Hugging Face's Clem Delangue on Open Source AI and the LLM Bubble | MTS Live

The open‑source tradition that built the modern internet is now a battleground for artificial intelligence. The United States pioneered frameworks such as the Transformer, yet recent years have seen many frontier labs hide their most powerful large language models behind proprietary APIs. In contrast, Chinese teams dominate today’s open‑source LLM ecosystem, with projects like DeepSeek and Kimi becoming the default choice for startups and academia worldwide. Clem Delangue argues that this reversal reshapes competition, accelerates innovation, and forces policymakers to reconsider where the next wave of AI talent will emerge.

Delangue also warned that the rapid influx of capital into large‑language‑model services creates a classic bubble. Massive data‑center builds and soaring revenue forecasts mask uncertain margins and long‑term sustainability challenges. While some call for tighter restrictions, he contends that releasing models openly—either via APIs or full weights—actually strengthens security. When defenders have the same tools as attackers, they can build more effective safeguards, turning a perceived risk into a collective defense mechanism that keeps the broader ecosystem safer.

Beyond text, Hugging Face is pushing AI into robotics with the Richie Mini, a consumer robot that already hosts hundreds of community‑built apps. The company’s platform now stores two petabytes of model and dataset files—roughly 500,000 two‑hour movies—demonstrating why it outpaces GitHub for AI artifacts. Delangue envisions deeper US‑China collaboration on open‑source standards, hoping transparency will drive faster adoption while preventing fragmented regulation. As AI moves from screens to physical devices, open‑source infrastructure will be the glue that binds innovation, safety, and global competitiveness.