Hugging Face

The GitHub Moment That Rewired AI Development

In 2018, three French entrepreneurs in New York made a decision that would quietly reshape the entire AI industry. Their original product — a chatbot app for teenagers called Hugging Face — had failed to find a mass audience. Rather than pivot to a new consumer gimmick, Clément Delangue, Julien Chaumond, and Thomas Wolf did something unusual: they open-sourced the NLP engine powering it, calling it the Transformers library. Within months, researchers worldwide were not just downloading it — they were contributing to it, and begging for a place to share their own models.

That accidental act of openness became the founding thesis of what Hugging Face is today: the definitive infrastructure layer for the global machine learning community. By 2020, they had launched the Hugging Face Hub, a Git-based repository for AI models and datasets modeled explicitly on GitHub's collaborative architecture. By 2023, the Hub hosted over 500,000 models and had attracted a valuation of $4.5 billion — a figure that, as multiple outlets noted, was roughly 100 times its annualized revenue, a testament to how irreplaceable the platform had become.

What Hugging Face Actually Does

Unlike OpenAI or Anthropic, Hugging Face does not build proprietary models for end consumers. Its business is infrastructure: a platform where the world's AI researchers, startups, and enterprises collaborate on open-source models, share high-quality datasets, and deploy interactive demos called Spaces. The Transformers library alone has been downloaded billions of times. Enterprise customers — including Intel, Pfizer, Bloomberg, and eBay — pay for dedicated compute, advanced security, and compliance tooling. Revenue hit $130 million in 2024, up from $70M in 2023 and just $15M in 2022, a trajectory powered by compute monetization (Inference Endpoints, AutoTrain) rather than model subscriptions.

• Model Hub: Over 1 million community-contributed models across NLP, vision, audio, and multimodal tasks
• Datasets Library: 120,000+ curated datasets covering virtually every ML domain
• Spaces: 50,000+ live, interactive AI demo apps hosted directly on the platform
• Transformers Library: The most-starred ML library on GitHub, with 121K+ stars and support for PyTorch, TensorFlow, and JAX
• Inference Endpoints: One-click production API deployment for any model on dedicated GPU infrastructure

In 2025, Hugging Face expanded beyond software with the acquisition of French robotics startup Pollen Robotics and the launch of LeRobot — an open-source robotics ecosystem that includes the $299 Reachy Mini desktop robot, which sold $1 million in units within days of launch. The move signals Delangue's ambition to make Hugging Face the open-source layer not just for language models, but for embodied AI.