If you only watch short videos, you might easily perceive Unitree Robotics as a "star robot company": performing the Yangge dance at the Spring Festival Gala, fighting, running, and doing backflips. Every viral clip seems to prove that the era of humanoid robots has arrived.
However, reading Unitree's prospectus, the first-round inquiry, the second-round inquiry, and the audit implementation letter together leads to a more interesting conclusion:
Today, Unitree's true leadership lies not in the robot's "brain," but in its ability to mass-produce and sell high-performance robots at a sufficiently low cost and rapid pace.
This perfectly explains why Unitree is already profitable and why it still needs to invest heavily in building its "brain" post-IPO.
Unitree is applying for a listing on the STAR Market this time.
I. First, Look at a Set of Numbers That May Exceed Many Expectations
In 2025, Unitree achieved a revenue of approximately 1.699 billion RMB, with a gross profit margin for its main business reaching 60.13%, and a net profit excluding non-recurring items of about 591 million RMB. In 2023, its revenue was only 159 million RMB, with a net loss excluding non-recurring items of 18.02 million RMB. In just two years, the company has transitioned from hundred-million-level revenue to nearly 1.7 billion RMB in revenue, and from losses to high profitability.
The growth engine has also completely switched.
In 2023, humanoid robots contributed only 1.88% of Unitree's main business revenue; by 2025, this proportion had risen to 51.78%, surpassing quadruped robots for the first time. Throughout the year, 5,511 humanoid robots were shipped, with confirmed sales of 5,215 units, at an average selling price of approximately 166,400 RMB/unit.
Notably, the gross profit margin remains high: in 2025, the gross profit margin for Unitree's humanoid robots was still 63.18%, and for quadruped robots, it reached 56.72%.
This does not look like a robotics startup still stuck at the "burning cash to build demos" stage.
It has become a real business.
II. Unitree's True Moat Is "Making Robots Cheap"
There is a frequently overlooked fact in the robotics industry:
Performance leadership does not necessarily guarantee commercialization, but cost leadership can directly determine who achieves scale first.
Unitree's technological path has carried a strong engineering focus since day one. After its establishment in 2016, the company independently developed motor drives, mechanical structures, and whole-body control systems along the motor-drive route, and redeveloped its first commercial product, Laikago, based on Xingxing Wang's early XDog prototype. Around 2019, it further achieved a breakthrough in high-torque motors.
Xingxing Wang himself is not a typical "capital-driven founder." The prospectus discloses that he has over 15 years of robotics R&D experience, developed the XDog driven by low-cost outer rotor brushless motors during his master's studies, and has led more than 130 patents for the company.
This technological DNA has ultimately formed a highly typical vertical integration capability.
Unitree's in-house R&D covers not only the complete machine but also a vast array of core components, including motors, reducers, joint modules, control systems, LiDAR, and dexterous hands. The inquiry reply discloses that if high-spec optional LiDAR and dexterous hands are excluded, the procurement cost of outsourced components typically accounts for only about 14%-18% of the total product cost.
This is the key to understanding Unitree's capability in price wars.
From 2023 to 2025, the average selling price of quadruped robots dropped from 38,300 RMB to 30,300 RMB, while sales surged from 3,121 units to 23,037 units. The average selling price of humanoid robots fell even more dramatically, from 593,400 RMB to 260,400 RMB, and then to 166,400 RMB, with sales growing from 5 units and 412 units to 5,215 units.
Unitree does not achieve high gross margins by selling robots at high prices; rather, it lowers costs even faster while continuously reducing prices.
This is a highly formidable competitive capability.
Because when an industry enters the scaling stage, the truly terrifying enterprises are not those that make the product first, but those that can continuously drive the flywheel of "lowering costs—reducing selling prices—expanding sales volume—lowering costs again."
III. But There Is a Reality That Must Be Stated: Today's Humanoid Robots Have Not Yet Truly Entered Factories to "Work" on a Large Scale
This is perhaps the most critical information in the entire set of IPO materials.
According to the second-round inquiry reply, from January to September 2025, among Unitree's humanoid robot revenue, scientific research and education accounted for 73.60%, commercial consumption for 17.39%, and industrial applications for only 9.01%. Within the industrial application revenue, corporate guided tours accounted for another 50%-70%.
Furthermore, the company disclosed that the revenue from humanoid robots explicitly used for scenarios such as intelligent manufacturing, intelligent inspection, and logistics distribution totaled 15.702 million RMB, accounting for 29.29% of industrial application revenue.
Based on a simple calculation using this metric, this portion of revenue is equivalent to only about 2.6% of the humanoid robot revenue during the same period.
