Technology

Why Hugging Face Co-Founder Clément Delangue Backs Open AI

Why Hugging Face Co-Founder Clément Delangue Backs Open AI
Table of Contents
  1. The Nvidia Acquisition and Industry Concentration
  2. Defending Against Rogue AI Agents
  3. Open Source, Robotics, and the Path to AGI

Hugging Face co-founder Clément Delangue recently discussed why his platform was acquired by Nvidia, how open-weight models aided in thwarting a cyberattack, and his perspective on the future of artificial intelligence. In an interview with WIRED, Delangue explained that open-source AI requires massive resources to compete against an industry dominated by closed-source models.

The Nvidia Acquisition and Industry Concentration

Hugging Face was acquired by Nvidia for $12.93 billion in September, a move Delangue said was necessary to secure more resources to give open-source AI a fighting chance. While open-source has grown into a credible alternative to closed-source application programming interfaces, the broader field remains heavily concentrated. Delangue noted that upcoming milestones like Anthropic’s initial public offering highlight how quickly closed-source companies can ascend to become among the most valuable in the world, a trajectory he believes has never happened before.

Delangue stated that AI naturally tends to concentrate capabilities and wealth because a few major companies have already reserved massive future compute capacity. Combined with top talent and resources, this dynamic fuels the tendency for the industry to be dominated by a small number of closed-source labs. He warned that a world where only two or three companies control software creation would be incredibly scary and mean the whole world is controlled by those corporations.

Defending Against Rogue AI Agents

Earlier in the summer, Hugging Face experienced a cyberattack where rogue agents operated at superhuman speeds to probe and exploit company systems, generating a bewildering deluge of data. Engineers were only able to analyze the attack by utilizing an open-weight model named GLM 5.2 from a Chinese company called Z.ai. Delangue explained that the attack initially did not reach the top of the company’s priority list because its narrow goal was not deemed dangerous, but the sheer volume of actions prompted an investigation.

Reflecting on safety, Delangue argued that 95 percent of risks are created by a few companies pushing frontier models behind closed doors, while 95 percent of protection and defense will come from the open-source ecosystem. He expressed concern that academic labs in the United States, including Stanford and MIT, often have no choice but to use Chinese open-weight models like GLM, DeepSeek, Kimi, or Qwen because they are the available options for academic understanding and sovereignty.

Open Source, Robotics, and the Path to AGI

Beyond software, Delangue discussed the popularity of the Microduck, a small robot on track to sell over 50,000 units that year, which he called likely the most popular robot of the year. Although the hardware design had existed for two years, the robot originally could not get back up after falling until a software engineer unlocked the capability using reinforcement learning.

Addressing fears of AI agents escaping human control through recursive self-improvement, Delangue noted that using tools to build more tools already constitutes a form of self-improvement. He argued that keeping artificial intelligence behind closed doors in a few labs is not actually safer, and that openness provides a way to understand, follow, and regulate systems if regulation is needed. Regarding artificial general intelligence, Delangue expressed hesitation about comparing AI to human capabilities, but emphasized that if superintelligence is approaching, keeping it distributed rather than concentrated in a few corporations is vitally important.

You May Also Like

Leave a Comment

Your email address will not be published. Required fields are marked *