More than half of the first year of mass production of humanoid robots has passed. After Unitree and Zhipu, the industry is calling for the "third player".
The humanoid robot industry is currently in a crucial phase of climbing the slope from technological breakthroughs to commercial validation. The parallel exploration of multiple paths not only demonstrates the authenticity and complexity of market demand but also accelerates the overcoming of technological bottlenecks and the reduction of cost curves in practice. This seemingly "wild" growth is actually an inevitable process for the industrial ecosystem to find the optimal solution through dynamic trial and error.
Looking back, the much - questioned statement of "the year of mass production" at the beginning of the year is now being vindicated by real orders.
When Zhipu and Unitree jointly won orders worth over 100 million yuan, Zhongqing achieved the delivery of thousands of units, and Leju reached the situation of "deliver as soon as the product comes off the production line" due to high demand, the industry landscape has quietly changed. Humanoid robot manufacturers are no longer confined to laboratory research and development and troubled by market development. Their core concern has shifted to how to quickly increase production capacity to meet the surging delivery demands.
Jiang Zheyuan, the founder of Songyan Power, told the Embodied Learning Club that the current order volume has exceeded 2,500. Galaxy General also said that it has received orders from 100 stores.
The dawn of commercialization undoubtedly illuminates the possibility of the industry's "self - hematopoietic" ability, but this light does not shine evenly. In the robot body track, as the pattern gradually becomes clear, the Matthew effect is becoming increasingly prominent: Unitree and Zhipu in the first echelon have abundant orders, while many mid - tier enterprises are still struggling with double - digit or even single - digit order volumes.
Even within the leading camp in the spotlight, there are significant differences in development paths: some enterprises adhere to the end - game thinking, continuously delving deep into "general - purpose" humanoid robots, aiming to cover the widest range of potential needs with a single product platform; others adopt the "laying eggs along the way" strategy, giving priority to meeting the customized needs of specific scenarios or large customers, achieving commercialization through rapid iteration, and gradually approaching their ultimate goals in practice.
Digging deeper, the underlying logic of the choice of commercial paths focuses on the debate over the commercial nature of humanoid robot products. Are they standardized products that can be mass - replicated, or non - standardized products that need in - depth customization to release maximum value? The former pursues to meet the common needs of multiple scenarios with a single hit product; the latter deeply releases the potential of productivity through customized solutions, although it may increase the cost of a single delivery.
The divergence of paths reveals different choices of enterprises in strategic positioning and core capability building. This is both a process of natural market selection and an inevitable path for the industrial ecosystem to mature.
We don't have a crystal ball to make judgments on the choices of any enterprise. However, we can catch a glimpse of the future through the current keywords of commercialization.
The first echelon is gradually taking shape, and the "third place" is on the horizon
Whether it is technological leadership or continuous growth in order volume, there is no doubt that Unitree and Zhipu have firmly secured their positions in the "first echelon" of humanoid robots. This leading advantage is accelerating the formation of a positive cycle: end - customers are increasingly inclined to choose leading products with market validation and brand recognition in their purchasing decisions, and the industry trend of "the strong getting stronger" continues to be strengthened.
Especially after rental companies purchased Unitree robots, they automatically became "traffic generators." In order to boost their rental business, they kept creating interesting content with the robots. It is no longer new for robots to make a fool of themselves. Robot rappers and robot short dramas are even more interesting. When Unitree robot videos are everywhere on social media, Unitree has completely opened up the C - end market thanks to the rental companies' sense of traffic.
The name "Unitree" is like a golden edge on humanoid robots, attracting everyone's attention.
Image source: Unitree Technology
According to public information, in March 2025, Unitree won an order for 10 Unitree humanoid robots from Tongji University with a winning bid price of 8.25 million yuan. After the CES 2025 exhibition, Unitree signed a procurement agreement for hundreds of quadruped robots with a certain warehousing and logistics company in the United States. Unitree won the bid for a subsidiary of China Mobile with 46.05 million yuan, including small - sized humanoid biped robots, computing power backpacks, and five - fingered dexterous hands...
Unitree has not yet announced its total order volume for this year, but according to the Embodied Learning Club, Unitree is one of the few humanoid robot manufacturers with real orders in the C - end market.
It is worth mentioning that traffic is just the icing on the cake for Unitree's commercialization. As early as 2024 before it became well - known, Unitree's annual revenue had reached the level of 100 million yuan. Investors once said that Unitree has been profitable for many consecutive years.
Different from Unitree, which benefits from the traffic boost of its robots, Zhipu has its own built - in traffic. The label of the genius founder has not only been deeply engraved in the minds of industry insiders but also left a mark on the outside world. Moreover, Zhipu is not just relying on this "calling card." Whether it is the cooperation with PI on the brain side to break through complex long - range actions, the new product combining the advantages of wheeled and bipedal robots, or the Lingxi X2 focusing on "interaction," all these indicate that Zhipu has "product strength."
Image source: Zhipu
In just July, Zhipu has frequently announced winning bids. First, it won the bid for the procurement project of data collection and exhibition robots of the Zhuhai Embodied Intelligence Application Innovation Center with a contract value of 12.7362 million yuan. Then, it won the bid for the "full - sized humanoid biped robot" package of a subsidiary of China Mobile with 78 million yuan.
Unitree and Zhipu have proved that in the commercialization of the humanoid robot body track, technological foundation and a "reason to be seen" are indispensable. This is exactly the key engine to achieve "self - hematopoietic" and continuous leadership. After all, humanoid robots need to be fed with a continuous stream of real - world scenario data. Enterprises that deliver products first have the initiative in obtaining scenario data, which can better feed back the technology of the next - generation products, thus forming a virtuous development cycle.
