Understand the industrial chain in "one page": Humanoid robots, Figure chain and Tesla Optimus chain

(1) Humanoid Robots: A Hotbed for the Next Hundredfold Opportunity

The humanoid robot industry is at a critical turning point from R & D demonstration to mass production and implementation.

Based on the mass - production plans of current leading enterprises such as Tesla (the Optimus project plans for mass production in 2026), Figure AI, and domestic company Ubtech, the years 2025 - 2026 are regarded as the key window period for industrial scale verification and accelerated commercialization.

Currently, humanoid robots are showing a trend of "cost reduction in hardware" and "intelligence enhancement in software" advancing in tandem, initially driven by B - end industrial scenarios, especially in the automotive manufacturing field.

Ubtech's Walker S1, XPeng's Iron, etc. have entered the factories of car - making enterprises such as BYD, Geely Zeekr, and XPeng for practical training to verify their application value on the production line.

The global competition pattern has initially taken shape, with different paths for domestic and overseas enterprises:

1) Overseas giants: Represented by Tesla and Figure AI, they occupy the technological high - ground with their leading advantages in AI algorithms, system integration, and end - to - end models.

2) Domestic enterprises: Relying on China's powerful automotive and 3C industrial chains, they have formed significant advantages in supply - chain collaboration, rapid iteration of complete machines, and cost control. The prices of some domestic products have dropped to the level of 100,000 yuan, creating an obvious price difference with overseas products, which lays the foundation for large - scale market penetration.

(2) Analysis of the Humanoid Robot Industrial Chain Structure

The humanoid robot industrial chain can be divided into three major links: upstream core components, mid - stream body manufacturing, and downstream scenario applications.

Among them, mid - stream body enterprises are analogous to the "OEMs" in the automotive industry and occupy a core leading position in the industrial chain. They are responsible for technology integration, product definition, and large - scale production.

*Note: Tables are made by FinGPT Agent, the same below.

a. Upstream core components: The value is highly concentrated in three major components.

Joints are the foundation for achieving movement and are divided into two types: rotary and linear.

Reducers are the key to ensuring movement accuracy. The mainstream solutions include harmonic reducers (used by Tesla's Optimus) and planetary reducers. Representative enterprises include Japan's Harmonic Drive Systems and domestic Laifu Harmonic, etc.

Sensors endow robots with perception ability, including force/torque, tactile, and visual sensors.

b. Mid - stream body manufacturing: Domestic and overseas enterprises are accelerating their layout, forming a technological competition.

Overseas, Tesla (Optimus) and Figure AI (Figure 01) are the representatives, leading the technological frontier.

Domestic startups such as Ubtech (Walker S1), ZhiYuan Robotics (Yuanzheng A1) have emerged. At the same time, car - making enterprises and technology giants such as XPeng (PX5) and Xiaomi (CyberOne) have also entered the market.

c. Downstream application scenarios: Current commercialization focuses on the B - end.

Industrial manufacturing is the primary implementation scenario, especially in automotive factories. For example, Ubtech's Walker S1 has entered the production lines of BYD and Dongfeng Liuzhou Motor for practical training.

Logistics and warehousing handling and sorting are another major potential market. In the future, it will gradually penetrate into commercial services and household scenarios.

(3) Main Challenges Faced by Humanoid Robots

Humanoid robots generally adopt a full - stack technology architecture of "brain - cerebellum - limbs". This architecture, through modular division of labor, efficiently and collaboratively supports the three core capabilities of robots: perception, decision - making, and execution. It is the key framework for achieving embodied intelligence.

Among them, the "brain" is responsible for high - level task planning and decision - making intelligence; the "cerebellum" focuses on real - time motion control and balance coordination; the "limbs", as the execution terminal, are responsible for direct interaction with the physical world.

This hierarchical and decoupled design aims to balance the generalization ability of complex tasks and the high - frequency, real - time physical control requirements. It is the current mainstream technology implementation path.

The industrialization process of humanoid robots depends on the collaborative breakthroughs in two major aspects: hardware and software.

At the hardware level, the focus is on three core challenges: "cost reduction, mass production, and battery life", while at the software level, efforts are made to solve three technical bottlenecks: "intelligent generalization, data scarcity, and real - time performance".

First, let's look at the challenges at the "hardware" level.

1) High cost and lack of standardization: Currently, the hardware solutions for humanoid robots have not converged, and there is a lack of unified standards, resulting in high BOM costs. Among them, joint modules and dexterous hands are the core for cost reduction.

2) Insufficient mass - production capacity: Leading humanoid robot companies can only achieve small - batch deliveries of hundreds to thousands of units in 2025, mostly used for non - commercial purposes such as data collection.

However, with Fourier Intelligence's harmonic reducers entering mass - production testing and Unitree Technology's self - developed M107 joint motor, the mass - production bottleneck is expected to be broken.

3) Battery - life limitation: The battery life of mainstream products is limited, falling short of all - day operation requirements, which restricts commercial implementation.

Battery life depends on the development of battery technology. Currently, GAC's GoMate uses all - solid - state batteries to achieve a 6 - hour battery life; Pudu Technology's PUDU D7 has a battery capacity of over 1 kWh, supporting over 8 hours of operation.

Then, let's look at the challenges at the "software" level.

1) Insufficient intelligent generalization ability: The "ChatGPT moment" has not yet occurred in the field of humanoid robots. The shortage of model representation ability and high - quality data is the core bottleneck for the emergence of intelligence, but the industry is evolving rapidly.

ZhiYuan Robotics released the general embodied base large model GO - 1, using the ViLLA architecture, with the average task success rate increased by 32%.

Figure's Helix VLA uses a "slow system + fast system" to balance generalization and real - time control, becoming the mainstream for engineering implementation.

