The Idea

In 2016, three Frenchmen sat in a New York lecture hall taking an online Stanford engineering course, surrounded by a study group three dozen strong they'd assembled themselves. Clément Delangue, a former teenage eBay power-seller turned startup product guy, had reconnected with Julien Chaumond, an elite mathematician then working inside France's economic ministry. Through the group they met Thomas Wolf, a trained scientist turned patent lawyer who happened to play in a band with Chaumond. By the time the class ended, the three had decided to build something together.

Their target was deliberately unserious. Siri and Alexa existed, and the founders found them lifeless. "They were extremely boring," Delangue said. "They would only do productivity stuff." So they set out to build the opposite: a fun, open-domain chatbot for teenagers, an "AI BFF" you could talk to about anything. They named it after their favourite emoji, the hugging face, and quietly told people they wanted to be the first company to IPO with an emoji instead of a three-letter ticker.

This was the beginning of Hugging Face.

The Execution

• 2016: Delangue, Chaumond and Wolf founded Hugging Face in New York City as a chatbot app aimed at teenagers, named after the 🤗 emoji. Early backing came via the Betaworks accelerator, which seeded the company and gave it a US base.
• Late 2018: Google released BERT, a new language model, and the team shipped an open-source PyTorch port within a single week. That weekend sprint, not the chatbot, is the moment most often cited as defining the company's future.
• 2019: They killed the consumer chatbot and pivoted entirely to open-source machine learning infrastructure. The Transformers library became the flagship, and the developer community made it the default toolkit for modern NLP.
• 2020: Launched the Hugging Face Hub, a GitHub-style repository for models, datasets and demos, cementing the platform's role as the town square for AI builders.
• May 2022: Raised a $100M Series C led by Lux Capital, with Sequoia and Coatue, after open-sourcing PyTorch BERT four years earlier and growing to 100,000+ hosted models.
• 2021 to 2022: Began monetising for the first time, posting roughly $10M of revenue in year one and about $15M the next, built on paid individual and team plans plus managed enterprise deployments.
• August 2023: Closed a $235M Series D at a $4.5B valuation, with Google, Amazon, Nvidia, Salesforce, AMD, Intel, Qualcomm and IBM all on the cap table, as ARR reached an estimated $70M.
• April 2025: Acquired French humanoid-robotics startup Pollen Robotics, maker of the Reachy robots, extending the open-source playbook into embodied AI and proprietary robotics data.
• June 2026: Delangue announced the company had crossed $100M ARR, serving 50,000 organisations and 13M+ users while keeping the platform free for 97% of users and monetising only the 3% who need enterprise compute and private hosting.

The lesson?

The product that made Hugging Face was the one they nearly gave away by accident. They set out to win teenage attention spans and instead built better internal tooling, and when the developer community wanted the tools far more than the chatbot, they followed the signal rather than the original pitch. The deeper bet was counterintuitive: give 97% of users the platform for free, capture nothing from them, and own the distribution layer every model has to pass through.