What does this mean?
It means that the two concepts most easily confused in the market today are actually completely different:
"Humanoid robots are beginning to be sold at scale" does not equal "humanoid robots have already replaced human labor at scale."
Unitree has already accomplished the first task and holds a significant lead; however, the entire industry still has a long way to go to achieve the second.
At this stage, those truly willing to pay are still primarily universities, research institutions, tech companies, developers, and clients for performances and commercial exhibitions.
This is not Unitree's problem, but rather the stage the industry is currently in.
Today's humanoid robots are more like high-performance "development platforms" rather than mature "labor products."
IV. Therefore, What Unitree's IPO Really Needs to Solve Is Transitioning from a "Cerebellum Champion" to a "Brain Player"
If over the past decade, Unitree's greatest strength has been its physical body and motion control, then the biggest variable for the next decade is the embodied AI "brain."
The prospectus is actually very restrained in its description of this.
Currently, Unitree's robots have integrated self-developed physical intelligence models, possessing clear advantages in motion control capabilities; some G1 models have also integrated relatively mature Large Language Models (LLMs). However, the company explicitly disclosed: during the reporting period, the self-developed general embodied large model has not yet been applied to robot products at scale.
More critically, the company's previous R&D investment was clearly skewed towards the physical body and the "cerebellum." In 2025, R&D expenses were 145 million RMB, accounting for only 8.53% of revenue; during the same period, the average R&D expense ratio for several comparable companies was 24.70%.
Unitree also acknowledged that it had not previously conducted large-scale real-world data collection and factory deployment training. It was only after releasing the WMA and VLA models in the second half of 2025 that it began to significantly ramp up its "brain" efforts.
Consequently, a very interesting figure emerged:
Unitree's total IPO proceeds planned for investment amount to approximately 4.202 billion RMB, of which the "Intelligent Robot Model R&D Project" alone is planned to receive about 2.022 billion RMB; another 1.11 billion RMB is allocated to physical body R&D.
What does 2 billion RMB mean?
It is roughly 14 times Unitree's total R&D expenses for the full year of 2025.
Of course, this is a multi-year investment and cannot be simply compared horizontally. But the direction of capital expenditure speaks for itself:
This IPO is not just about producing tens of thousands more of the existing robots; it is also about financing a technological paradigm shift.
Moving from leading in "body + cerebellum" to a complete closed loop of "brain + data + body."
V. Unitree's Real War Has Just Begun
Unitree has already solved the most difficult first question in robotics entrepreneurship:
How to transform a lab demo into a commodity that can be stably mass-produced, continuously reduced in price, and still remain profitable.
In this regard, it has clearly run ahead of many peers.
In 2025, Unitree's overall gross profit margin was 60.44%, while those disclosed for the same period by UBTECH, Dobot, DeepRobotics, and Leju Intelligent were 37.67%, 46.49%, 52.83%, and 40.78%, respectively. The prospectus attributes this difference to full-stack in-house R&D, cost control, and economies of scale.
But the second question is harder than the first:
When will robots no longer just be able to run, jump, and perform, but truly understand tasks, generalize to unfamiliar environments, and continuously create economic value?
This depends on models, data, computing power, scenario training, and the industrial ecosystem, which are precisely the areas where automakers, AI companies, internet giants, and even global tech companies are heavily investing.
Unitree itself has already felt this shift. In the first quarter of 2026, the company's operating revenue still grew by 68.49% year-on-year, but with the rapid increase in expenses such as R&D and sales, the net profit excluding non-recurring items dropped by 52.55% year-on-year.
This does not mean Unitree is losing speed.
On the contrary, it indicates that embodied AI is transitioning from an industry of "who can build the robot first" to an industry of "who can make the robot actually work."
Therefore, in judging Unitree's value in the future, the most critical metrics may no longer be how stunning the next backflip is, nor how many robots can be sold in a year.
Instead, they are three questions:
After entering real production scenarios, can robots continuously generate high-quality data?
Can this data train embodied models that truly possess generalization capabilities?
Can the model capabilities ultimately be transformed into productivity that customers are willing to pay for continuously?
If these three questions receive affirmative answers, then today's Unitree, with 1.7 billion RMB in revenue, may just be the beginning of the story.
If it can never cross this threshold, then no matter how strong the motion control or how beautiful the cost curve, the company may remain locked in as an excellent high-performance robot hardware enterprise.
Unitree has already proven that Chinese companies have the capability to turn robots from "expensive lab toys" into scaled commodities.
What will truly determine its ceiling next is whether it can complete another leap—