Official warning: Humanoid robots are in danger.

On November 27th, after Goldman Sachs' investigation found that there were no large - scale orders for humanoid robots at all, the National Development and Reform Commission also poured cold water on this industry.

Li Chao, the deputy director of the Policy Research Office of the National Development and Reform Commission, made the following statement:

Currently, the technological routes, business models, and application scenarios of humanoid robots are all immature. There are already over 150 enterprises across the country, and more than half of them are startups or companies entering the field from other industries. This is good for innovation, but two risks need to be guarded against: one is the concentrated listing of highly repetitive products, and the other is the compression of R & D space.

Note that the National Development and Reform Commission didn't mention "overcapacity". With 150 enterprises, the global shipments in 2024 are only in the thousands. It's clearly in the exploration stage. Where does the overcapacity come from?

However, the frequent popularity of humanoid robots this year and the resulting attention are not unfounded. The capital behind them has heated up. The financing this year has exceeded 10 billion yuan, and the capital has been hyped up. For financing, large - scale orders are not necessary, but demos and prototype machines are essential.

There are three key pieces of information hidden in Li Chao's words: among the 150 enterprises, more than half are startups or cross - industry entrants, and we need to guard against the "concentrated listing of highly repetitive products". These two points are easy to understand. Then what does "the compression of R & D space" mean?

Just after the overcapacity of computing power, here comes the same situation with robots

Upon seeing this news, the first thing that came to my mind was the news of overcapacity of computing power in our country. Computing power, especially GPUs, is regarded as the cutting - edge of hard technology in the eyes of ordinary people, a key area that is subject to technological bottlenecks, and also represents the future of industrial upgrading. In the wave of artificial intelligence, a large number of data centers and intelligent computing centers have been launched one after another. However, according to the calculations of the Inspur Artificial Intelligence Research Institute, the average utilization rate of computing power in intelligent computing centers in our country is only 30% or even less than 20%, especially in small - and medium - sized intelligent computing centers in third - and fourth - tier cities.

On the other hand, calculations show that in 2025, China will need 3 million GPUs. High - end computing centers that can truly support large - model training are even scarcer. The low - end general computing power is seriously in excess, while the intelligent computing power that meets high - end needs is still in serious shortage.

This situation should be familiar to everyone. In fact, it is what the National Development and Reform Commission calls "the compression of R & D space". This term is a bit abstract. Put it more simply: everyone is rushing to do the easy things, and no one is willing to tackle the tough problems.

All localities are building intelligent computing centers - this is easy to do. Just buy servers, build computer rooms, and do the supporting facilities. Throw in the money and you can see the achievements. But for the high - end computing power technologies that really need to be broken through, such as large - model optimization, computing power scheduling algorithms, and research on storage chips - these technologies that take 5 to 10 years to tackle, few companies are working on them.

This is a common feature of almost all strategically developing industries. Specifically, R & D personnel and R & D funds are thinly spread. The market and capital are more willing to invest in short - term, quick - return homogeneous projects, resulting in difficulties for exploratory R & D, original routes, and high - risk directions to obtain resources. In the end, a lot of money and resources are invested, but they are all for showy projects and superficial efforts, and no one touches the real technological bottlenecks.

Specifically for robots, it does seem bustling: the financing has exceeded 10 billion yuan, new product press conferences are held one after another, and videos of robots dancing and doing somersaults have flooded social media. But investors' private judgments are very calm: "The threshold for assembling a humanoid robot is no longer high."

What does "assembling" mean? Of course, they don't use the word "assembling". Instead, they say "the difficulty of overall machine integration has decreased". With China's supply - chain capabilities, almost all the components are ready - made. A demo can be made in 6 months, and then a demonstration video can be shot for financing.

But what about the core technologies?

Reducers, sensors, and joint modules - these core components that account for 50 - 60% of the cost of robots are still monopolized by foreign countries.

To my surprise, in conversations with industry insiders, I found that humanoid robots don't have as high a technological barrier as expected. For example, it was quite rare for a robot to do a somersault last year, and Unitree's performance was amazing. But now, after seeing all kinds of robots playing ball, dancing, and doing martial arts, to be honest, people are a bit tired of it. Now a lot of companies can do it. In fact, the cutting - edge progress is extremely fast, but most enterprises are still complacent about making homogeneous products recently, thinking that they have kept up with the pace of new - quality productivity.

