Where humanoid robot mass production stalls: Optimus setbacks and the pace of China's supply chain

Humanoid robots are the most watched hardware category in the current AI wave, but they are also the hardest to build at scale. The Information reports that Tesla's latest generation, Optimus V3, faces difficulties with its robotic hand design, production line equipment, and training data collection. Tesla has raised Optimus output to a few hundred units per week. The goal is to push capacity past 1,000 units per week before the end of 2026.

The hand is a major bottleneck. To match the fine manipulation of a human hand, Optimus's hand and forearm contain more than 100 small components, including many screws. Workers still assemble these parts manually. Some robots coming off the line require immediate manual repair. Tactile sensors have also shown reliability issues.

AI capability remains a challenge. One source said the current Optimus still needs to be programmed for specific tasks and operates in tightly controlled environments. This reflects a shared industry problem: how to let robots learn autonomously in the real world and perform a wide variety of manual operations.

China's supply chain is gaining importance in this race. The Paper reports that XPeng's robot has completed supply chain factory audits and designated core component suppliers, advancing its mass production process. On 22 September, XPeng held its first robot supply chain partner conference. Leading companies from the fields of actuators, dexterous hands, sensors, and AI chips attended. The report also noted that Tesla's robot team recently visited Chinese supply chain companies for factory audits. The two sides have significant overlap at the level of core suppliers.

Traditional automakers are entering the space too. Changan Automobile told investors it will accelerate its push into embodied intelligence humanoid robots, flying cars, and other emerging industries. It aims to reach mass production of humanoid robots in 2028. The company outlined a path of using robots to build cars, using robots to sell cars, and putting robots into cars.

There is a sobering reality behind the excitement. On 20 September, China's Zhongqing T800 defeated a human boxer at a robot combat event in the United States. However, an operator wearing a VR headset controlled the fight remotely. The robot itself only handled dynamic balance and posture correction. Demonstrations are still far from genuine general autonomy.