Unitree's IPO is missing a VisiCalc

I browsed through the draft of Unitree Technology's prospectus for the upcoming listing review.

The listing review will be held on June 1st. The company plans to raise 4.2 billion yuan and has a valuation of 42 billion yuan. It is aiming for the title of "the first A-share humanoid robot company".

In 2025, the annual revenue was 1.7 billion yuan. After deducting various expenses, the net profit was nearly 600 million yuan, a year-on-year increase of 674%. The gross profit margin of the main business was 60%. The company sold 5,500 humanoid robots, ranking first in the world.

When presented with these figures, my first reaction was: this company is really making money. Then I wondered, whose money is it making? Who is paying the bill?

01

With these questions in mind, I looked at the customer structure in the prospectus again, and my understanding of Unitree changed immediately.

According to the data disclosed by Unitree in the Shanghai Stock Exchange's inquiry letter, in the first three quarters of 2025, the humanoid robot business was segmented by application scenarios. The revenue from scientific research and education was 438 million yuan, accounting for 73.6%.

Commercial consumption accounted for 17.4%, mainly including store displays and cultural and tourism performances. The revenue from real industrial applications, such as intelligent manufacturing and intelligent inspection, was 15.7 million yuan, accounting for 2.64%.

To put it simply, for every 100 yuan of humanoid robots sold, nearly 74 yuan comes from universities and research institutions.

This proportion cannot be ignored. It shows one thing: Unitree's current humanoid robot business is essentially closer to an "educational tool business" rather than a "productivity business".

The payment logic for these two types of businesses is completely different.

Universities purchase robots with research funds. The procurement decision depends on the needs of research projects, academic frontiers, and domestic substitution policies. No one will calculate "how much labor cost this machine can save me in a year".

A recent single-source procurement announcement from Westlake University stated very clearly:

Mainstream foreign products are too expensive and are subject to technology control against China. Other domestic products cannot meet the requirements in terms of perception and movement capabilities. Unitree is the only choice at this stage.

The four words "the only choice" are the answer to the 60% gross profit margin.

In a market where competitors are not yet mature, customers are not sensitive to prices, and demand is driven by academic popularity and policies, the pricing power naturally lies in the hands of the seller. This is completely different from the industrial scenario.

Industrial customers calculate the input-output ratio when purchasing robots: how much work this machine can do in a year, how many people it can replace, and how long it will take to recover the cost. In this logic, a 60% gross profit margin is out of the question.

Unitree itself is also aware of this.

The exact words in the prospectus are quite long. Translated into plain language, it is: We know that we are currently mainly making money by selling to schools, and we also know that this is not the end. In the future, we need the industrial and consumer markets to take over.

It is honest and straightforward.

There is an interesting comparison: Ubtech. Also in the humanoid robot business, it chose a different path, directly targeting the automotive factory market and providing production line collaborative training for BYD, Geely, and Mercedes-Benz.

In 2024, it invested 478 million yuan in R & D, seven times that of Unitree in the same period. The corresponding result was that it still had an overall loss of about 790 million yuan in 2025, with a gross profit margin of less than 40%.

There is no right or wrong in these two paths. Unitree chose to first establish a foothold in the market with the highest certainty and generate profits, while Ubtech chose to first enter the scenario with the highest threshold and build a moat.

This is two completely different business judgments, betting on "when the industry will truly take off". The first step in understanding Unitree is to see clearly its revenue structure. It is a company that is doing very well in the "educational tool dividend period".

The 60% gross profit margin and nearly 600 million yuan in net profit are all products of this stage. However, the problem is that the educational tool dividend period has a characteristic: it has a time window.

02

I did a calculation. Based on the non - recurring net profit of 591 million yuan in 2025, the price - earnings ratio corresponding to Unitree's 42 billion yuan valuation is about 70 times. The average price - earnings ratio of similar companies on the Science and Technology Innovation Board is about 45 times. Unitree is much more expensive than the industry average.

What are we paying for with the extra cost? It certainly isn't the current educational tool business.

There is an expectation that Unitree will transform from a company selling robots to universities into a platform - level player in the era of embodied intelligence.

