NVIDIA acquires Hugging Face for $12.93 billion

NVIDIA has signed an agreement to acquire Hugging Face for $12.93 billion. The platform is expected to remain open, multicloud, and compatible with competing chips.

The leading hub for open models could soon come under the control of the leading supplier of AI chips. NVIDIA has agreed to acquire Hugging Face for $12,930,300,000, making it one of the largest transactions in the company’s history.

Jensen Huang’s wording matters: NVIDIA has “agreed to acquire” Hugging Face. The acquisition has not yet been presented as complete. The official announcement provides no closing date, list of required regulatory approvals, or details about the conditions that could prevent the transaction from going ahead.

For its part, Hugging Face confirms its “intention to join forces with NVIDIA” through a banner displayed on its homepage. The company has not yet published a separate announcement detailing its future governance, the roles of its founders, or how its teams will be integrated.

According to information obtained by Reuters, approximately $11.9 billion will go to Hugging Face investors. An equity-based package worth up to $1 billion would be used to retain employees joining NVIDIA. Jensen Huang’s announcement does not detail this breakdown.

Founded in 2016 by French entrepreneurs Clément Delangue, Julien Chaumond, and Thomas Wolf, Hugging Face initially developed a conversational app before becoming a central platform for building, publishing, and deploying models.

Its platform hosts models, datasets, and applications accessible through Spaces. It also develops widely used libraries such as Transformers, Diffusers, Datasets, Safetensors, Tokenizers, and Accelerate, alongside paid services for businesses, storage, and inference.

NVIDIA says more than 18 million developers, researchers, and creators now use Hugging Face. The company cites more than 3 million models, 500,000 datasets, 1 million applications, and 200,000 corporate users.

These figures come from the companies involved and do not distinguish between active accounts, paying organizations, and occasional users. Hugging Face’s homepage still separately displays more than 2 million models and over 50,000 organizations. The discrepancy may reflect different definitions or update schedules.

The acquisition price is nearly three times Hugging Face’s last publicly disclosed valuation. In August 2023, the company raised $235 million at a valuation of $4.5 billion. NVIDIA already participated in that round alongside investors including Salesforce, Google, Amazon, AMD, Intel, and IBM.

Hugging Face reportedly rejected a $500 million investment offer from NVIDIA in 2025 that would have valued the company at approximately $7 billion. The chipmaker is now moving from minority shareholder and technical partner to prospective owner.

The purchase price cannot be explained solely by the platform’s current revenue. Before the acquisition was confirmed, Reuters reported annualized revenue of approximately $150 million. Based on that unaudited figure, NVIDIA would be paying roughly 86 times annualized revenue.

That valuation suggests the buyer is seeking more than software or an immediate source of income. Hugging Face sits between model creators, the companies evaluating their work, and the providers that run it. Its search engine, model pages, leaderboards, libraries, and deployment services directly influence which technologies developers adopt.

NVIDIA already controls a substantial portion of the physical infrastructure used to train and run AI systems. Hugging Face brings the company closer to the point where users choose a model, compare its performance, download its weights, or select an inference provider.

The acquisition therefore adds a distribution layer to NVIDIA’s business. The company would no longer only provide the machines on which models run. It would also control one of the main places where those models are presented, documented, tested, and deployed.

That position is becoming more strategic as several of NVIDIA’s largest customers develop their own components. Google operates its TPUs, Amazon offers Trainium and Inferentia, while Microsoft, Meta, and OpenAI are pursuing different technologies intended to reduce their dependence on the dominant supplier.

Open models provide NVIDIA with another route to growth. Unlike closed services, they can be downloaded, customized, and deployed across a wide range of infrastructure. Each new project nevertheless creates demand for computing capacity that NVIDIA can attempt to convert into hardware sales, rented capacity, or use of its software tools.

NVIDIA says this commercial logic will not turn Hugging Face into a storefront reserved for its own products. Jensen Huang states that developers will remain free to choose their models, frameworks, inference providers, cloud services, and computing platforms. NVIDIA hardware will not be required to build or deploy a project through Hugging Face.

The company also promises to preserve multicloud development and support for several families of accelerators. Models from every developer should remain publishable, including those optimized for competing hardware.

This commitment addresses the transaction’s central risk. Hugging Face hosts organizations such as AMD, Intel, Google, Amazon, Microsoft, and Meta, several of which develop technologies that compete directly with NVIDIA’s. The platform’s value depends on its ability to serve the entire market without appearing to function as the commercial property of a single supplier.

Technical compatibility alone will not guarantee neutrality. A platform can continue supporting a competitor while giving its parent company’s products greater visibility, stronger integration, or faster optimization.

Future decisions concerning recommended models, search rankings, leaderboards, default providers, export formats, and inference services will therefore receive close attention. NVIDIA has not announced an independent governance mechanism capable of reviewing those choices.

The company has also made no measurable commitments concerning prices, free usage limits, API continuity, or support timelines for