds and providers instead of concentrated, and a slower march toward stacks where one company owns the model, the API, the cloud and the silicon underneath it. A few percentage points against $356 billion is worth billions of dollars.

The 86x multiple is the clearest signal in the whole deal that the obvious valuation model does not apply. Nvidia bought the conditions that keep its customers shopping, and it bought them at a price its own quarterly earnings absorb without flinching. Against a Data Center segment already clearing $89 billion a quarter, $13 billion is a line item.