The result is that 150 enterprises are making highly homogeneous products, all waiting for a market that hasn't arrived yet. And for the technologies that really need to be conquered, such as reducing the cost from $50,000 to $20,000 and replacing imported reducers with domestic ones - few companies are willing to tackle these tough problems.

A lot of money has been spent, and many enterprises have flocked in, but no one touches the key bottlenecks of the industry. Isn't this a repeat of the overcapacity of computing power?

Is the humanoid robot industry following the old path of computing power?

Let good money drive out bad money

Of course, it's an overstatement to say that all 150 enterprises are making homogeneous products.

On November 24th, the Ministry of Industry and Information Technology announced the list of members of the Standardization Technical Committee for Humanoid Robots, which includes many familiar names. Wang Xingxing, the founder of Unitree Technology, Peng Zhihui (ZhiHuijun), the co - founder of Zhipu Robotics, and executives from enterprises such as Ubtech, Xingdong Jiyuan, Xiaomi Robotics, SenseTime, Chery, Huawei, and XPeng were selected as members.

These can be roughly divided into three categories: leading startup robot enterprises that have already made shipments, such as Ubtech, Unitree, and Zhipu; enterprises with technological accumulation in software, such as Huawei, Xiaomi, and SenseTime; and automobile enterprises with an industrial foundation in hardware, such as Chery and XPeng.

Those who can participate in formulating standards must have made large investments, have high R & D costs, and are accumulating experience step by step, truly creating new products that meet the needs of the scenarios. Letting them participate in formulating standards is to let good money drive out bad money. Even if we can't clear the market, at least it will put more pressure on bad money.

So who are the bad money?

Based on my experience, they can be roughly divided into three typical types of players.

First are the concept - hyping companies in the Hong Kong stock market (mainly humanoid robot companies that are not listed in the A - share market and have flocked to the Hong Kong stock market). Originally, they were making sweeping robots or industrial robotic arms. Now, by adding the concept of "humanoid", their stock prices can double. There is a joke in the industry: some companies' humanoid robots are actually just traditional robots with two legs added, and they claim to be "laying out for the future". Buying components for assembly has limited technological content and is essentially just a way of market - value management.

The second type is those who are crossing industries to ride on the hot trend, which is also mentioned in official documents. Various pension companies, hotel management companies, etc., claim that they have scenario resources and it is reasonable and compliant for them to cross into the robot industry. They conduct some capital operations such as mergers and acquisitions or joint ventures and claim to be participating in this wave of the humanoid robot trend. Some even dare to call themselves high - tech enterprises.

The most numerous are assembly factories. This type is the most complex and needs to be divided into two kinds:

One kind has the ability to transfer technology. For example, automobile component manufacturers, which have accumulated experience in reducers, sensors, and motors, can reasonably transform to produce robot components. Even some automobile enterprises (Chery, XPeng) directly produce complete machines, which also makes sense - because the underlying technologies, supply chains, and customer resources are all connected.

But more are pure assemblers. They purchase standard parts from Tuopu and Sanhua, make a shell and do system integration by themselves, and then claim to be "independently developed". The logic of these manufacturing enterprises is very simple: "If I don't make robots, I'll be left with backward production capacity." So, regardless of whether they understand or not, they rush to get ahead. Don't understand? Can't they just poach high - paid talents? As a result, just like the new - energy vehicle industry back then, it has blossomed everywhere.

Be more tolerant of bubbles and more vigilant against blindness

When I make an analogy with new - energy vehicles, many people should easily think that local governments may be making blind investments again.

In the past year, at least a dozen provincial and municipal governments across the country have intensively announced the construction of robot industrial parks or introduced special support policies. Needless to say about Beijing, Shanghai, Guangzhou, and Shenzhen. What's even more dazzling is that this trend has spread not only to the provincial and municipal levels but also to the district and county levels.

Let's first look at the competition within Beijing. In August last year, Yizhuang released the "Action Plan for Building a World - Class Embodied Intelligent Robot Industrial New City", aiming to achieve an annual production capacity of tens of thousands of units by the end of next year and build a 250,000 - square - meter Yizhuang Robot Industrial Park; Haidian District, relying on the innovation resources of Zhongguancun Science City, has become a center for embodied intelligent R & D; Changping District plans to build a 300,000 - square - meter Changping Robot Industrial Park.

It's quite rare in other industries for at least three districts in one city to build robot industrial parks simultaneously.

