Unitree has made money, but humanoid robots haven't.

The humanoid robot track has been caught in a paradox in the past two years:

On one hand, there has been a crazy influx of capital, driving up valuations. On the other hand, the industry has always been haunted by the soul - searching question: "What else can it do besides dancing?"

On March 20, 2026, the application of Unitree Technology for IPO on the Science and Technology Innovation Board was accepted by the Shanghai Stock Exchange. This 363 - page prospectus put the answer on the table for the first time.

In 2025, its revenue was 1.708 billion yuan, net profit after deducting non - recurring items was 600 million yuan, the shipment volume of humanoid robots exceeded 5,500 units, and the gross profit margin exceeded 60%.

Selling robots is more profitable than selling cars. The gross profit margin of Unitree's humanoid robots was 62.91%, while Tesla's was about 18% and Xiaomi's was about 15% during the same period.

However, what really deserves to be questioned in the prospectus is not "why Unitree can make money", but "whose money is Unitree making?"

In the first three quarters of 2025, in the revenue of Unitree's humanoid robots, scientific research and education accounted for 73.6%, commercial consumption accounted for 17.39%, and real industrial applications only accounted for 9%.

In other words, Unitree's customers are mainly universities, scientific research institutions, and technology companies. They buy robots for research and development, rather than to replace workers and enter production lines.

This leads to a more fundamental question: If the "ultimate narrative" of humanoid robots is to become the next - generation productivity tool, then how far is a company that relies on scientific research and education for 70% of its revenue from that ultimate goal?

A "Misaligned" Profit

Unitree's profit is not a victory for the commercialization of humanoid robots, but a victory for the "dividends of the scientific research and education market".

This judgment needs to be analyzed in detail. In 2025, the unit price of Unitree's humanoid robots was 167,600 yuan, with a gross profit margin of 62.91%. This means that for each unit sold, the gross profit exceeded 100,000 yuan.

Behind the high price and high gross profit is a specific customer group: scientific research institutions and universities. These customers have several common characteristics: they have sufficient budgets and are not sensitive to prices; the procurement decision - making cycle is long.

Once on the supplier list, the repurchase is stable; their demand focuses on "being able to run and jump, and being open - source and open", rather than "being able to do specific work".

Unitree just meets these needs. Its robots have strong motion control capabilities, a perfect open - source ecosystem, and an early layout in the overseas market, and have accumulated a good reputation in the scientific research circle. In 2023 and 2024, the proportion of Unitree's overseas revenue continuously exceeded 55%, mainly due to the purchases of overseas scientific research institutions.

The characteristics of the market determine Unitree's profit model: it makes money not by "scale", but by "premium".

Because there are few competitors, customers have sufficient budgets, and the products have obvious differentiation, it can maintain a high gross profit.

There is an easily overlooked problem here: the ceiling of the scientific research and education market is much lower than that of the industrial market. How many universities around the world need humanoid robots? How many scientific research projects can afford the unit price of 167,600 yuan? How long can the market support Unitree's growth?

In 2025, the shipment volume of Unitree's humanoid robots exceeded 5,500 units. If we calculate based on each of the world's top 200 universities purchasing 50 units, the capacity of this market is only 10,000 units.

Unitree can fill this market in two years. The scientific research and education market can support a profitable company, but it cannot support the imagination of a "trillion - dollar track".

This is Unitree's "misalignment": it has proved that humanoid robots can make money with a beautiful profit report, but at the same time, it has also exposed that it is still far from the "ultimate narrative".

Two Five - Year Periods Behind Unitree's IPO

Unitree's development history can be understood through two "five - year" periods. What really matters is not the time nodes, but the two leaps in the business model.

The first five - year period (2016 - 2021): selling hardware.

Unitree started with quadruped robots, and its business model was very simple: build robotic dogs and sell them to overseas scientific research institutions and technology enthusiasts.

The core ability at this stage was "cost control". It self - developed all core components such as motors, reducers, and lidars, and pushed the price down to a fraction of its competitors'. By 2021, the cumulative sales volume of Unitree's quadruped robots exceeded 30,000 units, ranking first in the global market share.

The proposition at this stage was: survive and build a technical foundation. Unitree achieved this.

The second five - year period (2021 - 2026): selling "hardware + services".

In 2023, Unitree launched the humanoid robot G1. This time, Unitree sold not only hardware but also a set of "value - added services": open - source models, development platforms, and technical support.

