Author | Chen Wen

Source | Dongjian New Research Institute

Amidst the 2025 Spring Festival Gala spotlight, Hangzhou Unitree Technology's humanoid robot gracefully executed intricate yangko dance steps, propelling the embodied AI sector's valuation to the trillion-dollar mark in global capital markets.

Dubbed by MIT Technology Review as the "ultimate form of the Fourth Industrial Revolution," this domain's competition is witnessing the fiercest battle for market dominance in technological history.

On one front, humanoid robots are invading factories, transforming into "assembly line workers" on automotive lines. Concurrently, bionic robotic arms suture mouse cerebral vessels with microscopic precision, undetectable by the naked eye. This seemingly dichotomous landscape vividly portrays the industry's bipolar ecosystem.

01 Ecological Dominance and Niche Survival

The proliferation of embodied AI is evident in the diverse range of players involved.

In addition to established robot companies like UBTECH, Boston Dynamics, and Fourier Intelligence, specializing in robot bodies and hardware, and startups like Unitree Technology, Magic Atom, and Zhui Ji Power focused on the robot industry, two other groups are prominent: tech giants such as Tesla, Huawei, Baidu, and Xiaomi, along with numerous automakers branching out from other industries.

Large corporations and giants excel in perception and cognition algorithms, possess well-defined application scenarios, and share extensive industrial chains, enabling them to establish robust ecological dominance.

For instance, Tesla ventured into robotics with a "car-making logic." Its Optimus humanoid robot leverages autonomous driving technology, collaborating with the FSD system and Dojo supercomputer to initially handle tasks like transportation and inspection in factory settings, with plans to extend to external customers by 2026.

Tesla's core strengths lie in technology reuse (e.g., visual algorithms, battery management) and mass manufacturing experience, aiming to reduce robot costs below $20,000, leveraging "disruptive pricing" to penetrate the consumer market.

Xiaomi has adopted an "ecological penetration" approach. From the quadruped robot dog CyberDog to the humanoid robot CyberOne, Xiaomi integrates embodied AI into smart homes, wearable devices, and automotive scenarios through the synergy of its AIoT ecosystem.

This strategy mirrors Tesla's, leveraging mature supply chains and cost control to propel robots from the lab to the mass market.

Huawei's layout showcases more Eastern wisdom, with the Huawei (Shenzhen) Global Embodied Intelligence Industry Innovation Center connecting 16 industrial chain enterprises, exercising discourse power from chips to terminal devices across the entire chain.

Giants' advantages are apparent, while startups predominantly adopt a "deep technical expertise" strategy to penetrate the market.

For example, Unitree Technology's quadruped robot B2-W achieves dynamic balance under a 40kg load with its proprietary high-torque motor and reinforcement learning algorithm, commanding over 69% of the global market share.

Its unique selling proposition lies in full-stack self-developed hardware (motors, reducers, controllers) and an "agile and flexible" iteration strategy—from the first humanoid robot H1 to the mass-produced G1, it slashed the entry-level price of humanoid robots to RMB 99,000 in just two years, setting an industry price benchmark. This "geek gene" empowers it to establish barriers in vertical scenarios like industrial inspection and emergency rescue.

Similarly, startup Galaxy General has bet on the technological singularity of "embodied large models." Its Galbot G1 robot boasts a success rate exceeding 95% in grasping randomly stacked, unseen transparent, highly reflective objects on an international scale, further mastering generalized manipulation skills such as opening cabinets, drawers, and hanging laundry.

Currently, Galbot G1 independently executes complex tasks like dispensing and inventory management in pharmacies; in factories, it performs generalized tasks like pallet breaking and bin transfer, adeptly handling unexpected anomalies and visually guided stacked, arbitrarily placed bins (no on-site QR code positioning required).

The survival logic of Galaxy General and similar entities hinges on technological generational differences, using algorithmic breakthroughs to offset hardware scale disadvantages, seeking blue oceans in long-tail demands such as healthcare, manufacturing, and agriculture.

From the above analysis, embodied AI sector competition resembles a game of "elephants" versus "cheetahs"—the former dominates with resource depth, while the latter excels with agility.

02 The Innovation Equation for Startups

Industry competition is fierce, yet embodied AI still confronts triple challenges at the R&D and application levels: technology, data, and commercialization.

Firstly, technical route convergence is crucial.

Wang Tianmiao, Honorary Director of Beihang University's Robotics Institute and Dean of the Zhongguancun Zhiyou Research Institute, shared his views at the "2024 World Robot Conference," stating, "Currently, humanoid robots face two major hurdles: the robot's current 'soft' end, with general large models and vertical professional models suitable for robots still in nascent stages of development. Additionally, dexterous hands present technical and cost challenges that need addressing."

Secondly, the lack of data has trapped many companies in a "chicken and egg" dilemma.

Embodied AI relies on the accumulation of physical interaction data, but collecting such data is costly (daily data collection for a single robot exceeds RMB 10,000). Traditional AI's static datasets are ineffective in this domain, leading to inefficient algorithm training. Tesla's Optimus, for example, requires millions of hours of factory test data to achieve stable sorting, while startups often struggle to access scenario entry points.