The Yangtze River Delta has also blossomed everywhere. Not only is Shanghai Pudong preparing a 10 - billion - yuan fund and building the Zhangjiang Robot Valley, but also the Wuzhong District of Suzhou has set up a 10 - billion - yuan industrial fund to build the "Number One District in the Country". In Hangzhou, Wuxi, and Changzhou, as long as you search, various investment - promotion and industrial - park supporting preferential policies will come pouring in.

Midwestern cities are also not willing to lag behind. Chengdu released a three - year action plan last year, aiming to break through 60 billion yuan in the full - industry - chain scale next year. A 10 - billion - yuan industrial fund was registered and established in October last year; Xi'an has included robots in the 34 key industrial chains at the provincial level; The production line of humanoid robots in Wuhan Optics Valley was partially put into operation in April this year; The Zhengzhou High - tech Zone hosted the World Robot Contest last year...

At present, the competition in the humanoid robot industry is increasingly trending towards urban - ecosystem competition. The old methods of providing tax incentives, quickly approving industrial parks, and offering settling - in allowances may not work anymore.

For example, the marathon held in Beijing saved a lot of public - relations costs for embodied - intelligence companies and attracted the attention of many investors and consumers. Such innovative ecosystem - cultivation measures require local leaders to truly understand the needs of entrepreneurs.

What's even more noteworthy is the "ambition" and "similarity" in goal - setting. Yizhuang in Beijing aims to achieve an annual production capacity of "tens of thousands of units" next year, Shenzhen aims to reach 100 billion yuan in the related - industry scale by 2027, Suzhou aims to support the city's robot - industry scale to reach 200 billion yuan, and Chengdu aims to reach 60 billion yuan in the full - industry - chain scale next year.

But the reality is that according to the Bain & Company report, the global shipments of humanoid robots in 2024 are only in the thousands. Currently, they are mainly used in scientific research, guided tours, and a few industrial - manufacturing fields. Facing such a narrow range of application scenarios, it's difficult to achieve large - scale production. Will these lofty goals lead to a repeat of the scenarios during the boom of photovoltaic and chip industries?

In my experience, when I mention humanoid robots to people outside the industry, the first reaction is usually "bubble", "hype", and "making money". There are few other reactions. The most common question is why robots have to be in human form. Even investment tycoon Duan Yongping has asked such a question. It shows that people always look at new things with suspicion, which is understandable.

But I've always been calling for more tolerance for bubbles. When a large - scale and promising emerging industry emerges, severe overcapacity is almost inevitable. This is because no one in the market knows the future stable demand level of the industry. They can only keep trying and making mistakes. It's not until the expected demand growth fails to materialize that the market realizes the approximate level of stable demand. By then, the industry is already in a state of obvious overcapacity. Only after this situation ends can the industry truly become a mature and stable one. Even in the United States, a country that claims to have little government intervention in economic activities and is completely market - driven, such problems have occurred in industries such as railways, automobiles, telecommunications infrastructure, and computers. This shows the universality of this phenomenon.

China's investment system undoubtedly exacerbates this overcapacity. Not only do investors have a positive outlook based on their own investigations, but also the top - level design sets the tone. It's almost a foregone conclusion that everyone will rush in regardless of the cost.

Conclusion

In 2015, more than 200 prefecture - level cities across the country planned a production capacity of 20 million new - energy vehicles, while the actual sales volume that year was only 330,000. Currently, the humanoid robot industry is still in its early stage. It's only reasonable to formulate industry standards as soon as possible to correct the deviation.

But what's more worrying at present is not the "bubble" itself - it's precisely because of the bubble that a large amount of funds and resources are attracted. The chairman of Bain & Company said that planetary roller screws, multimedia sensors, and motors basically account for 50% - 60% of the current total cost of humanoid robots. Once cost reduction and technological breakthroughs are achieved, they will have significant competitiveness in the global industrial chain. Currently, foreign manufacturers are still leading in the core - component manufacturing of humanoid robots. Also, in terms of scenario development, the C - end has rich and diverse scenarios, but most manufacturers are still focusing on the highly certain industrial demands. There are still many meaningful tasks and R & D spaces worthy of investment.

It doesn't mean that the central and western regions and local counties and cities have no opportunities to develop humanoid robots. According to current projections, the future of humanoid robots will penetrate into every household and spread to various fields. Relying on the characteristic industrial bases and scenarios of each place, I believe there is a possibility of a "hundred - flower - blooming" situation in the future.

This article is from the WeChat public account "Rongzhong Finance" (ID: thecapital). The author is Fu Qisen, and the editor is Wu Ren. It is published by 36Kr with authorization.