Hugging Face

• 1. Introduction

  ›

  Hugging Face gets its name from a small emoji. Three French entrepreneurs started it as a consumer app. Within a decade, they were acquiring humanoid robotics startups and working with UNESCO on a free translator covering 200 languages. What happened between those two points begins with a decision about code.

  416173

• 2. From Chatbot to Platform

  ›

  Clément Delangue, Julien Chaumond, and Thomas Wolf were its founders, building the company's first product for a teenage audience: a conversational chatbot. When the underlying model was released publicly, developers and researchers built with it immediately. They showed far more interest in the base technology than in the consumer app it had powered. The founders read that as a signal and changed course.

  The Transformers library, designed for natural language processing applications, became one of the key tools the new platform offered. The platform itself was organized around sharing: researchers and developers could publish machine learning models, share datasets, and show their work publicly.

  The platform that emerged would soon become the backbone of a major open science collaboration.

  43

• 3. BLOOM and BigScience

  ›

  On the 28th of April 2021, Hugging Face launched the BigScience Research Workshop alongside several other research groups. The goal was to build and release a large language model that no single organization could claim to own.

  BLOOM, the model the workshop produced, was announced in 2022. It had 176 billion parameters and was built to work across many languages. At that scale, BLOOM sat alongside the largest proprietary models of the time. Unlike those models, it was released freely to any researcher who wanted to use it. The multilingual design distinguished it from most comparable models, which were built primarily for English.

  The following year, the company's relationship with commercial technology infrastructure would deepen in a new direction.

  567

• 4. Expanding Access

  ›

  Amazon Web Services joined Hugging Face as an enterprise partner in February 2023. The partnership made Hugging Face's products available to AWS customers as building blocks for their own applications. The next generation of BLOOM was planned to run on Trainium, a proprietary machine learning chip built by AWS.

  STATION F in Paris served as the home base for a new AI accelerator program. Hugging Face launched it in June 2024 alongside Meta and Scaleway. The initiative was aimed at helping European startups integrate open foundation models into their products. Selected participants received mentoring, access to AI models and tools, and computing power from Scaleway. The program ran from September 2024 to February 2025.

  On the 23rd of September 2024, Hugging Face joined Meta and UNESCO to launch a free online language translator. Built on Meta's No Language Left Behind model, it offered translation across 200 languages, including many low-resource languages. The project was framed as a contribution to the International Decade of Indigenous Languages.

  The following spring would see the company step into territory it had never entered.

  109141516

• 5. The Pollen Acquisition

  ›

  Matthieu Lapeyre and Pierre Rouanet founded Pollen Robotics in France in 2016. Hugging Face announced its acquisition of the humanoid robotics startup in April 2025, extending the company's reach from software into physical hardware.

  Clément Delangue stated his rationale in a post on X: to make artificial intelligence robotics open source. Where machine learning had developed strong norms around publishing models and sharing code freely, robotics had remained largely proprietary. The Pollen acquisition was the company's opening move in that domain.

  The open-access model the company had built would face an unexpected test before long.

  171819

• 6. Powerful Malware

  ›

  Android devices were the target when hackers hijacked the Hugging Face platform in early 2026. The malware they deployed was described as powerful enough to completely take over a compromised device.

  Open platforms face a structural challenge that closed ones do not: the same access that welcomes researchers also creates pathways for bad actors. The 2026 attack on Hugging Face made that challenge concrete. The platform's value rested on openness; the attack exploited precisely that.

  Malware powerful enough to hand attackers complete control of a device was a problem that open-access principles alone could not prevent.

  20