Hugging Face Is Nvidia's $13 Billion Answer to Amazon, Google & Microsoft
 |
| By Jurica Dujmovic |
Nvidia just agreed to pay $13 billion for Hugging Face in a mostly cash deal.
That’s the platform where much of the open artificial intelligence community publishes, evaluates and downloads models.
It’s also in addition to Nvidia’s $3.5 billion purchase of Taiwanese chip designer MediaTek’s bonds.
That isn’t even the biggest checks Jensen Huang has written lately.
AI is the one thing that’s driven the majority of stock market growth this year. But now, poll after poll shows Americans are not just souring on AI … but getting terrified of it.
Meanwhile, a quiet tech revolution is underway. One some experts believe could have a bigger impact than AI or even the internet itself.
Even better? Weiss members will soon get FIRST-DAY ACCESS to claim pre-IPO shares in one of the most important players in this emerging tech revolution.
See how you can claim yours here.
Last December’s $20 billion cash deal to license hardware from AI chip startup Groq holds that title.
The Key Word Here Is Cash
Those multibillion-dollar deals seem staggering.
Yet, when you look at Nvidia's latest results, those recent buys look more affordable.
After all, the chipmaker earned $59.7 billion in its most recent quarter.
So just the Hugging Face purchase alone only amounts to about 22% of one quarter's profit.
Measured against Hugging Face itself, the price looks far more aggressive. The company recently reached approximately $150 million in annualized revenue.
Nvidia is therefore paying roughly 86x revenue and almost 3x the $4.5 billion valuation Hugging Face received in 2023.
Current earnings cannot explain the deal.
The strategic explanation is stronger.
Hugging Face can place Nvidia in front of more than 200,000 companies as its largest customers develop their own AI processors.
Yet Hugging Face became valuable by serving the whole industry.
Nvidia has promised that developers will remain free to use competing chips, cloud providers and inference services.
The acquisition succeeds only if Nvidia can own this important distribution point without making developers feel that it controls their choices.
The Platform Behind the Models
Hugging Face is often described as the GitHub of AI.
Developers use it to find a model, examine its documentation, download its weights, adapt it and publish the result.
Nvidia says the platform now serves more than 18 million developers and hosts over 3 million models, 500,000 datasets and 1 million applications.
Those figures describe a distribution network rather than a conventional software product.
Most of that network remains free.
Hugging Face makes money by selling enterprise collaboration features and computing services to a fraction of its users.
Its current offerings include:
- Organizational plans starting at $20 per user each month,
- Managed GPU computing starting at 60 cents an hour, and
- Access to tens of thousands of models through a unified interface.
The free community creates demand for the paid layer.
Under the transaction disclosed to regulators, approximately $11.9 billion will go to Hugging Face shareholders, while as much as $1 billion will fund equity awards for employees who join Nvidia.
The deal is expected to close in the first half of 2027, subject to regulatory approval and other customary conditions.
The retention package also shows how much of the asset resides in its people and community relationships.
Nvidia Needs More Routes to Customers
Nvidia hardly appears short of demand.
It generated $96.2 billion of revenue in the quarter ended July 26, up 106% from a year earlier.
Data-center revenue reached $89 billion. That includes $48.7 billion from hyperscale customers and $40.3 billion from AI cloud providers, industrial customers and other enterprises.
The size of those numbers, however, could hide a structural risk.
Three direct customers represented 16%, 15% and 13% of Nvidia's revenue during the first half of its current fiscal year.
That’s about 44% combined, according to its latest quarterly filing.
Nvidia cautions that direct customers can differ from the companies that ultimately use or receive its products. So, the filing does not identify the final buyers.
It nevertheless shows that a relatively small number of purchasing relationships can materially affect the business.
Several large technology companies are also building alternatives.
- Amazon markets Trainium for AI training and inference.
- Google offers its seventh-generation Ironwood TPU.
- Microsoft's Maia 200 targets large-scale inference.
- And Meta says it is developing four new generations of its MTIA processors within two years.
None of these programs have stopped Nvidia's revenue from doubling.
Nor do they need to replace Nvidia completely to matter.
Moving recurring, predictable workloads onto internal chips can reduce the amount of business available to Nvidia and give major buyers more leverage when they negotiate prices.
Hugging Face offers a route around some of that concentration.
Its community includes startups, universities, researchers and ordinary companies that will never design an accelerator.
