As Model Approaches Converge, Physical AI's Decisive Factor Changes

The article argues that as physical AI technical approaches converge, the industry faces new bottlenecks, specifically 'fractured layers' between models and data, between vision and action, and between simulation and reality. To address these, DeepRoute, a Chinese autonomous driving company, has established Superfluid Lab, led by former DeepSeek core member Ruan Chong. Superfluid Lab focuses on foundation models, VLA models, world models, and multimodal understanding, aiming to create a zero-friction research loop. The lab's approach emphasizes organization-level alignment, where researchers from different fields work together around a unified model, contrasting with traditional product-centric development. This reflects a broader industry shift from product competition to model competition, and now to competition in foundational capabilities and organizational efficiency, similar to OpenAI's early strategy.