It's getting harder to tell the story of embodied intelligence.

Written by | Huahua

Unitree Technology is going to go public.

This news didn't cause a big stir. Instead, it's more like an industry topic.

In today's AI narrative, it's even a bit boring. After all, in the past period, the public's attention was on Agents and crayfish. In contrast, a robot company going public is neither new nor fast.

However, if you open its prospectus, you'll see a set of different data:

The proportion of humanoid robot revenue increased from 1.88% to 51.53% within two years.

The production and sales rate exceeded 95%.

30,000 quadruped robots were sold, and the shipment volume of humanoid robots exceeded 5,500 units.

The annual revenue is expected to exceed 1.7 billion yuan, and the net profit after deducting non-recurring gains and losses is over 600 million yuan.

Meanwhile, Unitree plans to raise 4.202 billion yuan, which will be invested in models, bodies, products, and manufacturing bases. It is expected to have an annual production capacity of 75,000 humanoid robots + 115,000 quadruped robots.

To be honest, these figures aren't explosive. But they point to something more specific: Robots are starting to be produced, sold, and delivered on a large scale like commodities.

This is the first time the embodied intelligence industry has presented verifiable commercial evidence.

1. The Moment a Robot Leaves the Factory

In the autumn of 2024, a video quietly spread in the industry circle.

There was no carefully designed demonstration, no lighting, and no background music in the video. A robot slowly moved in a factory workshop, bypassed the debris on the ground, stopped in front of the shelf, and completed a pick-up operation. The workers standing beside were looking at their phones, and no one applauded.

The protagonist of the video is the Unitree H1, the first-generation general-purpose humanoid robot launched by Unitree in 2023. It is about 1.8 meters tall and weighs 47 kilograms. It is driven by motors all over the body. Its design goal is to complete actual work in unstructured environments, rather than just running fixed programs for demonstrations. When the H1 was officially released, its price was about 90,000 yuan, making it one of the lowest-priced mass-produced humanoid robots in the world at that time.

This video was later forwarded by many people, not because of its amazing technology, but because it made people feel something different from before: This seems like work, not a performance.

Wang Xingxing once said in an interview, roughly: "We don't pursue making robots look cool. We pursue making people want to buy another one after using it."

Putting this statement together with Unitree's starting point, there is a bit of a sense of discord. When Unitree received a new round of financing in 2022, it was still in an unremarkable office building in Hangzhou. Wang Xingxing, who was 32 years old at that time, was already well-known in the industry, but far from as glorious as he is today.

However, it is this simple product logic that has allowed Unitree to take a different path from most robot companies.

2. A Demo is One Thing, and Shipment is Another

In the past five years, the embodied intelligence field has never lacked videos.

Robots do somersaults, serve coffee, play table tennis, and play the piano. Every few months, a new demonstration goes viral. Every time it goes viral, it is accompanied by a new round of financing and a new round of discussions: The iPhone moment of embodied intelligence is coming.

But the words "is coming" have been said for many years.

The three questions that have never really been answered are: Can it be mass-produced? Is there real demand? Can it be shipped continuously?

The figures in Unitree's prospectus answer these three questions head-on for the first time.

A production and sales rate of over 95% means that almost all the products made are sold. The buyers include factories, universities, research institutions, and overseas customers. It's not just one scenario; multiple scenarios are running simultaneously. This is not expected demand but real purchasing behavior.

Within two years, the proportion of humanoid robot revenue increased from less than 2% to over 50%. This speed shows that the market is not waiting for this product but quickly digesting it.

The annual production capacity is planned to exceed 190,000 units. By making an analogy: Cadillac delivered about 150,000 vehicles in China in 2024. Unitree's production capacity target is already approaching the shipment volume of a mature consumer product.

3. The Barrier is Not in the Model, but in the Ability to Mass-Produce

If you continue to read the prospectus, you'll see an interesting fund allocation: 48.13% of the raised funds will be invested in model research and development, and the rest will be invested in manufacturing bases and body research and development.

This ratio reveals the most core contradiction in embodied intelligence: It is both an AI problem and an engineering problem.

