Humanoid robot manufacturers are learning to live frugally.

Short - term capital bets cannot draw a sexy business curve. To tell a complete business story, it still depends on the humanoid robot manufacturers themselves. In other words, "self - hematopoietic ability" has become a new issue for complete machine manufacturers.

The attitude of humanoid robot manufacturers has changed. They are now calmer and discouraging the outside world from viewing the situation irrationally.

Entering 2025, an obvious trend is that robot manufacturers are busy "showing off their muscles". There are endless videos of robots dancing, doing somersaults, and demonstrating skills, and the manufacturers' promotional language depicts a prosperous development scene. Just a few months later, the promotional efforts have significantly cooled down. Manufacturers are no longer focused on how to stimulate the outside world's interest but are bringing the production line to the forefront, telling a story about the productivity of embodied intelligence. In their public statements, they no longer talk about the "era of every household having a robot" but respond in a more reserved manner.

"It's very difficult to reach every household at this stage."

"Humanoid robots are far from entering the era of productivity."

"They can't work in factories for now."

All of a sudden, major humanoid robot manufacturers are changing their strategies. Even the so - called "top - tier" companies favored by capital are trying to lower the outside world's expectations. The reason for this change is related to the word "health". Although excessive attention is important, technological development is not an overnight process. If the narrative is overly hyped while the technology is difficult to implement, it will inevitably lead to an unsatisfactory outcome, which is not conducive to long - term and healthy development.

This strategic shift is also reflected in the business lines. Several complete machine manufacturers are no longer pursuing the rather bold "full - stack self - research" and "universality" but are looking for more practical implementation methods to break through the barriers in the humanoid robot field.

Yes, in a seemingly least practical field, people are now starting to talk about practicality.

After all, short - term capital bets cannot draw a sexy business curve. To tell a complete business story, it still depends on the humanoid robot manufacturers themselves. In other words, "self - hematopoietic ability" has become a new issue for complete machine manufacturers.

Universality does not mean perfection. Specialized commercialization is clearer.

History has proven that absolute "universality" is not a perfect solution. When a new standard - setter emerges, universality can become "useless".

Twenty years ago, there was a must - have product called the "universal charger", which was one of the best solutions for charging all mobile phones at that time. Thus, the universal charger had a certain degree of "universality". However, when smartphones replaced feature phones and the batteries were no longer removable, and new interface protocols emerged in an endless stream, all other specialized mobile phone accessories could survive in different ways, but this absolutely universal product became useless.

The humanoid robot field is facing a similar enlightenment. At present, when the technical standards have not been established, the industrial consensus is yet to be formed, and the maturity of the application ecosystem is still unclear, if one is overly obsessed with building universal capabilities, it may not only hinder its long - term development path but also pose severe challenges to short - term commercial implementation and value realization.

From the perspective of implementation, whether it is for general use in factories or in family scenarios, there is still a long way to go. Currently, there are still obstacles for humanoid robots to enter relatively closed factory scenarios, and the difficulty of transitioning from factories to family scenarios is an exponential leap. That's why Wang Xingxing frankly said that it's not going to happen in the next few years for humanoid robots to enter families, and most enterprises also say that it will take 5 - 10 years.

Image source: Unitree Robotics

However, the real - world application scenarios have a real and urgent demand for robot productivity. The large - scale deployment of mobile robots in fields such as factory logistics and the verification of their commercial value have fully demonstrated the rigid demand for robots as a form of productivity.

So, when the general humanoid robot is still in its infancy, what kind of productivity can meet the urgent needs of the application scenarios considering the current supply capacity?

Fu Sheng, the CEO of Cheetah Mobile, has put forward the view that starting from market demand, focusing on "small scenarios" with clear structures and rigid demands may have more commercial value than blindly pursuing the "universal robot" that aims for comprehensiveness.

There are already real - world cases on the market to support this view.

One approach is to add features to products. For example, Roborock and Dreame both launched vacuum cleaners with robotic arms this year. By adding the picking - up ability to the original 2D floor - cleaning function, these vacuum cleaners have transformed into 3D space cleaning assistants.

This approach of adding features to products has also been verified among humanoid robot manufacturers. For example, for a shopping - mall guide robot, is it enough just to be able to listen and speak? Of course not. Digital Huaxia chose to make its humanoid robot's face more lifelike so that its image is more acceptable to customers. Or, the recently launched Lingxi X2 by Zhipu AI can initiate interactions, which means the robot can communicate with users spontaneously instead of just responding passively. Therefore, for shopping - mall guide robots, won't products with a more lifelike face and the ability to initiate communication have more market potential?

