Hugging Face Glossary
Hugging Face is a French-American AI company founded in 2016 by Clément Delangue, Julien Chaumond, and Thomas Wolf. The company operates the most-used open-source AI model hub on the internet: a platform where labs, independent researchers, and hobbyists upload pretrained model weights, datasets, training scripts, and demos under various licences. As of 2026, more than a million models and a comparable number of datasets are hosted on the platform.
For this Dictionary, Hugging Face is the infrastructure layer the open-weights ecosystem runs on. When the Dictionary’s Sovereign Compute calculator recommends downloading Llama, Hermes, Mistral, Gemma, Qwen, or any other open-weights model for local inference, the actual download is in nearly every case from Hugging Face. The company’s transformers and datasets Python libraries are the standard tooling for working with these models programmatically.
Hugging Face’s strategic position is unusual: it competes with no major lab directly, charges only for hosted enterprise services and compute, and has consequently become the neutral platform that even commercial-rival labs (Anthropic, Meta, Google) use for some artefact distribution. Its sustained existence is a non-trivial precondition for the Sovereign Compute argument as currently formulated.