The price of 42 billion yuan tells a story of the scarcity as "the first A - share humanoid robot company" and a pre - occupation of a multi - trillion - yuan track. It implies that the company can sell robots into factories, families, and all places that need labor in the future.

Currently, more than 70% of Unitree's revenue comes from universities, and the industrial scenario can almost be ignored. The founder himself has said that "it is still unrealistic to have robots do housework in the next few years".

There is a gap between the future depicted by the 42 billion yuan valuation and Unitree's current position.

The core reason why the market is willing to pay for this gap is that last year's growth curve was extremely impressive, with revenue increasing by 335% and profit increasing by 674%. In the face of such growth speed, all "not yet" can be regarded as "soon to be".

However, the latest data shows a different picture.

The draft for the listing review disclosed the data for the first quarter of 2026. The revenue was 423 million yuan, and the growth rate dropped directly from 332% last year to 68%. The non - recurring net profit dropped from 84.84 million yuan in the same period last year to 40.25 million yuan, more than halved.

In just one quarter, the growth rate dropped from three - digit to two - digit, and the profit was directly halved.

The company explained that the main reason is the sharp increase in R & D and sales expenses. The R & D expenses increased by more than 38 million yuan year - on - year, and the sales expenses included a brand promotion fee for the 2026 Spring Festival Gala. These expenses are necessary and there is no hidden agenda.

I believe that what is really worth paying attention to is the company's own forward - looking statement.

Translating the exact words in the prospectus:

As the revenue base gets larger and the industry's popularity gradually cools down, it will be difficult to maintain the previous high growth rate, and the performance may even decline significantly or worse.

For a company sprinting for an IPO to write "the performance may decline significantly" in its own prospectus, the weight of this signal is heavier than any external analysis.

The company has provided guidance for the first half of the year: the revenue is expected to be between 1.05 billion and 1.13 billion yuan, with a year - on - year growth rate of about 35% to 45%. The non - recurring net profit is expected to be between 236 million and 283 million yuan, a decrease of 6% to 22% compared with the same period last year.

In other words, the profit decline is not a short - term phenomenon in the first quarter and will at least continue until the semi - annual report.

The good news is that the situation will improve in the second quarter compared with the first quarter, and the decline is narrowing. The bad news is that the growth curve that made the market willing to give a 70 - times PE is starting to bend.

This is what the "window" means. The educational tool dividend period is called a window because it will not automatically upgrade to the "productivity dividend period".

University research funds cannot be spent at will; they have a budget cycle. The procurement demand for domestic substitution will eventually be exhausted. As we all know, academic fads come and go.

When these driving forces start to level off, a new growth engine needs to be quickly connected. If it is connected, 42 billion yuan will be the starting price; if not, the 70 - times PE will face the pressure of reality.

I'm not trying to talk down Unitree. I'm talking about a structural problem in an "intermediate state". The past growth logic is slowing down, and the future growth logic has not been realized. Unitree is now at the intersection of two curves.

Given this, has anyone successfully made the leap from "educational tool" to "tool" in history? I checked, and someone has. Moreover, almost everyone is using the descendant products of this case today.

03

In 1977, a small company called Apple Computer developed the Apple II.

This machine was later written into all technology history textbooks and was one of the starting products in the era of personal computers. However, few people noticed that the most core customers of the early Apple II were not enterprises or families, but schools.

I delved into this history.

In 1978, Apple won a contract with the Minnesota Educational Computing Consortium and supplied 500 computers to schools across the state.

This consortium had an educational software library, including the later - famous "Oregon Trail". These software programs could only run on the Apple II. Driven by the software, schools outside Minnesota soon came knocking.

Apple tasted the benefits. In 1982, when California introduced a tax - relief policy, Steve Jobs seized this opportunity and donated Apple IIe computers to about 9,000 schools across the state at once. By 1984, Apple had captured about half of the computer market in American primary and secondary schools.

Steve Jobs later said in an interview: "Schools' purchase of the Apple II was one of the key factors in Apple's success."

Does it sound familiar?

Replace "Apple II" with "G1 humanoid robot", "schools" with "universities and research institutions", and "Minnesota Educational Consortium" with "single - source procurement from Westlake University", and the structure of the first half of these two stories is almost identical.

Both had leading hardware capabilities, the education market was the first to pay, and they obtained pricing power through scarcity.

