Text / Sansheng

Source / Jiedian Finance

From the unveiling of Boston Dynamics' Atlas in 2013 to Unitree Technology's robotic handkerchief twirlers captivating audiences during this year's Spring Festival Gala, from Tesla's Optimus clumsily sorting batteries in factories to UBTECH's Walker S1 efficiently working in automobile factories like BYD and Geely, the humanoid robot industry stands on the brink of commercialization after over a decade of evolution.

Recently, the discourse around quantifying humanoid robots has intensified, largely due to the buzz created by Musk.

Musk, CEO of Tesla, has set clear targets for Optimus' mass production: 10,000 units by 2025, with deliveries to non-Tesla companies commencing in the second half of 2026. He believes that "Optimus has the potential to generate revenue opportunities exceeding $10 trillion." During Tesla's fourth-quarter 2024 earnings call on January 29, Eastern Time, Musk's remarks on the commercialization potential of humanoid robots propelled the topic into the limelight.

2025, the year often depicted in science fiction films as the "Year of the Robot," may be arriving sooner than anticipated. The robot revolution could be nearer than we think. For China's humanoid robot industry, this presents both opportunities for leapfrogging development and concerns about potential bottlenecks. How will technological advancements, cost competition, and domestic supply chain breakthroughs unfold?

01 Technological Leap: From "Staggering Steps" to "Human-Like Evolution"

Humanoid robots, defined as intelligent machines resembling humans that can walk on two feet, embody their core value in mimicking human form for seamless interaction with tools and the environment.

Early robots were limited to single tasks in fixed settings, whereas the new generation of humanoid robots strives for "versatility," achieved through the integration of language interaction, motion control, and environmental perception systems.

Currently, humanoid robots are heralded as the "ultimate AI carriers," boasting broad adaptability across various scenarios, such as navigating complex terrains, performing intricate operations, and engaging in emotional interactions. Their value extends beyond replacing repetitive labor to reshaping productivity paradigms.

Given these advantages, humanoid robots have become a focal point in robotics research. Goldman Sachs predicts that by 2035, the global humanoid robot market will reach $38 billion, with shipments of 1.4 million units, covering 10%-15% of hazardous and automotive manufacturing positions.

Thus, humanoid robots have emerged as a new frontier in the competition for advanced technologies like AI, high-end manufacturing, and new materials. Since 2023, major global players like Tesla, Amazon, and Google have aggressively entered this market, evidenced by Musk's enthusiastic stance.

China's Ministry of Industry and Information Technology (MIIT) has positioned humanoid robots as a "disruptive technology," with continuous policy support. In this high-stakes global tech race, Chinese enterprises cannot afford to be spectators.

For instance, as of June 2024, UBTECH holds over 2,450 patents, including over 450 overseas patents, with invention patents accounting for nearly 60%, making it the global leader in valid humanoid robot patents. With over 50% of its R&D investment and the pioneering commercialization of the Walker series, UBTECH has established itself as a frontrunner in China and globally.

To achieve these milestones, the humanoid robot industry has navigated numerous developmental challenges, one being the "black hole" of data training. As an industry insider puts it, "Training a robot is harder than raising a child."

Humanoid robots require vast amounts of scenario data for training, but the acquisition cost is prohibitively high. The National Humanoid Robot Innovation Center estimated that training a robot to sort batteries necessitates tens of thousands of grasping actions, involving variables like light, angle, and material, with each data point costing over RMB 1,000.

This explains why current robots predominantly remain at the "language interaction" stage. While public internet corpora can support large-scale dialogues, motion data in the physical world must be built from scratch.

For the domestic humanoid robot industry, another hurdle is the hardware "bottleneck" in the domestic supply chain. A humanoid robot requires over 40 precision joints, and core components like servo motors, reducers, and torque sensors have traditionally relied on imports. Overcoming this localization challenge and reducing product costs is crucial for the industry's growth.

