China’s Humanoid Robots Face New AI Challenge as Investment Surges

China’s Humanoid Robots : China’s robotics sector is progressing beyond machines that can run, dance, and perform complicated physical tasks. Companies are currently investing extensively in artificial intelligence (AI), which might enable humanoid robots do real-world tasks, learn new skills, and function in unexpected environments.

The drive to embodied AI is gaining financial traction. According to research company ITjuzi, Chinese enterprises engaged in this industry raised 93.5 billion yuan ($13 billion) in 322 agreements in the first half of 2026, five times the amount reported the previous year.

The focus is shifting as competition drives down the cost of robot gear. Companies increasingly want software that allows robots to understand their environment and do many activities without obtaining distinct instructions for each operation.

According to Dermot McGrath, head of Shanghai-based technology firm ZenGen Labs, robot companies have understood that future profit margins may be dependent on AI rather than the robots themselves.

The tournament has worldwide consequences. US businesses have a competitive edge in some advanced robot AI systems, whereas Chinese inventors are using their manufacturing skills and access to actual training environments to fight American rivals.

China’s Humanoid Robots Need Smarter AI Systems

Several Chinese businesses are creating software to help robots acquire and reuse abilities. According to business filings, X Square, Galaxea, and Spirit AI have all been valued at over $3 billion apiece.

X Square presented research indicating that robots can acquire handling abilities from a single human demonstration video without having to retrain the model for each activity. Yang Qian, co-founder and chief operating officer, stated that the company’s objective is to shift robots away from strict, code-based instructions and toward more flexible decision-making.

American firms such as Physical Intelligence, Skild AI, and Generalist are pursuing similar aims. Generalist claims that their GEN-1.5 model performed around 60% of easy tasks after seeing demos lasting three to twelve seconds.

Robotics expert Georg Stieler stated that prominent US research institutions set a high benchmark for how rapidly robots acquire and reuse abilities. He also sees China’s strongest startups as major rivals.

Established manufacturers are also investing in robotic intelligence. UBTech, a Hong Kong-listed company, advertized a chief scientist position focusing on AI model development, offering up to 124 million yuan ($18 million) in pay, including stock and benefits.

Unitree, a major humanoid robot maker based in Hangzhou, has about 30 researchers working on AI that predicts how objects behave, according to a person briefed on the company.

China’s Robot AI Race Depends on Training Data

Teaching robots to execute physical activities needs a different training material than that used to create chatbots. Machines require instances of actions such as grabbing items, manipulating wires, and interacting with their environment.

Chinese engineers are gathering this data through live demonstrations, wearable sensors, and computer simulations. Spirit AI has engaged over 1,000 individuals to record daily and industrial operations, whereas ACE Robotics use wearable sensors to collect worker motions and grip forces in stores and hotels.

Robbyant, Ant Group’s robotics business, said it trained a model on 160,000 hours of data, including 50,000 hours of human-guided robot motions. Its training also included 20 robot configurations from 17 companies, with the goal of employing the same AI system across several machines.

Robot training at DexForce takes place in computer-generated surroundings. Nicole Yang, chief operational officer, stated that the company can prepare some industrial activities for deployment in six to eight hours and has achieved success rates of more than 99.9% in specified precision operations. However, Manycore Tech’s Jerry Wang cautioned that simulations cannot perfectly replicate physical interactions.

Practical trials are now underway. Spirit AI said its robots connected flexible wires during battery-pack testing at CATL’s Zhongzhou factory, with insertion success rates exceeding 99%. Robbyant stated that its robots handle orders at two pharmacies in Shanghai, where personnel inspect medications before distribution.

Despite this development, powerful Nvidia processors are still difficult for Chinese enterprises to obtain due to US export prohibitions. Companies are looking at more efficient AI models, synthetic data, and indigenous chip alternatives, including Huawei technology, but the results are uncertain.

Wang Xingxing, the inventor of Unitree, has claimed that future AI models could lead to significant advances in robotics. He predicted that robots will reach that point when they could do 80% of tasks in 80% of unknown locations with oral instructions.

However, Stieler warned that successful demonstrations do not guarantee consistent commercial performance. The next hurdle for China’s robotics business is to demonstrate that AI systems can perform consistently across several activities, machines, and workplaces.