Furthermore, most industry players have yet to traverse the "valley of death" between cost and mass production.

Currently, embodied intelligent robots' core component costs account for over 60% of the total, with imported harmonic reducers priced at RMB 10,000 each, resulting in high overall costs (Boston Dynamics' Spot retails for RMB 530,000). Concurrently, inadequate precision manufacturing processes and supply chain management capabilities trap most enterprises in a dilemma of "impressive prototypes but no mass production in sight."

Global humanoid robot shipments in 2024 were less than 1,000 units, affirming the challenges of commercialization and scaling.

Startups' breakthrough lies in differentiated routes addressing industry pain points.

Taking Magic Atom as an example, technologically, it broke Boston Dynamics' hydraulic drive monopoly; its humanoid robot became the world's first electrically driven acrobatic robot thanks to its proprietary D190 joint module.

The self-developed humanoid robot MagicBot leverages the MagicNet multi-robot collaboration system to automate transportation, assembly, quality inspection, and warehousing at the Dreame factory.

To enhance adaptability and reduce overall humanoid robot costs, it also unveiled the proprietary dexterous hand product MagicHand S01 in February this year, featuring 11 degrees of freedom, a hand load capacity of up to 5kg, and an operational scenario maximum load of over 20kg.

In terms of data, Magic Atom employs its proprietary MagicData AI data engine to accurately annotate and efficiently process various data in the cloud and locally, then completes model training in a simulation environment, and finally integrates it seamlessly into real-world applications, using scenario data to inform technological iterations.

Regarding application deployment, Magic Atom delves into vertical scenarios like automotive manufacturing and 3C electronics, partnering with China Telecom to deploy the home service ecosystem, planning to deliver hundreds of mass-produced models in 2025.

It's evident that the core competitive advantage for startups like Magic Atom isn't direct competition with giants but rather breaking cost thresholds through hardware innovation, accumulating data assets through deep scenario diving, and building industrial barriers through ecological collaboration, thereby establishing irreplaceability in niche areas.

03 The Multi-Dimensional Game in Winning Factors

Looking back from the threshold of 2025, the embodied AI industry is transitioning from "technology competition" to "ecological warfare." The breakthrough experiences of Magic Atom and similar entities reveal that future competition will focus on three core dimensions: technical capabilities, scenario understanding, and ecological aggregation.

Technology serves as the foundation, universally acknowledged, while scenario and ecological competition more vividly reflects players' resource integration capabilities within the sector.

As a new company established in January 2024, Magic Atom's rapid progress is largely attributed to the resource empowerment of its investor, Dreame Tech.

Magic Atom's RMB 150 million angel round, completed last year, was led by Zhuichuang Venture Capital (a robotics industry-focused venture capital fund initiated by Dreame Tech) with follow-on investment from Yipu Fund.

Beyond funding, Magic Atom received supply chain and scenario landing support from Dreame Tech. Dreame Tech's global cleaning robot channels provided trial and error scenarios, while the RMB 150 million funding accelerated core component self-research and domestic substitution.

From industrial scenarios, where bipedal humanoid robots adapt to human working environments, reducing factory deployment costs through multi-robot collaboration, to home service robots exploring new scenarios like "humanoid shopping guides" and "elderly care" in partnership with China Telecom, Magic Atom has established a combination of "deep vertical scenario diving + horizontal ecological expansion."

Another often overlooked detail is that initially established in Beijing, Magic Atom soon relocated to Liangxi District, Wuxi, aiming to tap into its robust industrial ecology.

Robots have always been hailed as the "crown jewel of manufacturing," and UBTECH, the "first humanoid robot stock" listed on the Hong Kong Stock Exchange, hails from Wuxi.

To foster the robotics industry, Liangxi Science and Technology City, where Magic Atom is situated, boasts a relatively mature industrial ecosystem, encompassing industrial robots, service robots, rehabilitation robots, agricultural robots, surgical robots, and other industrial tracks.

Neighboring Liangxi District, Wuxi Economic Development Zone and Huishan District also prioritize the robotics industry.

The former defines humanoid robots as a new hallmark for future industries and has already attracted leading enterprises in robot R&D, including UBTECH, iFLYTEK, Annimai, JD Technology, and UDI Technology.

The latter has invested RMB 1.02 billion to establish a 163-mu Huishan Humanoid Robot Industrial Park. Though still in infrastructure construction, it has already signed and landed five projects related to the humanoid robot industry chain.

Undoubtedly, this resource network and industrial ecology have accelerated Magic Atom's technological iteration and market expansion.

The rise of embodied AI startups underscores the underlying logic of "new manufacturing": leveraging hard technology as a spear and scenarios as a shield, harnessing capital and policy tailwinds to excel in global competition.

In the future, those who transform "lab acrobatics" into "factory productivity" will occupy the commanding heights in this intelligent revolution. The key to this success may very well lie in the hands of teams that "understand both code and screws."