Hugging Face's CEO on Open Source AI, Model Routing, and Competition
Clément Delangue, co-founder and CEO of Hugging Face, joins the a16z Show (with hosts appearing to be Theo Jaffee and Sofia Puccini) to discuss the state of open source AI. He argues that open source AI is inherently safer than proprietary frontier models because the open source community tends to build specialized models rather than those trained on dangerous capabilities like cyber security. He contrasts this with closed frontier labs that have historically marketed risk (e.g., GPT-2 being "too dangerous to release") and now face new government scrutiny, including reported U.S. government involvement in restricting the release of GPT-5. Delangue reveals Hugging Face has crossed $100 million in annual recurring revenue, which he frames as validation of the open source business model. He highlights surging interest in local models—running on phones, laptops, and devices like Mac Minis—for privacy-sensitive use cases (health, corporate data) and sustained agentic workloads, driven by Hugging Face's llama.cpp library and models like GPT-OSS, Qwen, and Llama 4. He pushes back on Anthropic's recent accusation that Alibaba engaged in distillation "attacks," calling distillation a standard practice used across the industry and arguing frontier labs already dominate value capture as the fastest-growing companies in the world. He frames the real risk as too little competition, not too much, and advocates for a maturing AI ecosystem built on model routing, open source, and specialized models rather than monolithic proprietary APIs. He also discusses Europe's potential to build a frontier lab through ecosystem investment.
Original sources
Podcast The a16z Show
Episode:Hugging Face's CEO on Open Source AI, Model Routing, and the Future of Competition ↗
AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.