Julien Chaumond

Who he is

Julien Chaumond is a French entrepreneur and the co-founder and chief technology officer of Hugging Face, the company he started in 2016 together with Clément Delangue and Thomas Wolf^[1]. Hugging Face runs a platform on which people publish and download machine-learning models and datasets — the part of the company usually referred to as the Hub — and Chaumond's public role has been closely tied to building that platform and to arguing for the open approach it rests on.

Speaking for open models

One of the moments that put that argument in public view was the April 2023 launch of HuggingChat, a chat interface built on open models. At the launch Chaumond described it as the first open-source alternative to ChatGPT^[2]. The distinction he was drawing is a practical one: ChatGPT is reached through a closed API, meaning the company behind it keeps the model itself private and users interact with it only over the network, whereas an open-source alternative lets people download the model and run or modify it themselves. At the same launch he addressed the claim, made by some, that closed APIs were winning, and said that Hugging Face would never give up the fight for open-source AI^[2].

Rebuilding the Hub's storage layer

In August 2024 Chaumond was the author of the post announcing that Hugging Face had acquired XetHub^[3]. In that post he wrote that the new storage backend would help "unlock the next 5 years of growth" of the Hub's datasets and models^[3]. He also explained the reasoning behind the move: since 2020 the Hub had been built on Git LFS, a standard extension of the Git version-control system for handling large files, and the company wanted a storage backend of its own that was more optimized for what the Hub actually does^[4]. Git LFS keeps large files outside the main repository and stores pointers to them, which works well for ordinary code repositories but becomes a bottleneck when a platform is serving very large models and datasets to many users.

XetHub, the company being acquired, was based in Seattle. Its founders — Yucheng Low, Ajit Banerjee and Rajat Arya — had previously built and scaled Apple's internal machine-learning infrastructure^[4]. That background is what made the team relevant to the storage problem: the same engineering questions, how to store, version and serve very large machine-learning artifacts efficiently, come up whether the user is a single company or an open platform with a public catalogue.

Where he sits in the story

As CTO, Chaumond's public footprint is mostly of this kind: explaining technical decisions the platform makes and stating the company's position on openness. The two threads visible in his statements fit together — one is about getting open models into as many hands as possible through products like HuggingChat, the other about the unglamorous infrastructure underneath, the storage and serving layer that determines whether a catalogue of models and datasets can keep growing. Both are the work of a platform company rather than a single model release.