While Unitree and Zhipu are running blindly on the order path, the rest of the players in the track are left in the smoke of the "Matthew effect." If they don't become well - known soon, they may be permanently and comprehensively outpaced by the strong.
So, the "third place" in the humanoid robot field is on the horizon.
Some enterprises want to replicate Unitree's traffic myth and keep going on the path of creating interesting content; some are polishing their technological differentiation in order to break through; some choose to strongly bind with a single scenario to form a small - scale brand image; others aim for everything and complete validations in multiple scenarios to cultivate a "versatile warrior."
However, despite the bustling competition in the industry, there is still no final conclusion on who will be the third place (the evaluation criteria are multi - dimensional, not just single dimensions such as technology and commercialization). The Embodied Learning Club has learned from many industry insiders that there are currently several popular candidates, each with its own strengths.
An indisputable fact is that commercialization has become the "turning point" of the industry pattern. To grasp the turning point well, technology, orders, and traffic are all essential. Planting a mental anchor can leverage long - term development.
The three - layer divergence of commercialization routes
Commercialization has become an inevitable requirement for the industry's development, and different enterprises have different views on the key path choices.
Divided by levels, the top - level path divergence lies between commercial and industrial scenarios. The "factory faction" believes that humanoid robots are more suitable for industrial scenarios. The reason is that factories have real labor shortages, and the rigid demand can not only maximize the productivity value of humanoid robots but also represents the potential for future commercial scale. Secondly, the structured environment and standardized processes in factories are more suitable for the current technological level of humanoid robots, which can first enable humanoid robots to be used and iterated in real scenarios.
The "commercial faction" believes that the automation standard in factories is very mature. The work processes that can be replaced by machines have been deployed for many years, and the un - automated work processes are all complex operations, which are not suitable for the current application of humanoid robots. Moreover, the fault - tolerance rate in factories is extremely low, and the operation and maintenance costs and trust costs after delivery are higher. In contrast, the commercial scenario has a lower fault - tolerance rate and higher tolerance. Although it requires extremely strong obstacle - avoidance ability in non - structured and highly dynamic scenarios, the demand instructions in the commercial scenario are relatively standardized, and overall, it does not set a high threshold for the technological level.
For example, Wang Qian of Independent Variable Robot once said that the factory scenario avoids the part where the model is best at. The service industry is relatively more open, with relatively complex long - sequence tasks. Complex scenarios can truly promote the effective improvement of the model's ability and are also the areas where there are real customer needs, users are willing to pay, and can replace the things that humans are reluctant to do.
Although the two factions have different focuses, they are actually exploring the same logic: to make humanoid robots truly useful.
With the same underlying logic and clear goal orientation, the path divergence becomes a "minor issue." On the contrary, the efforts of the two factions are constantly expanding the application boundaries of the track.
Under the choice of scenarios, the next path divergence lies in the choice of product logic. Some enterprises choose the "laying eggs along the way" approach, including products, scenarios, and technologies. For example, the current products and matching scenarios may not be the ultimate path that the enterprise wants to focus on, but they are the most suitable for making money at present. First, start running to accumulate enough real data, proven technologies, and real orders. After having sufficient resources, polish new products to achieve the ultimate general skills and fulfill the vision of widespread application. The other part directly enters the end - game, deeply engaging in the end - game battlefield with a single product and iterating in the end - game scenario.
The most fundamental divergence, which is also the most critical choice affecting the future direction at present, is the "debate over standardized products."
Some enterprises believe that humanoid robots should be standardized products, using technology itself to meet general requirements and thus satisfying the majority of needs in most scenarios. Other enterprises believe that humanoid robots should adopt a model of standard + customization. According to the requirements of different scenarios and different customers, the functions of products can be customized. For example, for operations with low requirements for fine - grained manipulation, a two - finger gripper can be selected as the end - effector, and for higher - requirement operations, a five - fingered dexterous hand can be selected. The battery life can also be selected according to the operation intensity.
The debate over "standardized products" and "non - standardized products" actually lies in returning to the essential attributes of the application scenario. Standardized products, with their economies of scale and cost advantages, are more suitable for the commercial service field with relatively homogeneous demand. The industrial manufacturing scenario, due to the high differentiation of production lines, clear customization requirements, and strong customer payment ability, often becomes an ideal testing ground for the "standard platform + in - depth customization" solution, thus achieving better application efficiency and value release.
In short, there is no optimal solution, only the most suitable one.
The three - layer divergence indicates that the industry's commercialization is still in a wild growth stage, but it is not chaotic. Instead, it is an inevitable stage before the industry matures. The path choices at different levels and dimensions are actually different answers of enterprises to the core proposition of "how to most effectively transform technology into user value" based on their own endowments, technological reserves, and market judgments. The "rigid demand" and "structured environment" in the industrial scenario provide a solid foundation for technological verification and value release; the "inclusiveness" and "relatively standardized demand" in the commercial service open up a unique channel for early - stage large - scale development and user education. The "laying eggs along the way" approach accumulates valuable practical experience and survival capital, while the "end - game thinking" guides the ultimate direction of technological innovation. The debate over "standardized products" and "non - standardized products" will also finally find the optimal balance point under the principle of scenario adaptation, with the standardized core platform ensuring efficiency and cost, and the scenario - based customized modules releasing in - depth value.
Looking at the overall situation, the humanoid robot industry is currently in a crucial phase of climbing the slope from technological breakthroughs to commercial validation. The parallel exploration of multiple paths not only demonstrates the authenticity and complexity of market demand but also accelerates the overcoming of technological bottlenecks and the reduction of cost curves in practice. This seemingly "wild" growth is actually an inevitable process for the industrial ecosystem to find the optimal solution through dynamic trial and error.
This article is from the WeChat official account "Embodied Learning Club", author: Lv Xinyi. Republished by 36Kr with permission.