China's Robot Production Lead Is Really a Data Play

Inside a facility in Shanghai, hundreds of humanoid robots stand idle, waiting for humans to move their arms. Workers in exoskeletons and VR headsets sit at control stations, controlling the machines remotely to generate the movement data that trains the next generation of robots. This is not a research lab. It is a factory floor where the product being manufactured is training data, and the workers doing it are generating the competitive advantage that China is betting on to stay ahead in the global robotics race.

The setup sounds absurd. It is also the spine of a manufacturing strategy that has already produced 10,000 humanoid robots — the most of any company in the world — with the last 5,000 of those completed in just three months. Agibot announced on March 30 that its 10,000th robot had rolled off the production line, crossing a production threshold that Western competitors have not yet reached. But the milestone that analysts who follow the industry are actually watching is not the headcount. It is what those robots, once deployed, will generate in data — and whether that data will compound into a lead that hardware roadmaps alone cannot close.

China dominated global humanoid robot shipments in 2025, capturing roughly 87 percent of the 13,318 units delivered worldwide that year, according to Omdia, a technology research firm. Agibot was the single largest producer, accounting for about 39 percent of all humanoid robot deliveries globally — more than 5,100 units — in a market where China's national target is deploying 100,000 humanoid robots in factories across the country by the end of this year, a goal that reflects government industrial strategy. Agibot's own stated target is more measured: 100,000 robots delivered by the end of 2027, according to reporting from Gasgoo covering the company's April 17 partner conference. Tesla and Figure AI, the two most prominent Western competitors, each delivered approximately 150 humanoid robots in 2025.

Agibot's revenue surged from roughly 60 million yuan in 2024 to about 1.05 billion yuan in 2025, a twentyfold increase year over year, according to the company's disclosed figures. At a tablet assembly line belonging to Longcheer Technology, an Agibot G2 robot achieves a throughput of roughly 310 units per hour, with a task cycle of about 19 to 20 seconds and a success rate above 99.9 percent, according to a case study published by the company. Those numbers describe a company that has crossed a manufacturing threshold — not a research milestone, not a demo, but the ability to put a bipedal robot next to a human on a live production line and have it keep pace.

The Shanghai government co-funded the data collection site where hundreds of robots are teleoperated to produce training data. China Mobile, a state-owned telecom operator, purchased humanoid robots from Agibot for use in data collection facilities. Whether Agibot captures all the data generated by its robots in commercial factory deployments is a question the company has not answered publicly. What is not in dispute is the manufacturing scale it has already achieved and the training data infrastructure it has built to feed its learning systems.

Tesla has begun large-scale preparation for Optimus production at its Fremont, California facility in the second quarter of this year, according to reporting by the Robot Report. Musk himself stated earlier this year that Tesla was targeting 50,000 to 100,000 Optimus units in 2026, according to 36Kr, though on Tesla's most recent earnings call he declined to provide a specific production target for the year. Neither Tesla nor Figure is at the scale that would allow their fleets to generate training data at the rate Agibot's now can.

The gap is structural, not aspirational. Western companies are not building fewer robots because they chose to — they are building fewer because the supply chains, manufacturing partnerships, and factory-floor integration work that Agibot has spent years developing take time to replicate. The question is not whether Western robotics companies will try to close it. It is whether a lead built on real production lines and real data can be closed at all.