By Hamza Faisal

China's humanoid robots can already run, dance and perform backflips. The harder problem — and the one now attracting billions — is teaching them to think. According to a Reuters report, Chinese companies are pouring resources into embodied AI, the software that lets robots understand their surroundings, learn unfamiliar tasks and adapt without being programmed for every job.

The money has followed the shift. Chinese embodied AI companies raised 93.5 billion yuan, about $13 billion, across 322 deals in the first half of 2026 — five times the amount a year earlier, according to data firm ITjuzi. Funding is increasingly reaching specialist developers of robot intelligence, not just the manufacturers building the machines.

As hardware prices fall, the economics are flipping. Robot makers have "concluded that the margin won't come from the hardware but from the brain," said Dermot McGrath, founder of Shanghai-based consultancy ZenGen Labs.

The race for intelligence

American startups still set much of the pace. Physical Intelligence, Skild AI and Generalist are building software meant to teach robots different skills without extensive retraining; Generalist says its GEN-1.5 model completed around 60 percent of simple tasks after watching a demonstration lasting just seconds. China's X Square has published research showing robots picking up handling skills from a single human video.

"We want robots to shift from being driven by code to being driven by common sense," said X Square co-founder and chief operating officer Yang Qian. Specialist Chinese startups including X Square, Galaxea and Spirit AI have each reached valuations of roughly $3 billion, according to company disclosures.

Established manufacturers are investing too. Hong Kong-listed UBTech advertised a chief scientist role overseeing AI model development with pay of up to 124 million yuan, about $18 million including equity. Unitree, a leading humanoid robot maker, has about 30 researchers working on AI that predicts how objects behave.

The training race

Unlike chatbots, which learn from vast amounts of internet text, robots need examples of physical interaction — gripping objects, handling flexible wires, moving through real spaces. Chinese companies are attacking that data problem in different ways: some hire people to demonstrate tasks, others use computer simulations, and some combine both.

Spirit AI has hired around 1,000 people to record everyday and industrial activities. Ant Group's robotics unit Robbyant says it trained a model on 160,000 hours of material, including 50,000 hours of human-guided robot movements. DexForce takes the simulation route, claiming it can prepare some industrial tasks for deployment within six to eight hours.

But simulations cannot perfectly reproduce physical contact, X Square's Yang said. Chinese firms also face shortages of advanced Nvidia chips under US export controls, and a squeeze on IPO funding amid concerns over inflated valuations.

A 'ChatGPT moment' for robots?

Unitree founder Wang Xingxing said this month that the company had tested OpenAI's GPT-6 on robots with encouraging results, and suggested GPT-7, if released next year, could bring robotics its own "ChatGPT moment." He said he would consider that threshold reached when robots can complete 80 percent of tasks in 80 percent of unfamiliar environments from spoken instructions.

Others urge patience. A breakthrough in demonstrations does not necessarily mean robots are ready for broad commercial use, consultant Georg Stieler said: "If the question is when a customer can count on it, I would add years, not months." Early deployments are already working: Spirit AI robots wire battery packs at CATL's Zhongzhou plant with success above 99 percent. Whoever wins that race will own the most valuable part of the robot.

By Hamza Faisal