What scientific research institutions buy is not a dancing robot, but an experimental platform that can run algorithms.

The proposition at this stage was: stand out and prove the business model. Unitree also achieved this. It became well - known through the Spring Festival Gala, supported its revenue with the scientific research and education market, and proved its profitability with a high gross profit margin.

In these two five - year periods, Unitree's business model has one thing in common: it has always been "selling tools", rather than "selling productivity".

Scientific research institutions buy Unitree's robots for their own research; enterprises buy Unitree's robots for image display in showrooms.

In these scenarios, Unitree's role is a tool provider, rather than a solution provider.

The proposition for the third five - year period (after 2026) becomes: transitioning from selling tools to selling productivity.

This leap is much more difficult than the previous two five - year periods. Because "selling productivity" means:

Unitree is no longer just providing a dancing hardware platform, but needs to solve real problems in real scenarios, such as replacing workers in factories for assembly and completing sorting and handling in warehouses.

This cannot be solved by simply making the robots cheaper. It requires in - depth understanding of the scenario, adaptation to the process flow, and accurate grasp of customer needs.

In other words, Unitree needs to transform from a hardware company into an industry solution company. This transformation is much more difficult than the transition from quadruped to humanoid robots.

Why Can't the Dancing "Academic Genius" Enter the Factory?

For humanoid robots to enter factories, they need to cross not only the technical gap, but also a misaligned value logic.

Let's first look at the technical level. The fact that Unitree's robots can dance proves that they have strong motion control capabilities, can maintain balance, and can complete complex movements.

Factories need robots that can work, not dance: they need to be able to screw, carry, and adapt to environmental changes.

These two sets of capabilities point to different technical routes.

"Being able to dance" relies on "cerebellum" capabilities, such as motion control algorithms, balance ability, and dynamic response. Unitree has polished this set of capabilities to a high level; "being able to work" relies on "brain" capabilities, such as environmental perception, task planning, and generalization ability. The entire industry is still in the exploration stage for the latter.

Unitree mentioned in its prospectus that in early 2026, the company's self - developed industrial - grade embodied large model had been deployed and tested in its own factory, with the task of "autonomously completing the assembly of joint motors".

This is an important attempt, but the "own factory" is a controlled environment, which is fundamentally different from real industrial scenarios.

Now let's look at the value level. Even if the technology is mature, why would factory owners use robots to replace workers? The core consideration is the "return on investment".

A humanoid robot costs 167,600 yuan. With deployment, maintenance, and algorithm adaptation, the comprehensive cost may exceed 300,000 yuan. What is the annual salary of a worker? If the robot cannot recoup its cost within three years, factory owners will not buy it.

There is an easily overlooked detail in Unitree's prospectus: from January to September 2025, the total sales revenue of humanoid robots used in clear scenarios such as intelligent manufacturing, intelligent inspection, and logistics distribution was only 15.702 million yuan, accounting for less than 30% of the industrial application revenue.

The rest is mainly used for corporate tours and reception. That is to say, currently, enterprise customers who are willing to pay for humanoid robots mainly use them for "image display", rather than as "productivity tools".

This is the deep - seated dilemma that Unitree faces: its customers are willing to pay for "dancing", but no one is willing to pay for "working".

Conclusion

2026 is the tenth year since Unitree was founded.

Ten years, two five - year periods. In the first five - year period, it survived in the overseas market with quadruped robots; in the second five - year period, it stood out with humanoid robots and the Spring Festival Gala effect.

Now, Unitree stands at the starting point of the third five - year period. This time, the proposition is no longer "survive" or "stand out", but "keep going". Keeping going is more difficult than the previous two five - year periods combined.

In the "Statement to Investors" in the prospectus, Wang Xingxing wrote a sentence: "Let's realize the ultimate human dream of AGI together!"

Ten years ago, when Wang Xingxing was tinkering with XDog in the laboratory, he might not have expected to stand at the door of an IPO. But the "AGI Declaration" written on the front page of the prospectus is actually the same as the light in that young man's eyes ten years ago. They are both ultimate questions about "what machines can do".

Now, this question has a more specific direction: not "can robots dance", but "can robots work".

When the lights of the Spring Festival Gala go out and the tide of capital recedes, what determines how far Unitree can go is no longer the ability to "dance", but whether it can answer the more difficult question: What else can it do besides dancing?

This article is from the WeChat official account "Zhi Xie Island", author: Meng Zhaoye, published by 36Kr with authorization.