Image source: Digital Huaxia

From an investor's perspective, the humanoid robot is currently in the product introduction stage. At this stage, the core is supply - driven rather than demand - driven. Borrowing Steve Jobs' words, at this stage, people's attitude towards humanoid robots is similar to when the first Apple smartphones came out. "Consumers don't know what they need until we show them our products."

Therefore, by adding features to products, users can see more uses of humanoid robots.

Another approach is to simplify the scenarios.

Some domestic investors believe that the name "humanoid robot" is a bit confusing. For products like vacuum cleaners and delivery robots, it's easy to understand their functions. However, at this stage, humanoid robots are labeled as "universal", but they still can't be mapped to a specific and rigid - demand scenario, often only leaving some room for imagination for users to pay for the future.

At present, robot manufacturers focusing on specialized and single scenarios are doing quite well. For example, Digital Huaxia has chosen to focus on the exhibition and performance scenarios. Its Xia Lan humanoid robot, with its high - value female appearance and high emotional intelligence, is applied in government service halls, exhibition halls, and other scenarios. It is reported that it has won hundreds of millions of yuan in orders this year. Moreover, companies deeply involved in hotel service robots, such as Yunji Technology, have been narrowing their losses year by year and filed for an IPO on the Hong Kong Stock Exchange in April this year.

Humanoid robot manufacturers may consider first accumulating experience and making breakthroughs in a certain field and then expanding outward to win phased victories. This can be seen from the situation of the "Six AI Dragons". Initially, they aimed for AGI. Later, some gave up training large - scale models, and some entered the vertical medical field. As Li Kaifu said in an interview, "Everyone can clearly see that only large companies can afford to train large models." The goal is to survive first.

We can't call this kind of giving - up a failure. In the industry, while some people are looking 10 years ahead, others need to focus on the present.

Maybe it's more promising to focus on quadruped robots

There is a popular internet meme, "It's so hard to be a human. It's better to be a dog." Applying this saying to the humanoid robot industry may seem like an extreme view. However, the technological challenges and commercialization paths for quadruped robots are significantly easier compared to humanoid robots.

The motion control ability of quadruped robots is more stable than that of bipedal humanoid robots. While humanoid robots are still struggling with human - like gaits and straight - knee walking, quadruped robot dogs have already achieved movement speeds that humans can't reach.

In addition to the relatively easier technological research and development, the commercialization path for quadruped robots is also very clear. Quadruped robot dogs are mostly positioned as tools. Unlike humanoid robots, there are no disputes and wild imaginations about "personality". They can be the guide dog robots launched by Magic Atom and Zhipu AI, or the industrial inspection dogs and fire - rescue dogs launched by DeepRobotics. Even if they just act cute like emotional companions, there will be a large market for them.

Image source: Magic Atom

Meanwhile, in terms of price, quadruped robots have first reached the consumer - grade level of a few thousand yuan, becoming real consumer products. In the future, ordinary consumers can expect to experience the latest products in the robot industry with the price of a mid - to high - end mobile phone.

The price of humanoid robots remains high. The price tag in the hundreds of thousands means that ordinary consumers can't afford them. So, can it be sold as a luxury item? This leads to a paradox. At the current stage of development, robots are data - driven, just like large - scale models. The more they are used, the more data is generated. Once the flywheel starts to turn, the product has the basis for continuous iteration. However, if the robot is made into a product that only a few people can afford or is only needed by a specific group (such as for scientific research), it will hinder the rotation of the flywheel.

According to Fu Sheng's view, the current market is still in a very early stage. Some false demands and mediocre products mean that "it can't even be said that humanoid robots have become an industry yet".

From this perspective, is it more practical to focus on quadruped robots first?

Since the beginning of this year, we have seen that companies well - known for humanoid robots are all launching quadruped robots. For example, after several of its humanoid robots became popular, Zhongqing Robotics launched the JS01 quadruped robot in May this year. According to Yiou.com, Zhongqing's entry into the quadruped robot field is to further supplement its product portfolio.

Zhipu AI also revealed at a press conference before the launch of its Lingxi X2 humanoid robot that it will also launch a quadruped robot D1 (tentatively named) in the second quarter of this year. When asked about the reason for this layout, a Zhipu AI representative frankly said, "There is a lot of market demand."

