At the 2026 Aggregated Intelligence Industry Development Conference, co-hosted by the Aggregated Intelligence Industry Innovation Center and Hubei Sci-Tech Investment in Wuhan's Optics Valley, Deng Kun, Vice President of Application Engineering at BLACK SESAME, stated that with the continuous advancement of edge-side AI, the era of physical AI has arrived.

He explained that physical AI is not simply AI itself, but rather AI combined with a physical carrier, which could be a vehicle, a drone, or a robot. Within the development of these carriers, VLA (Vision-Language-Action) models and world models are evolving in a symbiotic relationship.

In his view, relying solely on VLA models is insufficient for understanding the world and spatial contexts, which is why world models play a crucial role in enhancing VLA development. On one hand, world models enable better simulation of the real world, improving the fidelity of environmental representation and facilitating more effective simulation training for VLA models. On the other hand, world models can act as an engine for VLA advancement, making actions more rational and temporally coherent, addressing the black-box problem, and improving interpretability and safety.

He noted that combining physical AI with world models allows AI chips and AI models to be applied across various industries, such as embodied intelligence and aggregated intelligence.

From the perspective of the chip supply chain, Deng indicated that the industry is shifting from an efficiency-first approach to a security-first one. Previously, the efficiency-first model involved stocking chips based on orders, avoiding inventory accumulation and production delays. However, this approach cannot keep pace with the rapid growth seen in sectors like automotive, robotics, low-altitude economy, and the explosive demand for AI computing servers.

"The previous efficiency-first model no longer aligns with current industry trends. Now, the industry is increasingly adopting safety stock strategies to accommodate the fast-evolving AI landscape while reducing reliance on any single supplier," he said.