When those organizations turn an open model into a production service, they need chips, networking, optimized software and technical support.
Nvidia can now participate closer to the point where those choices begin.
Neutrality Is the Asset
Nvidia's announcement contains an unusually explicit promise.
Hugging Face will continue supporting every model builder, multiple clouds and multiple accelerators.
Nvidia compute will not be required.
This matters because rival hardware vendors and cloud providers must believe their products can compete fairly on the platform.
Nvidia could damage that confidence without imposing a formal restriction.
Search rankings, default deployment options, benchmark presentation and the timing of support for new hardware can influence what developers choose.
Even subtle favoritism could persuade model creators to publish simultaneously on their own websites, GitHub or competing repositories.
Open model weights can be downloaded and redistributed, leaving Hugging Face with less control over its inventory than a traditional app store has over applications.
The same constraint explains the purchase price.
Nvidia is paying for voluntary participation from millions of developers.
The community cannot be ordered to remain. And squeezing it too quickly for revenue would weaken the network Nvidia wants to acquire.
The company must improve Hugging Face's reliability and enterprise services while keeping the free layer useful and the hardware choices credible.
How the Deal Can Pay Off
There are several realistic ways to earn a return without forcing customers onto Nvidia hardware.
Hugging Face can sell more private repositories, security controls and managed inference to its large organizational user base.
Nvidia can ensure that new open models run well on its software from the day they appear.
It can also make the path from experimentation to production shorter for companies that already want Nvidia infrastructure.
The broader benefit is defensive.
Open models allow companies to download and modify AI instead of buying every query from a proprietary model provider.
That can spread AI adoption across a much larger group of businesses.
More deployed models create more demand for computing. Even when no single model company dominates.
Nvidia's hardware business benefits from that expansion …
As long as its chips remain the preferred place to run the workloads.
This makes the acquisition a form of insurance against a fragmented AI market.
Nvidia does not have to predict whether OpenAI, Anthropic, Meta, DeepSeek or another developer will produce the most popular model.
Hugging Face gives it exposure to many of them and to the derivative models that users create.
What Investors Should Watch
The platform may help Nvidia remain central even as the model layer changes quickly.
- The first test is community growth.
Model uploads, active developers and participating companies should continue rising after the acquisition closes.
A slowdown relative to competing repositories would suggest that neutrality concerns are becoming costly.
- The second is commercial conversion.
Nvidia should eventually disclose whether Hugging Face's enterprise and computing revenue is growing quickly enough to justify more of the purchase price.
At $150 million in reported annualized revenue, even rapid growth will take time to become material beside Nvidia's $96 billion quarterly business.
- The third is equal treatment across hardware.
Investors should look for continued support of AMD accelerators, Google TPUs, Amazon Trainium and other platforms in Hugging Face's deployment tools and evaluations.
Broad compatibility may appear counterintuitive for Nvidia.
But it is essential to preserving the marketplace it has agreed to buy.
- Finally, investors should follow Nvidia's non-hyperscale data-center revenue.
The company's AI cloud, industrial and enterprise category grew 138% from a year earlier to $40.3 billion last quarter.
If Hugging Face helps that customer group expand faster, the strategic return could matter far more than the acquired company's standalone revenue.
A Costly Test of Restraint
Nvidia can easily afford Hugging Face.
Affordability, however, does not make an 86x-revenue purchase cheap.
The company is wagering that a central position in open-model distribution will protect and expand a hardware franchise that produced $89 billion of data-center revenue in just one quarter.
The investment case rests on restraint.
Hugging Face becomes more useful to Nvidia when developers trust it to remain useful to Nvidia's competitors.
If that trust survives, the acquisition can connect Nvidia to a much broader customer base.
It could also keep its technology close to the moment companies choose how to deploy AI.
But if the platform begins to resemble a captive Nvidia storefront, the community can gradually take its models and attention elsewhere.
Investors should therefore judge the deal through two numbers:
- Hugging Face's reported $150 million revenue run-rate, and
- Nvidia's $89 billion quarterly data-center business.
The first explains why the purchase looks expensive. The second explains why Nvidia is willing to pay.
Best,
Jurica Dujmovic
P.S. Nvidia has delivered record financial results this year. But its stock gains, while impressive, haven’t made similar records.
However, there is a new tech revolution underway. And one of the most intriguing ways to play it is with a company that hasn’t yet gone public.
And you don’t have to wait much longer to claim your own pre-IPO stake. Details here.
