Japan is building robots to cut China out of its training data

Tetsuya Ogata, a professor at Waseda University, said something unusually candid in Tokyo. Facing prototype machines from seven Japanese companies at the Physical AI Summit, he put it plainly. If geopolitical problems make Chinese data-collection robots unavailable, the development of Physical AI becomes difficult to sustain.

That assessment carries weight coming from a senior academic in what was once called the robot kingdom.

On 16 September the AIRoA association publicly showed prototypes from seven companies competing to develop domestically produced general-purpose robots. The participants include THK, Nachi-Fujikoshi, Keigan, Enacti, Emplifi, HatsuMuv and ugo. The competition runs through prototype, improved and mass-production stages, with the goal of selecting domestic robots suitable for Physical AI data collection and reducing dependence on Chinese hardware. On the same day ugo launched Nova, a domestic semi-humanoid robot, with mass production planned for 2027.

What Japan wants to localise is the data pipeline

The positioning of the competition is clear. It is not about building more attractive humanoid robots to compete with China on complete machines. It is about making sure data collection cannot be choked off.

Japan’s anxiety is specific. Chinese humanoid robot shipments accounted for 97 per cent of the global total in the first half of 2026, and AgiBot’s cumulative output has passed 10,000 units. Japan’s share of the global robot market has fallen from around 80 per cent to about 40 per cent by 2024. That hardware gap can still be closed. Dependence at the data-collection stage is more lethal. If every piece of data used to train a Physical AI model has to be gathered by Chinese-made robots, Japan never gains autonomy in this race.

Put simply, Japan is betting that the endgame of Physical AI competition is not the complete machine but the data. Whoever controls data collection controls the next round of model iteration. That is a different problem from being unable to build robots.

The primary market logic is clear. Japan is using state money to lever a Physical AI data infrastructure sector. Behind the AIRoA competition is a commission from NEDO, the national research and development agency, under its post-5G information and communications system strengthening programme. The approach mirrors the embodied intelligence training-ground system led by China’s CAICT. Build the data infrastructure first and argue about products afterwards.

What travels into listed equities is worth unpacking too. The core constraint on Japanese localisation sits in actuators and sensors. McKinsey stated at the Humanoids Summit Tokyo 2026 that actuators and sensors are the two highest-risk component categories in the humanoid supply chain, and that production of harmonic reducers is concentrated in the hands of Japan’s Harmonic Drive.

That points to a subtle mutual dependence. Chinese humanoid robots also use Japanese components, while Japanese Physical AI depends on Chinese data-collection hardware. Each side holds something the other needs.

This mutual dependence is the real picture of the China-Japan robotics contest today. Among the seven prototypes there is the EmplifAI alliance formed by Mazda and Asratec, THK with its roots in linear motion systems, and long-established specialists in precision reducers. They share one trait. Their accumulation in bottleneck components far exceeds that of Chinese peers, while they have been left far behind in complete machines and data scale.

The real question about Japan’s localisation push is not what the seven prototypes look like. It is whether, three years from now, these companies’ components become a substitute route around the Chinese supply chain. If Japan closes the loop on data-collection robots, its precision component capability connects directly to Physical AI training demand, forming a complete internal cycle that does not need China.

Bluntly, Japan’s move is not about whose robot looks better. It is about being able to play the game without the Chinese supply chain. Japan’s robotics activity is worth watching closely.

Editor’s note: This is an adapted translation of the original OFweek report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://robot.ofweek.com/2026-09/ART-8321206-8420-30703678.html.