However, the intensity of competition for these two problems is diverging.

The model layer is accelerating towards publicization. Whether it's the open-source community or commercial APIs, the threshold for obtaining AI basic capabilities is rapidly decreasing. The visual models and language understanding capabilities you use in robots today may be accessible to all competitors at a lower cost next year.

But the mass-production ability will not be publicized. It depends on the supply chain relationships accumulated over the years, factories capable of stably producing complex hardware, and a large number of engineering pitfalls that have been encountered. These things can't be written in papers and can't be bought. You can only endure and overcome them by yourself.

Unitree started with quadruped robots, and this choice itself contains a logic. The market scale of quadruped robots is much smaller, but it is the category in which Unitree first ran through the productization path, and it is also a training ground for the team to truly understand what "mass production" means.

From quadruped to humanoid robots, what Wang Xingxing is doing is not a brand-new start but migrating the verified engineering system to more complex hardware.

Being able to make a demo is just an admission ticket. The ability to stably ship products is the moat.

4. The Bubble in the Embodied Intelligence Industry is Starting to Layer

When telling Unitree's story, we can't avoid a larger background. Where are most companies in this industry now?

From 2023 to 2025, more than 60 humanoid robot companies emerged in China, with a total financing of over 50 billion yuan. The logic of investors is that this is the next explosive track, and they need to secure a position in advance.

However, if you look closely at the current situation of these companies, you'll find an obvious stratification.

In the first layer, there are very few companies that have truly entered mass production. Unitree is the one closest to large-scale shipment at present, followed by several companies with backgrounds in large enterprises or locked-in major customers.

In the second layer, companies have completed product definition, can produce prototypes, and have small-scale deliveries. However, the shipment volume is still in the range of dozens to hundreds of units. The technical route is not bad, but there is still a long way to go before industrialization.

In the third layer, companies are still raising funds, making demos, and participating in exhibitions. The videos they shoot are very good, and they can tell a complete technical story, but there is no verifiable shipment data. Companies in the third layer are not without value. Some core technology accumulations may be absorbed in future integrations.

But for today's investors and observers, this is an important reminder: The bubble in the robot industry is not an overall bubble but a layered bubble.

Companies that can ship products are real, and the rest need to be treated more cautiously.

This is also another meaning of Unitree's IPO. It is making a mandatory classification for the entire industry.

5. A New Combination Appears

Looking at Unitree's shareholder structure, you'll find some interesting names: Wang Xing and his related system (Meituan-related entities hold a total of 9.65% of the shares), Shunwei Capital of Lei Jun (holding 4.42% of the shares), and top institutions such as Sequoia and Matrix Partners.

Explaining this combination by capital support is too superficial. What's more worthy of attention is what each of these forces can bring:

Lei Jun's system: Manufacturing experience, consumer electronics supply chain, and hardware scaling strategies

Wang Xing's system: Local life scenarios, delivery networks, and service implementation channels

Unitree itself: Robot bodies, motion control, and mass-production engineering capabilities

Putting these three together actually answers a more crucial question than whether the robot is useful: Where will the robot ultimately be used, who will operate it, and how can it cover enough scenarios?

The story of the AI application layer often involves a single-point breakthrough, where one product penetrates a certain type of user. But this is not the logic for robots. It is a systematic project. Hardware, software, scenarios, and service networks are all necessary. Without any one of them, it is difficult to achieve scale.

This is exactly why the competition in embodied intelligence is ultimately not between companies but between ecosystems. The one who can piece these parts together the fastest will have an advantage in the next round.

Unitree's IPO is the moment when the outline of the first large piece in this jigsaw puzzle appears.

6. Is a Robot Worth the Price?

The starting price of Unitree's quadruped robot Go2 is 9,997 yuan, which is the first time the industry product price has been reduced to less than 10,000 yuan. The starting price of the humanoid robot G1 is 99,000 yuan (now reduced to 85,000 yuan), the starting price of the R1 is 39,900 yuan, and the Air version is 29,900 yuan. Including deployment and service fees, the complete cost of a deployed robot is approximately in the range of 100,000 to 200,000 yuan.

