BOCI has released a research report stating that humanoid robots, as the ideal carrier of embodied intelligence, have crossed the threshold for mass production in 2025. The primary bottleneck hindering their large-scale commercial deployment is the insufficient generalization capability of large models. The firm believes that with the rapid advancement of models, computing power, and data, this issue is likely to be progressively resolved. Leveraging unique policy support, a deep industrial chain, and cost advantages, China's supply chain holds a leading global position and is poised to benefit significantly from the large-scale rollout of humanoid robots. The report recommends focusing on core component segments that are expected to deliver orders and revenue first, assigning an Outperform rating to the sector.

BOCI's key views are as follows. Humanoid robots are the ideal carrier for embodied intelligence, having crossed the mass production threshold in 2025. With the gradual maturation of humanoid robot hardware technology and the rapid development of AI large models, propelled by multiple factors including policy support and increased investment across various countries, the humanoid robot industry has reached a historic inflection point. Numerous manufacturers, both domestically and internationally, achieved scaled production in 2025. According to statistics from Omdia and BOCI, global humanoid robot shipments in 2025 reached approximately 14,800 units, representing substantial year-on-year growth from 2024. Concurrently, the scale of domestic robot orders achieved a significant breakthrough in 2025. Based on the plans of various domestic and international robot manufacturers, the next two years will be a critical period for the industrialization of humanoid robots.

Hardware motion capabilities have improved rapidly, while large model generalization capability remains the primary commercialization bottleneck. In the past two years, humanoid robot manufacturers have made significant progress in hardware motion capabilities. Based on the current hardware level demonstrated by humanoid robots, BOCI believes it is sufficient to meet basic deployment needs. The primary bottleneck hindering the large-scale commercialization of humanoid robots is the relatively low level of intelligence centered on the "brain." On one hand, currently mature VLA (Vision-Language-Action) large models have certain limitations in generalization capability, while world models with stronger generalization ability are still in an early stage. On the other hand, there is a severe shortage of high-density, high-quality data required for the deployment of embodied intelligence models. This results in current downstream application scenarios for humanoid robots being confined to structured environments like industrial and vertical applications, with a significant distance remaining before they can be applied in unstructured scenarios such as home services.

Rapid progress in algorithms, computing power, and data is helping to overcome the model generalization capability bottleneck. In terms of models, world models are developing rapidly. Regarding computing power, with algorithmic optimization of large models, improvements in inference engineering, and the expansion of computing infrastructure, the supply of computing power is growing quickly. Inference costs have fallen sharply in the past two years, and the cost of training and inference for the humanoid robot "brain" is expected to decrease in tandem. On the data front, governments and manufacturers are accelerating their布局 (deployment) to fill the data gap. The rapid development of algorithms, computing power, and data, the three pillars of physical AI, is expected to help humanoid robots overcome the bottleneck of insufficient model generalization capability.

Policy, funding, and technology are resonating, giving the domestic supply chain significant advantages poised to benefit fully. Leveraging unique domestic policy and funding support systems, a strong manufacturing base and supply chain depth, and cost advantages from scale, the industrialization of humanoid robots and component cost reduction in China are already globally leading. In 2025, the top six global humanoid robot shippers by volume were all Chinese companies, collectively accounting for 87.87% of the global shipment share. As humanoid robots move towards large-scale commercial applications, the domestic supply chain, particularly core component segments, is expected to benefit fully and will be the first to realize related orders and revenue.

Motion solutions are gradually converging, while segments like dexterous hands, electronic skin, and thermal management continue to iterate. Breaking down the components of a typical humanoid robot, its architecture includes a perception system, a decision-making and control system, an actuation system, and other components. Among these, the actuation system, comprising rotary actuators, linear actuators, and dexterous hands, accounts for the majority of the cost. Currently, electric actuation systems, with motors, reducers, and ball screws as core components, have become the choice for most manufacturers. The technical roadmap for humanoid robot motion solutions is converging and offers high certainty. In contrast, segments like dexterous hands, electronic skin, and thermal management, constrained by current technological levels and costs, are still in the exploratory phase of their technical roadmap. While there is some uncertainty, their development potential remains considerable.

Regarding valuation, the humanoid robot industry is currently in a critical transition phase from technical validation to scaled production. The report recommends focusing on core component segments expected to be the first to realize orders and revenue. Stocks to watch include Hengli Hydraulic, Leaderdrive (Harmonic Drive), Zhejiang Rongtai, Beijing Beite Technology, Zhenyu Technology, Tuopu Group, Sanhua Intelligent Controls, Wuzhou Xinchun, Moons' Electric, Buke Technology, Riying Electronics, Hanwei Technology, and Keli Sensing.

Key risks to the rating include the risk of humanoid robot commercialization progress falling short of expectations, the risk of technological iteration, and the risk of declining profitability.