02 Triple Catalysts for Localization: Policy, Scenarios, and Open Source Ecosystems

Recently, China's first heterogeneous humanoid robot training ground officially opened at Zhangjiang Model Force Community in Pudong, capable of accommodating over 100 robots for simultaneous training in the first phase. Here, robots practice walking and Tai Chi, resembling a robot skills academy.

As mentioned earlier, humanoid robots require extensive training data. "Practice makes perfect." More data input and training sessions enhance the robot's intelligence. Such training grounds can significantly reduce the costs of developing embodied intelligence technology.

This is just one aspect of the domestic humanoid robot industry's development. The entire Chinese industrial chain is quietly fortifying its position. According to the MIIT's "Guidance on the Innovative Development of Humanoid Robots," key technologies for the "brain, cerebellum, and limbs" will be broken through by 2025, with international competitiveness anticipated by 2027. Policy-driven, the Yangtze River Delta and Pearl River Delta regions have emerged with over 100 humanoid robot-related enterprises, engaged in a race for localization breakthroughs.

In Jiedian Finance's view, the localization of humanoid robots manifests in three areas: core components, scenario revolution, and open source ecosystems. In core components, the domestic robot industry has adopted a "rural surrounding cities" approach.

For years, domestic humanoid robots have relied on imported core components like servo motors, reducers, and torque sensors. However, in recent years, domestic products have begun to make breakthroughs, with Chinese enterprises venturing into niche areas like lead screws, encoders, and coreless motors. For instance, Jiangsu Leili has launched a full series of joint motor solutions with precision close to Japanese products but at a 30% lower price; Changying Precision has collaborated with foreign enterprises to develop precision robot parts, achieving small-batch deliveries.

With any breakthrough, China's vast manufacturing and population scales provide a unique testing ground for the domestic robot industry, fostering a "scenario revolution." For example, UBTECH's Walker S series of humanoid robots have entered automobile factories to perform quality inspection tasks, achieving millimeter-level precision in inspecting car logos and headlights. These scenarios not only validate technology but also push enterprises to develop cost-effective solutions.

Furthermore, the domestic humanoid robot industry's open source ecosystem approach is gradually challenging foreign giants' monopoly. For example, the National Humanoid Robot Innovation Center's release of the world's first open-source public version machine, "Qinglong," signifies China's attempt to bypass Tesla's closed system.

As the humanoid robot industry poised for an explosion, Chinese companies will undoubtedly be at the forefront of the most competitive sector.

03 The Future: Will 2025 Mark a "Tipping Point"?

Currently, the humanoid robot industry's hottest discussion revolves around mass production.

On February 11, Unitree Technology launched two humanoid robot models, H1 and G1, on its official JD.com flagship store. The next day, the first batch of G1 humanoid robots, priced starting at RMB 99,000, sold out.

In factories, UBTECH's robots have also achieved remarkable results. In the BYD automobile factory, the first phase of practical training for UBTECH's Walker S1 humanoid robot has yielded initial results, doubling efficiency and increasing stability by 30%. Relevant optimization work is ongoing, with expectations of meeting large-scale delivery conditions in the second quarter of 2025.

Additionally, UBTECH has collaborated with multiple automobile enterprises like Dongfeng Liuzhou Automobile, Geely Automobile, FAW-Volkswagen Qingdao Branch, Audi FAW, BYD, and BAIC BJEV, as well as renowned enterprises like Foxconn and SF Express, on humanoid robot practical training.

Jiedian Finance dares to predict that with the continuous advancement of generative AI, actuators, machinery, battery storage, and other industries, humanoid robots will permeate various life and production scenarios in the coming years. Behind this lies not just competition among individual industries but a broader game within the AI era's vast ecosystem.

As Figure AI in the US secures $675 million in funding and Tesla nurtures robot training with autonomous driving data, Chinese enterprises must grapple with establishing a sustainable business model before the capital frenzy wanes and finding a balance between open source ecosystems and patent barriers.

Perhaps the answer lies in the factories of the Yangtze River Delta, the laboratories of Shenzhen, and the relentless efforts of countless engineers striving to reduce costs by even 1%.

*The title image is AI-generated