Therefore, whether it is to enrich the product portfolio or due to market demand, quadruped robots are becoming a must - have for humanoid robot manufacturers. After all, in the capital - intensive humanoid robot field, Unitree Robotics has been profitable for many years since 2020, and its 60% global market share in quadruped robots is the key to its self - sustainability.

Focusing on humanoid robots may be a choice that aims high but may not achieve the desired results. In this wave of robot entrepreneurship, the first wave of cold - start has ended, and how to achieve self - sustainability is a real problem that investors are posing to enterprises.

More friends, more opportunities

The technological landscape of humanoid robots mainly consists of three parts: the body, the "cerebellum", and the "brain". The industry consensus is to develop software and hardware in an integrated and coordinated manner. However, the current situation is that software and hardware development are mutually restrictive, and they can't even be synchronized, let alone develop together. Therefore, some enterprises are reluctant to give up any part of the technological landscape and regard their full - stack R & D capabilities in software and hardware as their core competitiveness, constantly building moats in the industry.

Moreover, the concept of full - stack development sounds very promising, and even investors will look more favorably on it and are willing to invest.

Currently, the so - called "full - stack software and hardware development capabilities" claimed by enterprises seem more like an ambitious goal. In the early stage of humanoid robot development, it is very difficult for enterprises to cover all technological segments. In fact, at least from a phased development perspective, humanoid robot companies generally start from developing the body, accumulate technology and funds, and then start to focus on the embodied brain. Or some enterprises directly outsource the development of the embodied brain and cooperate with model companies.

Meanwhile, enterprises focusing on the body are also being squeezed by upstream companies. The upstream core component manufacturers still have significant influence over the mid - stream body manufacturers. Motors, lead screws, and dexterous hands account for most of the cost of the body. And upstream companies can easily enter the complete machine business at any time due to their cost advantages in core components.

Therefore, in the current entrepreneurial rhythm, going it alone will slow down the product iteration process, and the urgent need for commercialization will inevitably accelerate product cooperation.

For example, the cooperation between Huawei and Leju Robotics. As early as November 2023, Leju Robotics launched its first humanoid robot, KUAVO (Kuafu), based on Huawei's open - source Hongmeng OS (KaihongOS).

In March 2024, Leju Robotics also reached a cooperation agreement with Huawei Cloud, a subsidiary of Huawei, to jointly explore the application scenarios of "Huawei's Pangu large - scale model + Kuafu humanoid robot" and build an open - ecosystem platform for "humanoid robots +". The Kuafu robot equipped with the Pangu large - scale model has significantly improved its intelligence and generalization ability, which has been verified in industrial and family scenarios.

It is understood that the total delivery volume of Leju's humanoid robots is expected to reach the thousands level this year. "Now, basically, as soon as a robot comes off the production line, it is delivered," Leju's general manager Wang Song said in an interview with the Science and Technology Innovation Board Daily.

Currently, the ecological cooperation in the embodied intelligence industry is accelerating. In the past month alone, there have been several cases:

• Kepler cooperated with Zhaofeng Machinery. Zhaofeng's humanoid robot lead screw project can provide core components for Kepler and also offer discrete industrial scenarios.

• Huawei, Ubtech Robotics, Zhipu AI, and Zhongjian Technology are cooperating on a health - care humanoid robot. The parties will cooperate in computing power, large - scale model platforms, data collection, solutions, and application scenarios.

• Yuejiang Technology cooperated with Tencent Cloud. Yuejiang values Tencent Cloud's infrastructure, audio - video solution capabilities, and technological advantages such as a full - stack self - developed multi - modal large - scale model and a full - stack cloud - native security system.

• Huawei and Ubtech Robotics are cooperating. Huawei's Ascend, Kunpeng, Huawei Cloud, and large - scale model capabilities, as well as its R & D and production experience, can help Ubtech accelerate product iteration...

To sum up, most current cooperation mainly involves large companies providing computing power and model support for robot manufacturers. For robots, the "brain" determines the upper limit of the body. In this regard, Huawei is actively deploying through ecological cooperation, and its role as a "shovel - seller" in the embodied intelligence field is becoming increasingly prominent.

In general, technology companies often tend to work in isolation for the so - called technological barriers. In fact, in the "technology - product - market" chain, technology must be quickly transformed into usable products to achieve commercialization and marketization. In the emerging humanoid robot industry, the first company to launch products and continuously iterate them will gain a first - mover advantage.

Being first is better than being the best.

Conclusion: Ambitious technological ideals need to be practical

The changes in the humanoid robot industry do not mean the abandonment of technological ideals. Instead, in the face of harsh