China Mobilizes Millions of Workers to Win Robotics Data Race
Key Takeaways
- ▸JD.com plans to mobilize 500,000+ workers to collect 10 million hours of robotics training data in two years, vastly exceeding U.S. industry capacity
- ▸Chinese companies leverage low labor costs and direct workforce access to build localized datasets, creating competitive advantage in robot development for domestic markets
- ▸Egocentric video data (first-person hand movements) collected in real homes, factories, and retail environments is the new frontier for robot training, moving beyond costly teleoperation
Summary
Chinese tech companies are rapidly scaling humanoid robot development by mobilizing massive local workforces to collect millions of hours of training data in real-world environments—homes, factories, and retail spaces. Companies like JD.com and X Square Robot are leveraging low labor costs and government support to generate egocentric video data showing human hand movements and complex tasks, addressing a global bottleneck in robotics training data.
The strategy differs sharply from the U.S. approach: while American companies outsource data collection to workers in developing countries due to high domestic labor costs, Chinese firms are building vast, localized datasets. JD.com's program in Suqian exemplifies the scale—the company plans to collect 10 million hours of data over two years, eventually involving 100,000 employees and 500,000 external workers. Workers are paid to film themselves doing household chores, while factory workers wear head-mounted cameras and wrist sensors to record assembly-line tasks.
This decentralized data ecosystem provides China with a structural advantage in physical AI development. Locally-collected data better reflects domestic environments and user behaviors, potentially enabling Chinese robots to outperform competitors trained on more generic international data. As robotics development worldwide remains constrained by training data shortages, China's ability to rapidly generate massive, contextualized datasets positions the country to accelerate physical AI commercialization and robot deployment.
- U.S. companies outsource data collection to developing countries due to high labor costs, while Chinese companies build in-house data ecosystems with government support
- This data-collection advantage could help China lead in physical AI and humanoid robotics commercialization within the next 2-3 years
Editorial Opinion
China's human-labor-powered data strategy represents a decisive structural advantage in the race to commercialize physical AI. By mobilizing massive domestic workforces at scale, Chinese companies are solving the training-data bottleneck that constrains global robotics development—a problem U.S. companies are trying to outsource rather than overcome. The real risk isn't just that China will build better robots faster, but that this model fundamentally shifts the labor economics of AI development, making domestic data collection a competitive moat rather than a solved problem.
