NVIDIA Sets Its Sights on Hugging Face: $12.9 Billion for a Key Piece of Open AI

NVIDIA has reportedly reached a deal to acquire Hugging Face for $12.9 billion, a move that, if it goes through, would extend the chipmaker’s reach from GPUs and CUDA into one of the most widely used platforms for publishing, downloading, and deploying AI models. The Information reports the deal has been reached, though other reports suggest no definitive contract existed as of recently. NVIDIA and Hugging Face have not officially announced the acquisition, so it cannot yet be considered a done deal.

The possible Hugging Face deal in 20 seconds

• The Information puts the possible deal at $12.9 billion.
• Hugging Face was valued at $4.5 billion in 2023, and NVIDIA already holds a stake in the company.
• The platform occupies a central position in the distribution of AI models and datasets.
• The purchase would extend NVIDIA’s influence beyond hardware and CUDA.
• Official confirmation is still pending, and any deal would still need to close.

NVIDIA’s interest has a logic that goes beyond simply adding another company to its portfolio. Hugging Face has become standard infrastructure for developers, researchers, and companies working with open models. It’s the place where models are discovered, weights are downloaded, datasets are shared, and applications and tools are published.

Controlling that layer would bring NVIDIA even closer to the point where developers decide which model to run and on what infrastructure to run it.

From CUDA dominance to model distribution

NVIDIA has built much of its current position by combining two mutually reinforcing elements.

On one side are its data center accelerators, from the Hopper and Blackwell generations to its newer platforms. On the other is CUDA, the programming environment that has spent years enabling a huge catalog of software optimized for its GPUs.

Hugging Face would add a third layer.

The platform doesn’t manufacture accelerators or develop a single flagship foundation model. Its importance lies in connecting models, developers, and infrastructure.

For a company that wants to deploy Llama, Qwen, DeepSeek, Mistral, or other open models, Hugging Face is usually part of the discovery, evaluation, or distribution process, even if the model later runs on entirely different tools and infrastructure.

That cross-cutting role also explains why a potential acquisition raises questions.

Hugging Face has maintained relationships with numerous manufacturers and vendors. Its 2023 funding round included, alongside NVIDIA, companies such as AMD, Intel, Google, Amazon, IBM, Qualcomm, and Salesforce.

If NVIDIA completes the acquisition, several of those partners would end up relying on a platform owned by one of their main competitors.

Why open models matter so much to NVIDIA

There’s a particularly relevant explanation from an infrastructure standpoint.

NVIDIA’s biggest accelerator customers are trying to develop their own alternatives.

Google has TPUs. Amazon Web Services is developing Trainium and Inferentia. Microsoft has also made progress on its own AI silicon. The major labs are likewise looking to diversify suppliers and reduce training and inference costs.

NVIDIA therefore faces a long-term risk: that an increasing share of AI ends up concentrated within a handful of large vertical platforms capable of controlling the model, the software, the cloud infrastructure, and the accelerators.

The open ecosystem works differently.

A company can download a model, fine-tune it, and run it in its own data center, hire a neocloud, use a European provider, or deploy it on one of the major hyperscalers. NVIDIA’s GPUs still show up in many of those scenarios.

That’s why keeping a broad ecosystem of open models alive can be economically favorable for the chipmaker.

Hugging Face wouldn’t just be another software business. It could become a way to keep demand for compute capacity open outside the large proprietary labs.

The $12.9 billion shows where value is shifting

The reported valuation is hard to justify based on Hugging Face’s current revenue alone.

The company was valued at $4.5 billion in 2023, when it raised $235 million in a round backed by some of the world’s biggest tech companies.

A $12.9 billion price tag would multiply that valuation by roughly 2.9x in three years.

Available reports also put its annualized revenue at around $150 million. If both figures are comparable, NVIDIA would be valuing the company at roughly 86 times that revenue.

That ratio makes it fairly clear the buyer wouldn’t just be paying for current revenue.

It would be paying for Hugging Face’s position within the AI technology stack.

