After years of developing humanoid robots capable of running, dancing and performing backflips, Chinese robotics companies are shifting their focus toward the software that could make those machines useful in a broader range of jobs. The field, known as Embodied AI, aims to help robots understand their surroundings, learn new tasks and adapt without being programmed for every individual action.
Chinese embodied AI companies raised 93.5 billion yuan ($13 billion) through 322 deals during the first half of 2026, five times the amount raised during the same period a year earlier, according to research firm ITjuzi. Investment is increasingly flowing to specialist developers of robotic intelligence, alongside established hardware manufacturers.
The change reflects growing pressure on robot makers to find value beyond physical machines as competition pushes down hardware prices. Dermot McGrath, founder of Shanghai technology consultancy ZenGen Labs, said companies increasingly believe profits will come from the robot’s “brain” rather than its hardware.
The competition is also becoming part of the broader US-China technology rivalry. American companies have an advantage in some advanced robotic AI systems, while Chinese developers are relying on manufacturing expertise and access to real-world industrial data. The country that most effectively combines AI with machines could gain an edge in automating factories and logistics.
US startups such as Physical Intelligence, Skild AI and Generalist are building systems intended to teach robots multiple skills without extensive retraining. Generalist says its GEN-1.5 model completed about 60% of simple tasks after watching demonstrations lasting between three and 12 seconds. China’s X Square has reported that robots can learn handling skills from a single human video without retraining the model for each task.
Chinese specialists including X Square, Galaxea and Spirit AI have each reached valuations of roughly $3 billion, according to company disclosures. Established manufacturers are also increasing their investment. UBTech advertised a chief scientist role overseeing AI model development with compensation of up to 124 million yuan ($18 million), including equity and benefits. Unitree has about 30 researchers working on AI designed to predict how objects behave, according to a person briefed by the company.
Training robots requires different data from training chatbots. Instead of mainly learning from internet text, robots need examples of physical interactions such as gripping objects, handling flexible wires and moving through workplaces. Chinese companies are collecting this information through human demonstrations, wearable sensors, computer simulations and combinations of these methods.
Spirit AI has hired around 1,000 people to record everyday and industrial activities, its co-founder Gao Yang said. ACE Robotics uses wearable sensors to capture workers’ movements and grip forces in stores and hotels. Robbyant, the robotics unit of Ant Group, said it trained a model with 160,000 hours of material, including 50,000 hours of human-guided robot movements, and used data from 20 robot configurations made by 17 brands.
DexForce is using computer-generated environments to train robots. The company said some industrial tasks can be prepared for deployment within six to eight hours, with success rates above 99.9% in specified precision operations. However, simulated environments cannot fully reproduce physical interactions, according to Jerry Wang of Manycore Tech.
Chinese developers also face constraints that could slow progress. US export controls have limited access to advanced Nvidia chips, while concerns over inflated valuations and overcrowding have made IPO funding more difficult. Zhu Mingxuan, who leads development of JoyIn Technology’s Aether robot AI, said the shortage of advanced chips could become a serious issue as robotic systems grow more capable.
Companies are testing whether their systems can help robots solve unfamiliar problems. Zhu described an experiment in which a robot repeatedly failed to place headphones in a cupboard beyond its reach. After engineers showed it videos of people completing the task, the robot pushed a box toward the cupboard and attempted to stand on it.
Some systems are already being used in controlled industrial and commercial settings. CATL said Spirit AI robots connect flexible wires during battery-pack testing at its Zhongzhou plant, with an insertion success rate above 99%. Robbyant said its robots process orders at two pharmacies in Shanghai, while staff check the medicines before dispatch.
The larger goal is software that can learn new jobs without extensive adaptation and work reliably across different customers and workplaces. Chinese companies are attempting to reduce their computing needs with more efficient models, synthetic data and simulations, while domestic chipmakers such as Huawei develop alternatives to Nvidia hardware. It remains uncertain how much these efforts can compensate for the chip shortage.
Unitree founder Wang Xingxing has suggested that robotics could experience a major capability leap if future AI models can handle a broad range of unfamiliar tasks. He said he would regard that point as reached when robots could complete 80% of tasks in 80% of unfamiliar environments from spoken instructions.
Robotics consultant Georg Stieler cautioned that impressive demonstrations would not necessarily indicate readiness for widespread commercial deployment. Robots may soon handle a broad selection of unfamiliar tasks in demonstrations, he said, but dependable performance for customers could still be years away.
Sources: Investing.com News.