Nvidia paid $13 billion for Hugging Face — and promised not to close the library

Hugging Face is where most people who run AI models locally go to download them. Its repository holds more than 3 million models, 500,000 datasets, and 1 million applications, used by 18 million individual developers and more than 200,000 companies. Nvidia has agreed to buy it for $12.93 billion — the company’s second-largest acquisition ever, after its $20 billion purchase of Groq’s assets last year.

The deal creates an unusual concentration of power in the AI ecosystem. Nvidia already supplies the chips that most AI systems run on. It also contributed more than 500 models to Hugging Face itself. Now it owns the storefront where competitors, researchers, and individual developers publish and share their work. For users who rely on Hugging Face to find and run AI models without paying cloud providers, the question is whether that access survives the transition.

Nvidia’s CEO Jensen Huang addressed the concern directly. ‘Developers will choose the models they want, the frameworks they want, the clouds and inference service providers they want and the computing platforms they want,’ he said in the company’s announcement. Hugging Face CEO Clément Delangue framed the deal as enabling the scale the platform needs: more compute, more support, more visibility. Neither addressed what happens if Nvidia’s business interests conflict with keeping competitor tools easily accessible.

The skepticism is worth holding onto. Nvidia’s incentive is to increase demand for its GPUs. Hugging Face makes it easier to run models on hardware from AMD, Intel, and Qualcomm as much as on Nvidia’s own chips. Owning the platform gives Nvidia commercial leverage it did not have before — including the ability to package Hugging Face access with its own unused compute capacity for enterprise customers. That is a business model, not a charitable act.

What the deal does not change immediately: Hugging Face operates under its existing structure until the acquisition closes, which is expected in the first half of 2027. Models currently hosted on the platform remain available. Open-source licensing on those models does not change under the acquisition terms alone — though future decisions about platform policies will be Nvidia’s to make.

For developers who use Hugging Face to run local AI without cloud dependency, the concern is whether a chip company with commercial interests in cloud AI infrastructure will maintain the same environment that made Hugging Face the default home for open-source AI. Nvidia says its goal is to grow the platform from 18 million to 100 million users. How it plans to do that profitably without altering what made the platform valuable has not been explained.

The acquisition was confirmed on September 3, 2026. Regulatory review is expected in the European Union and the United States before the deal closes in H1 2027. No changes to the Hugging Face platform or its open-source model hosting policies have been announced.

Tags: nvidia, ai, Hugging Face, Jensen Huang, Open Source, Local AI, Machine Learning