This price has actually started to make some buyers seriously calculate the cost.

For example, if a security guard has a monthly salary of 5,000 yuan, the annual cost is 60,000 yuan, and it is close to 180,000 yuan in three years, not including management costs and holiday issues. A quadruped robot for night patrol has about the same cost over three years, but it runs for continuous machine hours.

Not every factory can figure out this account clearly.

Robot deployment also involves site renovation, system docking, and daily maintenance. Many people haven't thought clearly about these hidden costs before making a purchase. The tasks that a robot can do are often narrower than expected. It can complete repetitive tasks in fixed scenarios, but complex judgment and handling of exceptional situations are still beyond its scope.

But one thing is certain: The cost of robots will only become lower, and their capabilities will only become stronger. This is the basic law of industrial production.

GGII data shows that in Q1 2026, the unit cost of humanoid robots has dropped to 100,000 yuan, a 33% decrease compared to 2025, and the investment payback period has been shortened to 12 months.

What Unitree is doing now is to use large-scale production to lower the price curve and continuous iteration to raise the capability curve. When the intersection point of these two curves is low enough, there's no need to calculate the cost anymore. Buying a robot will be as natural as buying a computer.

This is the real logic behind Unitree's plan to produce 190,000 units annually: It's not that there is so much demand today, but to use production capacity to drive down the price and create demand for tomorrow.

7. This Industry No Longer Belongs to Storytellers

All industries have a story-telling period in their early stages.

The characteristics are new technology, good-looking videos, great imagination, and active capital, but there is no real delivery. In this stage, telling a good story is more important than doing a good job.

Unitree has officially put an end to the story-telling period of this industry with a prospectus.

When an industry starts to talk about production capacity and delivery cycles, its competition logic changes - from technological leadership to efficiency, cost, and the ability to deliver continuously. The entry threshold is raised, players are forced to be stratified, and capital's patience starts to tighten. They are increasingly looking at delivery data, not just at PPTs.

This industry has become more substantial.

Being substantial is not a bad thing. The automotive and consumer electronics industries have both gone through this moment. From new technology to industrial products. Companies that cross this threshold truly have long-term value.

8. From Understanding the World to Entering the World

If we draw a line for the evolution of AI in the past decade, we'll see a clear trajectory.

The starting point is understanding. Machines have learned to read languages, view images, and process knowledge. For the first time, they can understand what people are saying.

Then comes intervention. AI enters the workflow, generates content, and assists in decision-making. It goes from being able to understand to being able to help.

And now, the third thing is happening: AI is starting to have hands and feet and can enter factories and warehouses to complete the physical operations that used to be done by humans.

From understanding the world, to participating in the world, to entering the world. Each step is more difficult and more real than the previous one.

Unitree is at the point where the third step is just starting to be stable. This is not a story that will happen someday in the future but something that is happening right now, as evidenced by the prospectus, shipment data, and the factories that are starting to seriously calculate the cost and prepare to place orders.

In the past two years, AI has proven that machines can think. Unitree is proving that machines can work.

The distance between these two things is closer and faster than most people expected.

[Beyond the Page] Words:

The significance of Unitree's IPO is not how much money it has raised, but that it has given the industry a quantifiable reference for the first time.

In the past, when we said a robot company was good, we relied on videos, parameters, and financing amounts. Now we can look at more tangible things: production and sales rate, shipment volume, revenue structure, production capacity planning, and customer composition.

This is a turning point for an industry from storytelling to data-driven. From this moment on, companies that are still making demos and those that have already achieved mass production no longer belong to the same track.

Wang Xingxing said that Unitree's shipment target in 2026 is 10,000 to 20,000 units. If this target is achieved, it means that at least 10,000 humanoid robots will enter factories, warehouses, and real work scenarios this year.

By then, the term "embodied intelligence" will no longer be a concept but a procurement item on the bill.

This is the real reason why Unitree's IPO is worth remembering. It has made robots start to be talked about like commodities.

This article is from the WeChat official account "Beyond the Page", author: Huahua, published by 36Kr with authorization.