2) Scarcity of high - quality data: The data modalities of movement and operation are complex. The high cost of collection in the real environment, difficulty in generalization, and lack of standards restrict the training effect of models.

In terms of data, NVIDIA released the Cosmos platform, which provides video world models to generate physical synthetic data to solve the problem of data shortage.

Galaxy Universal launched the end - to - end grasping large model GraspVLA, which is pre - trained based on synthetic big data.

3) Real - time performance and computing power constraints: The action frequency of existing models (such as π0's 50Hz) does not meet the requirements of complex scenarios (the target is 100Hz), and end - to - end models have extremely high requirements for computing power.

The action frequency of the model is currently an important bottleneck for humanoid robots. Think of a robot as a person, and the action frequency is how many times you can "refresh" your actions per second.

50Hz means 50 frames per second, which seems quite fast. But in a scenario where you fall and need to support yourself with your hands instantly, one crucial frame might be missed in 50 frames, and the hand could break.

From RT - 1 to Helix, the model control frequency has increased from less than 10Hz to 200Hz, gradually meeting the real - time requirements. Only when the robot's "cerebellum" has a high enough "refresh rate" can it handle more complex emergencies.

*Note: Tables are made by FinGPT Agent, the same below.

Currently, the industry is accelerating the solution of the above problems through supply - chain collaboration, autonomy of core components, innovation in large models, and the construction of an open - source ecosystem to promote commercial implementation.

(4) Market Size and Application Scenarios of Humanoid Robots

The humanoid robot industry is on the verge of a commercial explosion. The year 2025 is generally regarded as the "year of mass production", marking a critical turning point for the industry from prototype demonstration to large - scale implementation.

In the long run, humanoid robots are expected to be popularized in the C - end market, with global shipments exceeding 70 million units and the market size exceeding 10 trillion yuan.

The industrialization of humanoid robots follows a scenario - progression path of "industrial manufacturing → commercial services → household services".

Currently, industrial manufacturing is the core breakthrough point, focusing on flexible production links such as automotive production lines; commercial services are expanding rapidly, verifying their value in scenarios such as unmanned retail and office services; household services, as a long - term goal, are still in the early exploration stage of commercialization due to high technical complexity and costs.

(5) Major Global Participants in the Humanoid Robot Industry

Currently, the humanoid robot industry presents a diversified competition pattern. The major participants can be divided into four categories: overseas giants leading the technological frontier, domestic first - tier startups supported by capital, cross - border car - making enterprises with in - depth layout, and technology giants providing intelligent bases.

All parties, with their different endowment advantages, engage in fierce competition in terms of technology routes, commercialization paths, and ecosystem construction, jointly promoting the industry from the laboratory to large - scale application.

Overseas giants lead: Overseas giants occupy a leading position in core algorithms and system integration, leading the technological development direction of the industry.

a. Tesla (Tesla): Relying on its technological accumulation in AI and autonomous driving, the Optimus series is a typical representative of the end - to - end model route.

b. Figure AI: Its technical path uses a hierarchical decision - making model. The top layer uses GPT - 4V for visual reasoning, and the bottom layer realizes high - frequency control, achieving excellent engineering implementation results.

Domestic startups are divided into different tiers: Domestic startups have rapidly emerged with the help of capital and industrial resources, forming a clear tier - differentiation pattern.

a. The first tier: Represented by Ubtech, ZhiYuan Robotics, and Unitree Technology, they all have valuations exceeding 10 billion yuan, with strong financing capabilities and the ability to integrate industrial resources.

Ubtech's Walker S1 has entered the practical training in car - making enterprises such as BYD and Dongfeng Liuzhou Motor; Unitree Technology completed a 700 - million - yuan Series C+ financing, and its Unitree G1 actively explores the consumer market with a pricing strategy of 99,000 yuan.

b. The second and third tiers: Enterprises such as Leju, Pudu Technology, Fourier, and Galaxy Universal either rely on local governments and industry leaders (such as Huawei and Meituan) for resources or focus on niche areas such as open - source platforms and core components, forming differentiated competitive advantages.

Cross - border car - making enterprises enter the market: The core logic for car - making enterprises to enter the market is to use their mature supply - chain systems, large - scale production capabilities, and lean management experience to achieve rapid cost reduction and scenario implementation.

a. XPeng Motors (XPeng): The PX5/Iron robots released by XPeng's AeroHT have participated in the general assembly line practical training of the P7+ model in the Guangzhou factory and plan to achieve L3 - level mass production in 2026, with a cost target controlled within 120,000 yuan, following a clear path.

b. GAC Group (GAC): Its GoMate robot emphasizes the complete self - research of core components, relying on the automotive supply chain to achieve hardware reuse and cost control, and plans to promote large - scale implementation of the complete machine in 2026.

c. Other car - making enterprises: BYD, Chery, etc. also participate deeply in the industrial chain through direct investment (such as BYD's investment in ZhiYuan Robotics) or by establishing joint ventures.

Technology giants provide enablement: Technology giants mainly play the roles of "enablers" and "investors", accelerating the industrial intelligent process by outputting AI large - model capabilities and injecting capital.

a. Model and platform enablement: Huawei's Pangu large model empowers Leju robots, Baidu's Wenxin large model cooperates with Ubtech to optimize task - planning efficiency, and Xiaomi's self - developed "Xiaomi Brain" system provides the "brain" for humanoid robots.

b. Capital and ecosystem layout: Tencent, Alibaba, JD.com, LG Group, etc. deeply bind leading startups through strategic investment. For example, Tencent invests in ZhiYuan Robotics and Unitree Technology, and Alibaba invests in Unitree Technology and Zhuji Dynamics to