The interesting part comes later. The Apple II had been popular in schools for several years, with an increasing user base and decent profits, but it never really entered the business world.

Before 1979, how did business owners view personal computers? They were just toys for enthusiasts and had nothing to do with my business.

In 1979, a Harvard Business School student named Dan Bricklin watched his professor manually modify spreadsheets on the blackboard. Changing one number meant changing a large area. He thought this was so stupid and wondered why computers couldn't do it.

So he and a programmer named Bob Frankston developed a software called VisiCalc. The spreadsheet software cost $100 per copy and could only run on the Apple II.

This thing broke the deadlock.

VisiCalc gave business users the first reason to say "I must buy a personal computer". Before VisiCalc, businesses could do without a computer; after VisiCalc, a computer became a necessity.

I checked, and VisiCalc sold about 700,000 copies in six years.

The key is that a large number of business users spent $2,000 to buy an Apple II just to run this $100 software. Steve Jobs later admitted that VisiCalc's promotion of the Apple II exceeded any other single event.

This history gives us a clear conclusion:

The Apple II's transformation from an "educational tool" to a "tool" was not because the hardware itself became more powerful. There was hardly any change in the Apple II's hardware before and after the emergence of VisiCalc.

What really changed the game was the emergence of a killer application, which redefined "who pays and why".

Before that, people who bought the Apple II paid for "learning" and "exploration". After that, they started paying for "efficiency" and "output".

For the same machine, once the payment logic changes, everything is different: the market space, the growth curve, and the valuation logic.

This rule has been repeatedly verified in the technology circle, and someone even gave it a name: "killer - application - driven hardware popularization".

Desktop publishing software brought up the Macintosh, and mobile games popularized smartphones. Every time a hardware platform truly explodes, there is an application behind it that makes users feel they "have to buy".

Back to Unitree.

Unitree's current position is almost the same as that of the Apple II before the emergence of VisiCalc. It has leading hardware capabilities globally, has expanded in the education market, and has brand recognition. What it lacks is something that makes industrial users and consumers say "I must buy one".

In my opinion, this is the most crucial key to understanding Unitree's valuation. The 42 billion yuan is buying an option that "a VisiCalc will eventually appear".

04

Where is Unitree's VisiCalc? Actually, Unitree has been looking for it and has made some progress.

There is a small detail in the prospectus that few people noticed. At the beginning of 2026, Unitree's industrial - grade embodied large model, UnifoLM - X1 - 0, was piloted in its own factory. The robots could assemble joint motors on their own.

Using its own robots to assemble its own motors. This scene is a bit like "raising oneself".

This is the third model in the UnifoLM series. The first two have been open - sourced: WMA - 0, released in September 2025, can understand how the physical world operates. VLA - 0, released in January 2026, can see, hear, and perform operations.

VLA - 0 is modified from Alibaba's Qwen2.5 model and was trained with about 340 hours of real - machine data. One model can handle 12 different types of operation tasks. In academic evaluations, its spatial perception ability is on par with Google's Gemini robot model.

The path from the paper to open - source to factory implementation is clear. 2 billion yuan out of the 4.2 billion yuan in fundraising is being invested in the embodied large model along this line.

However, the prospectus immediately states:

Currently, global embodied large models are still in the R & D and testing stage. The company has not yet installed its large model on machines for large - scale use.

There is a wide gap between "pilot deployment" and "large - scale use".

Being able to assemble motors in its own factory is completely different from being able to do various tasks in an unfamiliar factory. This is similar to the problem faced by the creators of VisiCalc. They developed the prototype of the spreadsheet but were initially worried that it was too general and users didn't know what to do with it.

Later, they created a large number of demonstration templates for budget - making, real - estate analysis, inventory management, etc., and taught users "you can use it like this" for each scenario.

A killer application needs to be closely associated with specific scenarios and wait for users to discover that "they really can't do without it".

Wang Xingxing has his own definition of this critical point.

At the Yabuli Forum in March this year, he said that the "ChatGPT moment" for embodied intelligence is when you take a robot to an unfamiliar place and it can complete 80% of the tasks with voice commands in 80% of the scenarios. He believes this moment is at least two or three years away.

The "double 80%" is Unitree's version of VisiCalc.

The