A $399 robot duck that became a surprise hit in overseas markets recently has become a symbol of how the global robotics supply chain is intertwined.

Microduck is a 25-centimeter, open-source biped that can be trained with reinforcement learning.

It was developed by a French company, made in China, powered by a Chinese chip, and sold by a robotics team acquired by the United States-based open-source artificial intelligence platform Hugging Face.

The interdependence demonstrated in the duck's supply chain shows that the world's two largest economies remain connected across key parts of the robotics value chain.

In June, US chipmaker Nvidia unveiled an open humanoid reference design whose "brain" is built around its own computing platform and Isaac GR00T robot models. But the "body" comes from China's Unitree Robotics.

Wang Tianmiao, a professor and honorary director of the robotics institute at Beihang University, said that this system-level integration is decisive.

"Nvidia does not need to build an entire low-cost humanoid body from scratch. Unitree does not need to recreate Nvidia's GPUs, simulation tools, foundation models and developer ecosystem. Put the two together, and researchers can move more quickly from an AI model on a screen to a machine walking across a laboratory floor," he said.

"The real contest in embodied AI will not be won simply by producing a flashy robot prototype. Models, data, edge computing, hardware, knowledge of real-world scenarios and a viable business model all have to work together," Wang said.

Tesla's Optimus humanoid also puts that interdependence under a brighter spotlight. Previous media reports disclosed that Tesla officials and engineers have begun auditing parts of the company's Chinese supply chain as it prepares to ramp up Optimus production, though no official confirmation of the development has been available from the Chinese side yet.

Companies identified as potential suppliers included Zhejiang Sanhua Intelligent Controls, Ningbo Joyson Electronics and Ningbo Tuopu Group — all veterans of the automotive supply chain, according to media reports.

The details of Tesla's final supplier list may still be up in the air, but the industrial logic is already hard to change. China spent years building dense clusters of suppliers capable of producing components with high cost efficiency, speed and capacity.

Market consultancy McKinsey estimates China accounts for about 90 percent of global permanent-magnet processing capacity, 40 percent of precision-bearing capacity, 35 percent of motors and 30 percent of power electronics. Many of the components most critical to humanoid robots overlap directly with the EV supply chain, the consultancy said.

Wang He, a Peking University researcher and founder of Chinese robot maker Galbot, has seen both sides of that equation.

Wang He said, "Nvidia has stronger computing resources and a broader large-model stack, while Chinese robot makers can design hardware and entire supply chains around the needs of AI, allowing models, data and the physical machine to be improved together."

He added in an interview that Nvidia has purchased Galbot's robot hardware for experiments, and said the two sides share an interest in areas such as synthetic data for robot training.

McKinsey also said Chinese humanoid makers benefit from dense industrial clusters in real working scenarios, including in Hangzhou and Ningbo in Zhejiang province, Shenzhen, Guangdong province and Suzhou, Jiangsu province, where motors, circuit boards, gear systems, sensors and batteries can often be sourced within relatively short distances. That supplier density can speed up hardware iteration.

Wu Zuxuan, deputy dean of Fudan University's Institute of Trustworthy Embodied Intelligence, has described China's position in the global race as one in which its robot "body" is slightly ahead while the "brain" is broadly keeping pace.

China's biggest advantages are its supply chain and the sheer number of places where robots can be put to work. The next step is to connect those advantages more tightly to AI models so that robot bodies and brains can evolve together, Wu said.

(Source: China Daily)