Hugging Face is leaving with NVIDIA

huggingface.co still opens on a Hub. "The AI community building the future." Then a strip under the nav:

We are happy to share our intention to join forces with NVIDIA.

— huggingface.co

That is not a rumour in the sidebar. That is the product telling you who will own the git remote.

Their line for what the site is. Not ours:

The platform where the machine learning community collaborates on models, datasets, and applications.

— huggingface.co

The Hub is git for models. You clone a card. You import Transformers. You pin a Gradio demo to a GPU. Hugging Face was founded in Paris. The software is still there. Independent European software it is not.

Three million cards, one remote

The Models page on 3 September counted 3,038,993 cards. Filter by Text Generation, by Transformers, by whether inference is already wired. Each card is a repository: files, commits, a model card that is supposed to say what the weights will and will not do. The Hub documentation is blunter still:

The Hugging Face Hub is the reference AI platform for open ML. It hosts over 2M models, 1.5M datasets, and 1.5M AI apps (Spaces), all open and publicly available.

— huggingface.co/docs/hub

Transformers is the library that made that remote a default. The docs call it "the model-definition framework for state-of-the-art machine learning models in text, computer vision, audio, video, and multimodal models, for both inference and training." Over a million of those checkpoints sit on the Hub. Pipeline for a quick run. Trainer if you are actually training. generate if you want tokens back. You do not have to like the metaphor. You already typed `from transformers import pipeline`.

Spaces is the demo layer. The page calls itself "The AI App Directory." 1,463,544 Spaces on the day we captured it. Gradio and Streamlit apps you run in the browser. CPU Basic is free. A T4 medium GPU is $0.60 an hour. ZeroGPU hands out an NVIDIA card when the demo needs it, then takes it back.

The invoice is optional. The remote is not.

Public models, datasets, and Spaces stay free. PRO is $9 a month. Team is $20 per user per month. Inference Endpoints start at a few cents an hour on CPU. You can live on the free Hub forever if the repos stay public. What you cannot do is keep the remote European.

Open is not independent

On 26 August 2026, TechCrunch wrote that Nvidia had agreed to buy Hugging Face for $12.9 billion, citing The Information. Reuters ran the same figure that night. NVIDIA later confirmed the deal. Jensen Huang, on the company blog:

Hugging Face will remain an open platform for the entire AI ecosystem.

— blogs.nvidia.com, 3 September 2026

He also promised NVIDIA compute will not be required to build on or deploy through the Hub. Fine. Open weights on a Californian chipmaker's platform is still a Californian chipmaker's platform. EuroMakers lists independent European software. This is no longer that.

Their own Hub docs, still live, make the cut sharper:

No single company, including the Tech Titans, will be able to "solve AI" by themselves – the only way we'll achieve this is by sharing knowledge and resources in a community-centric approach.

— huggingface.co/docs/hub

NVIDIA is the Tech Titan that sells the GPUs those Spaces already rent. Buying the Hub is not sharing. It is owning the git remote.

On 3 September we deleted the listing. Pull request 300. The card that used to call Hugging Face a "European alternative to closed AI platforms" is gone. The Hub you still open is becoming an NVIDIA asset. That is why it is no longer on EuroMakers.

The Artificial Intelligence category still has European work: DeepL for language, PhotoRoom for product shots, Aleph Alpha for private models, Mistral's Vibe for an agent. None of them is a second Hub. There is no second Hub on this site. That is the point of taking this one down.

New to the EuroMakers Journal? Start with how the Journal works.