Y Combinator CEO Garry Tan is encouraging a more open approach to AI development in the United States, arguing that American labs should be able to distill frontier models as well. His view is that open-weight AI projects can strengthen the domestic ecosystem and expand access to advanced tools.
Distillation is a training method in which one model is queried extensively to help another model learn patterns and reasoning. It is already a common technique in AI research and model building, and Tan sees it as a practical way to support more competitive open-weight alternatives.
His comments come as major AI companies continue debating how much control should exist over model outputs and downstream use. Tan believes that restricting how customers use model-generated intelligence can limit innovation, especially when large models were themselves trained on broad public data.
He also argues that the U.S. should not rely only on a few closed systems. In his view, a healthy AI market needs both frontier labs pushing performance forward and open-weight labs giving developers more flexibility, transparency, and choice.
Tan's message reflects a wider industry question: how to balance scale, access, and competition as AI becomes more central to software, research, and business. If this approach gains traction, it could help shape a more diverse and resilient AI landscape in the years ahead.