The same phenomenon can be seen in other recent deals. Platforms that connect users and developers with different models are gaining value precisely because they let companies switch providers without rebuilding an application from scratch.

For much of the AI race, the model itself looked like the most valuable asset. Now there’s also a competition to control the interfaces, repositories, and platforms through which those models reach developers.

What would change for AMD, Intel, and other rivals

A potential purchase doesn’t mean Hugging Face would immediately stop working with hardware other than NVIDIA’s.

There’s no information suggesting that would happen.

But ownership does matter.

AMD is growing its presence in AI accelerators with Instinct and its ROCm platform. Intel maintains its own compute technologies, and other chipmakers — including AMD itself, which recently acquired inference-chip startup Taalas — are developing architectures aimed specifically at inference.

Hugging Face gives all of them a channel to make their technologies accessible to developers working with open models.

Under NVIDIA’s ownership, a question of neutrality would inevitably arise: would competing architectures get the same treatment on the platform?

The answer will depend on how NVIDIA manages Hugging Face if the deal is ultimately completed.

There’s even an incentive to keep it open. Turning Hugging Face into a showcase for NVIDIA-only technology could push developers and manufacturers toward alternative repositories, undercutting the very value of the asset being acquired.

Hugging Face’s usefulness to NVIDIA may, paradoxically, depend on it remaining a sufficiently open platform.

CUDA, Hugging Face, and NVIDIA’s new stack

From a technical standpoint, the deal can be understood by looking at the different layers NVIDIA is trying to cover.

Hugging Face would fill a gap where NVIDIA has its own products but lacks a comparable community around a neutral, widely used repository.

Its value also isn’t limited to file hosting. Around the Hub sit libraries, inference tools, datasets, Spaces, and enterprise services that form part of the workflow for many projects.

That’s likely where a significant part of the technological interest lies.

A deal that could also draw regulatory attention

The acquisition shouldn’t be viewed simply as a purchase between two software companies, either.

NVIDIA holds a dominant position in AI accelerators, and Hugging Face is infrastructure used by companies that are competing precisely to reduce their dependence on NVIDIA’s GPUs.

That could draw the attention of antitrust regulators if the deal is finalized — regulatory scrutiny of dominant chipmakers is already a live issue: the FTC is separately investigating Arm’s licensing practices amid the broader AI chip boom.

There’s also a known precedent. NVIDIA announced a deal in 2020 to acquire Arm for $40 billion, but ended up abandoning it in 2022 after facing strong regulatory opposition.

The circumstances are different, and there’s no basis to assume Hugging Face would meet the same fate. But the Arm case showed the difficulties that can arise when a dominant supplier tries to acquire a technology platform used by numerous competitors.

For now, the most basic step is still pending: confirmation.

The Information reports a $12.9 billion deal, while Business Insider had previously reported talks that could still fall apart. NVIDIA and Hugging Face remain publicly silent on the matter.

Until there’s a corporate announcement, it’s more accurate to talk about a possible acquisition or a deal not yet officially confirmed, rather than a completed purchase.

If it does go through, the technological significance will lie less in the $12.9 billion price tag than in the integration of three pieces: the GPUs that run the AI, the software used to program them, and one of the platforms through which millions of developers find the models they’ll later run on that hardware.

Frequently Asked Questions

Has NVIDIA officially bought Hugging Face?

No. The Information reports a deal valued at $12.9 billion, but NVIDIA and Hugging Face have not yet officially announced the acquisition.

Why is Hugging Face important to NVIDIA?

Hugging Face is a widely used platform for distributing AI models, datasets, and applications. Acquiring it would extend NVIDIA’s reach from hardware and CUDA into the distribution of open models.

Would Hugging Face stop working with AMD or Intel?

There’s no information indicating that would happen. A potential acquisition would raise questions about the future neutrality of a platform currently used by different hardware makers.

Why might NVIDIA be interested in protecting open models?

An open ecosystem lets companies run models on different providers and their own infrastructure. For NVIDIA, that means maintaining a broad market of organizations that need accelerators, rather than seeing AI concentrate among a handful of large providers with in-house chips.