BOC International: Moving from “showing ability” to “verifying value”, speeding up commercial verification of robots
The Zhitong Finance App learned that BOC International released a research report saying that the valuation logic of humanoid robots is gradually shifting from technological beta to delivery of alpha, while continuing to be optimistic about core components with high value and ability to be reused across entities. The two conferences further confirmed the bank's previous judgment, that is, 2026 is an important stage in moving from proof of concept to commercial deployment verification; with the introduction of real scenarios and the gradual implementation of orders, the market pricing focus is expected to further shift to order fulfillment, scale delivery, and scenario replication capabilities. As the core component supply system gradually matures, the bank expects that the valuation differentiation of the industrial chain will further accelerate. Upstream core component companies with technical and mass production barriers are expected to continue to benefit from the volume of complete machines, while in-house companies that can steadily replicate display capabilities to real scenarios and continue to fulfill orders and revenue are also expected to receive more continuous valuation premiums.
The main views of BOC International are as follows:
incident
The 2026 World Robotics Conference and the 2nd World Humanoid Robot Games were held in Beijing in August, and the robot industry entered an intensive verification window. The 2026 World Robotics Conference (WRC) was held in Yizhuang, Beijing on August 19-23. With the theme of “Human-robot symbiosis, integration of production and demand”, a total of 373 domestic and foreign companies and 3,000 innovative products were exhibited, and 311 new products were unveiled at the conference. Compared with previous sessions, this conference gave more prominence to actual applications, and scenarios such as manufacturing, warehousing and logistics, catering and retail, medical care, and emergency rescue became the focus of the exhibition. The 2nd World Humanoid Robot Games opened on August 22 at the Beijing National Speed Skating Hall. The competition was held on August 22-26. 666 teams and 2,056 robots from 16 countries participated in 51 events and 1,301 games. The scale of the competition and autonomy requirements were significantly higher than the first.
The bank believes that the two conferences together reflect that the industry evaluation system is shifting from “can do it” to “whether it can complete tasks stably and economically”, but the verification functions performed by the two are different
WRC is closer to commercial testing for real customer needs. The core is to verify whether robots can operate autonomously in factories, warehouses, retail, etc., and form an acceptable return on investment in terms of efficiency, stability, and cost; humanoid robot sports events are closer to engineering stress tests for the body, joints, motion control, and autonomous algorithms. Seen from this perspective, industry competition is further shifting from a single ontological parameter to a “hardware+model+data+scenario” system capability. The bank determined that with the gradual improvement of the core component supply system, the ontology integration threshold is expected to drop marginally, while the ability to truly differentiate will be reflected more in generalization capabilities, closed loop data, large-scale delivery, and stable operation over a long period of time.
The most obvious change in WRC is that the importance of actual work scenarios has further increased, and commercial evaluation standards have begun to shift from demo effects to efficiency and economy
In this year's WRC, “live operation” has become an important display direction for most exhibitors, and robots have further entered practical application scenarios such as transportation, logistics, medical care, and catering. The industry has begun to pay more attention to generalization capabilities, operation accuracy, and operating rhythms under different materials and environmental changes, and public testing and actual deployment data outside the exhibition has also begun to provide quantitative references. For example, independent variable (unlisted) completed a 1-hour public test of fully autonomous logistics sorting on August 12, before the opening of the WRC, achieving 1,816 items/hour, with an accuracy rate of over 98%; Starlight Era (unlisted) previously had a maximum supply volume of 1,200 pieces/hour in the actual deployment of China Post Guangzhou Post Center, and the operating efficiency reached more than 85% of manual labor. The bank believes that the commercialization of robots is gradually entering the stage of quantifiable verification, and the importance of indicators such as success rate, UPH (output units per hour), continuous operation time, manual handover rate, and unit task cost is expected to increase. At the same time, in some logistics and industrial scenarios, if a more simplified hardware solution can achieve better efficiency and cost, large-scale deployment may be achieved faster.
As the ontology and execution hardware continue to be iterated, model generalization and high-quality data are becoming important variables for robots moving from “can do” to wider deployment
Judging from the exhibits and manufacturer presentations at this year's WRC, VLA is still an important engineering direction. Many companies are also exploring the use of world models for environmental understanding, state prediction, and motion planning, and model capabilities have further evolved to cross-task and cross-ontology generalization. G0.5, which was previously released by Xinghaitu (unlisted) in June, unifies heterogeneous motion codecs to make a set of motion expressions cover different robot bodies, reflecting the trend of model reuse across ontologies. On the data side, there has been a marked increase in multi-modal acquisition solutions such as first-view devices, data gloves, and haptics in this year's WRC. Obi Zhongguang (688322 CH/purchase) also focuses on showcasing unstructured data acquisition hardware such as EGO, UMI, WristCam, and Hub. The bank believes that the key to improving the generalization capabilities of robots is shifting towards “whether there is sufficiently high quality data and whether an effective closed loop of data can be formed.” Subsequent competition focuses on whether a closed loop of “deployment - data backflow - model iteration - redeployment” can be formed through actual deployment and efficient simulation. The actual scenario covers a wider range of enterprises with more operating equipment, and it is expected that the ability to generalize the model will continue to improve.
The World Humanoid Robotics Games further enhanced the verification of the ability of software and hardware systems to collaborate
The autonomy requirements for this year's competition have increased markedly. According to the latest rules of the organizing committee, with the exception of the 100-meter obstacle and the 400-meter obstacle, all track races and other events must be completed completely by robots. At the same time, the number of events this year was increased from 26 events in the first edition to 51, including 30 contests and 21 scene events, and high-intensity events such as long jump, weightlifting, tug-of-war, and table tennis were added; the 21 events revolved around nine real scenes of home cleaning, hotel services, industrial production, emergency rescue, book management, supermarket retail, office services, garden inspection, and vehicle charging. The bank believes that high dynamics, continuous motion, and adversarial tasks can further test the performance and reliability of core components such as joint motors, reducers, body structure, and drive control, while autonomy requirements further test environmental perception, decision planning, and motion control capabilities. The industrial value of the competition lies more in providing a high-intensity, repeatable comprehensive software and hardware verification environment than simply breaking sports records.
Supply chain improvement is expected to continue to drive overall machine cost reduction, but competition for core components will also further shift from “presence or absence” to performance, reliability, and mass production capacity
This year's WRC has formed an entire industry chain display covering the whole machine and core components. The upstream categories include sensors, speed reducers, ball screws, motors, batteries, dexterous hands, and integrated joints. The bank believes that improving the core component supply system will help shorten the development and iteration cycle of the main unit and reduce the cost of the whole machine, but it will also reduce the room for differentiation by simply relying on hardware combinations. In the future, upstream competitiveness will be reflected more in performance/cost ratio, mass production consistency, reliability, and cross-entity adaptability; for machine companies, supply chain integration is the foundation for mass production, and the importance of model, data and scenario delivery capabilities is expected to further increase.