So, yeah, a Chinese robot humanoid smashed the 100-meter race record set by the human Usain Bolt in the recently concluded World Humanoid Robot Games in Beijing. But unlike Usain Bolt, the Tiangong Ultra humanoid didn’t know when to stop. The robot finished the race by crashing into a barrier and had to be stretchered off.

This and other robot mishaps produced a highlight reel of chuckles mixed in with images of robot hordes marching straight out of a Star Wars sequel. No, these aren’t the droids you’re looking for as Jedi master Obi-Wan Kenobi would say.

So what is the real state of China’s humanoid development? The robots are hungry.

That’s the conclusion of China’s influential National Development and Reform Commission (NDRC) which decried an industry that is rushing herd-like into development and that is already crowded with a reported 400 models that are often imitations of each other. What China’s robots need is a fresh diet of training data that reflects authentic settings and closed-loop applications driven by real demand.

The NDRC says the next step is to build a high-quality real-machine data collection system that will address the “data hunger” currently constraining development. This means the NDRC will emphasize world models and vision-language-action models in addition to expanding the library of embodied-training scenarios. The goal is to speed practical deployment of humanoids in fields like manufacturing, healthcare, consumer markets, services and public safety. No more fun and games seems to be the message.

That message was seemingly echoed at the World Humanoid Robot Games by the CEO of Unitree Robotics, a star of China’s humanoid scene. Unitree made its much anticipated stock market debut during the Games but CEO Wang Xingxing sought to curb expectations almost immediately.

Xingxing expanded on the “hungry robot” theme by noting that Unitree’s robots remain less efficient than human workers and must be retrained from scratch for each new task. The inability of robots to generalize across tasks and environments is the main obstacle facing the entire robotics industry, said Xingxing.

Xingxing says robots will have their “ChatGPT moment” when they can be placed into a new environment like a person’s home and complete 80 percent of tasks through text or voice commands alone. That moment could arrive in two to three years if things move fast or in five to ten years if they move slowly. The bet may be on quick as China reportedly already has 53 specialized training centers with another 34 planned or under construction. A significant portion of robot sales reportedly go to these state-run training centers. Chinese companies also are paying hundreds of thousands of workers to record “egocentric” data that documents their movements for humanoid training purposes.

Words of caution haven’t slowed down investor interest, however. For example, Xpeng, a Chinese car manufacturer, says it just raised an additional $900 million in support of its Dogotix robotics unit, putting the company’s overall valuation up to $6 billion. Xpeng is notable in that it offers both male and female versions of its IRON humanoid, anticipating the day when designing robot clothes become an actual profession (as per Isaac Asimov’s 1976 “The Bicentennial Man” science fiction tale).

Growth is so fast-paced that overall projections for 2026 Chinese robot sales are fluid. Some suggest Chinese manufacturers may ship as many as 60,000 units in 2026 and may control 90 percent of the market despite an import ban from the Trump administration. However, the Trump administration says it will spend $2 million for Boston Dynamics robot dogs to be used by the Immigration and Customs Enforcement (ICE). Boston Dynamics is a subsidiary of South Korea’s Hyundai entity but it appears “Spot” may benefit from a “grandfather clause” as the robot dogs already are used by the Secret Service. The U.S. robot import ban extends to new versions of Chinese non-humanoid robots that include robotic vacuum cleaners made by Roborock and others.