Summary of Key Points
At the 2026 World Robot Conference, the domestic embodied intelligence (perceptible and actionable intelligent robots) industry chain exhibited two major trends: Firstly, manufacturers of complete robots are rushing to go public (such as Yushu Technology and Yunshenzhu), with their products evolving from mere demonstration pieces to practical models capable of working in factories. Secondly, upstream component companies (chips, magnetic materials, automation equipment, etc.) have led the way in achieving growth, becoming the "surefire winners" in this field. However, the industry is still far from widespread commercial use, with the key to breakthroughs lying in the AI's ability to generalize from one situation to another and the accumulation of real-world data.
1. Manufacturers of Complete Robots Rush to Go Public: From Showpieces to Workers
Companies that build the entire body of the robots are currently focusing on going public to raise funds. For example, Yushu Technology has just gone public, while Yunshenzhu and Leju Intelligence are also in the process. Why? Developing robots is costly, and going public allows them to secure more capital to expand their operations. These companies showcased their capabilities at the conference: Yushu's robots danced and boxed, while Yunshenzhu displayed humanoid and quadruped robots. More importantly, they began to use the robots for practical tasks—Leju Intelligence's robots could collaborate on assembly tasks without human intervention. However, the efficiency of these robots is still lower than that of humans, and they need retraining for new tasks, which limits their widespread adoption.
2. Upstream Component Companies Profit First: Greater Stability than Complete Robots
While manufacturers are still in the trial and error phase, upstream component companies have already started reaping benefits. For instance:
- Automation Equipment Company Leisai Intelligence: Its revenue increased by 38%-42% in the first half of the year, with net profits rising by 55%-65%, due to the demand for its motors and dexterous hand components, which are used in 80% of mainstream humanoid robot manufacturers' supply chains.
- Magnetic Materials Company Jinli Yongci: Its revenue from robot motor rotors doubled year-on-year, and it has built specialized automated production lines to supply these core components. Analysts say that the business of component companies is more stable because robots, regardless of the technology they use, rely on chips, motors, and sensors. Additionally, costs are rapidly decreasing due to China's advantageous manufacturing ecosystem. For example, Fuleixin Materials supplies tactile sensors for dexterous hands, and Jifeng Shares has partnered with such companies to develop and sell industrial robots.
3. Widespread Commercial Use Still Lags: Generalization Ability is the Key
Robots currently lack the ability to generalize like humans—they need reprogramming for different environments or tools, which is a major bottleneck for commercialization. Fine tasks like screwing or wiring in factories still require human intervention, and household applications face high costs and reliability issues. Experts suggest starting with internal use by robot manufacturers and high-risk scenarios (such as nuclear facility maintenance) before entering the home market. There is no consensus on the best technical approach (layered large models or end-to-end models), but there is a consensus that accumulating real-world data (e.g., various situations encountered in factories) is essential for improving robot intelligence.
4. Data Collection Becomes a New Frontier: Feeding Robots Experience
To become smarter, robots need extensive data. Companies like Danghong Technology have developed multi-modal terminals that can read instrument readings, detect temperature/gas leaks, and transmit data back to optimize algorithms, creating a "data loop" that enhances robot performance over time. Aobizhongguang has launched a data collection platform that captures data from the robot's perspective and hand interactions, which is sold to companies that need to train models. There are already 64 data collection centers nationwide, with over 2,500 robots in use to gather this valuable information. This data acts as a training tool, enhancing the robots' generalization abilities.
Conclusion
Embodied intelligence is at a stage of transitioning from the laboratory to the factory. In the short term, component companies are the most stable option; in the medium term, the focus will be on mass production and cost reduction to penetrate more applications; in the long term, the goal is to develop robots with generalization capabilities and a "universal brain." Once there is sufficient data and a unified technical standard, the day robots enter households may arrive sooner than expected. (The mentioned companies are used as examples only and do not represent investment advice.)