Embodied artificial intelligence companies in China attracted investments totaling 93.5 billion yuan ($13 billion) across 322 deals during the first half of 2026, in an accelerated shift from kinetic displays toward building software minds for robots.
China’s embodied AI funding surge hits $13 billion
Challenges of transitioning from motor skills to physical perception and interaction
Data collection methods and initial trials in factories and pharmacies
Competitive comparison between China and the United States and future standards
Frequently asked questions
China’s embodied AI funding surge hits $13 billion
An extensive investigation published by Reuters examined the radical transformation underway in China’s robotics sector, where companies are moving beyond simply showcasing humanoid robots capable of dancing or performing acrobatic movements to developing the artificial intelligence systems necessary to perform useful and practical work. The report revealed that approximately 93.5 billion yuan, equivalent to $13 billion, was invested in Chinese embodied AI companies across 322 deals during the first half of 2026.
Data firm IT桔子 estimated that this amount equals roughly 5 times the total recorded in the corresponding period of the previous year, noting that these figures describe historical funding recorded in the first half of the year rather than funds raised in a single day.
Challenges of transitioning from motor skills to physical perception and interaction
Embodied artificial intelligence combines perception, planning, and motor execution capabilities within real physical environments; a robot may possess arms mechanically capable of precise movement, yet it still struggles to interpret unfamiliar objects or respond appropriately to changes around it. Unlike software agents operating in purely digital environments, physical machines must contend with the laws of friction, weight, obstacles, humans, and the consequences of physical errors.
Reuters interviewed developers at prominent Chinese companies including X-Square, Spirit AI, and Galaxea, where engineers gather training examples from live human demonstrations, wearable sensors, factory environments, and virtual simulations to train models capable of generalization and moving beyond pre-programmed movements.
Data collection methods and initial trials in factories and pharmacies
Data quality remains a major challenge for developers; repeating a demonstration of a specific task does not guarantee the model’s ability to safely perform the same task under different lighting conditions or using alternative tools. Some companies report conducting field trials in factories or pharmacies, including industrial activities associated with major manufacturers.
Nevertheless, these trials remain limited deployments or company-announced tests, and do not constitute conclusive proof that humanoid robots are capable of performing any office or industrial task completely independently. Furthermore, computer simulation reduces some training costs, but the physical world differs in ways that are difficult to model completely.
Competitive comparison between China and the United States and future standards
China may hold competitive advantages through its robust manufacturing supply chain and direct access to industrial environments, while American companies may retain strengths in certain foundational models and advanced computing systems. Both evaluations depend on technological evolution and market conditions.
The significance of this report lies in its departure from promotional displays to examine data collection mechanisms, investment economics, and the actual capabilities required for commercial robots. Future evidence will consist of independent success rates, safety standards, and the presence of paying customers for useful services in unfamiliar environments.
The field investigation highlights the technical gap separating dazzling visual displays from true industrial readiness. Over past years, many Chinese companies successfully developed mechanical structures and joint motors enabling humanoid robots to walk, jump, and dance with striking balance. However, transitioning to actual work inside factories or pharmacies requires embodied AI models capable of understanding changing physical environments and handling immediate interactions with objects varying in shape, weight, and texture.
For this reason, the 93.5 billion yuan investments across 322 deals in the first half of 2026 focus on building the “software minds” and kinetic-sensory data collection platforms utilized by companies such as X-Square, Spirit AI, and Galaxea. Despite the advantages provided by Chinese supply chains and extensive industrial environments for prototyping, the decisive benchmark for this sector’s success remains contingent upon proving independent competence, safety standards, and commercial viability outside controlled testing environments.
Frequently asked questions
Question: What was the volume of investment in embodied AI companies in China during the first half of 2026?
Answer: Investments reached approximately 93.5 billion yuan ($13 billion) distributed across 322 deals, which equals roughly 5 times the corresponding period of the previous year according to IT桔子 estimates.
Question: What is the difference between embodied AI and traditional software agents?
Answer: Embodied AI combines perception, planning, and action in physical environments, and must deal with weight, friction, obstacles, and humans, unlike software operating strictly digitally.
Question: How do Chinese companies like X-Square, Spirit AI, and Galaxea collect training data?
Answer: They collect data through live human demonstrations, wearable sensors, factory work environments, as well as virtual simulation environments.
Question: Do current trials in factories and pharmacies prove humanoid robots have achieved complete autonomy?
Answer: No, announced trials remain limited in scale, are subject to controlled conditions, and do not imply that humanoid robots can execute various work tasks independently and safely in unfamiliar environments.