激光雷达赋能物理人工智能数据采集：中国汽车与共享出行

We expect both Hesai and RoboSense's 3Q26 results (to be released in November) to meet reduced market expectations, after seeing weak 3Q earnings from A-share auto parts players. Over the longer term, we see LiDAR - alone or integrated with cameras - collecting data for physical AI training. Data collection is becoming a strategic bottleneck for physical AI because understanding a task is not the same as executing it reliably. On its 3Q26 earnings call, Tesla also sharpened its push to scale Optimus training data, outlining a need for 10K-20K robots to learn through practice. Within the auto supply chain, we highlight LiDAR suppliers as providing the "picks and shovels" for physical AI. Hesai's opportunity set is extending from selling LiDAR mounted on robots to selling its Kosmo 3D world data platform to collecting and generating training data;