Economic Observer reporter Zhou Yue
In the field of embodied intelligence, a rare "thousand-unit order" has emerged.
On September 2nd, Stardust Intelligent (Shenzhen) Co., Ltd. announced a strategic partnership with Xian Gong Intelligent, planning to deploy over a thousand AI robots in phases for scenarios such as industrial manufacturing and warehousing logistics within the next two years. This figure is relatively small in the current market - most of our peers only place small-batch orders of a few dozen or a hundred units.
Behind this company, which has only been established for three years, stands a string of heavyweight investors: Ant Group, Yunqi Capital, Daotong Capital, and a fund with a deep connection to the former ByteDance Group - Jinqiu Fund.
Jinqiu Fund was founded by Yang Jie, the former head of the Finance and investment department of ByteDance Group. The core team members mostly come from the investment line of ByteDance Group in the past. The fund name is derived from the first office of the former ByteDance Group, "Jinqiu Home". This fund has continuously heavily invested in the smart home sector over the past two years. It first led the Series C investment in Ushu Technology and now is betting on Stardust Smart.
At present, Stardust Intelligence has completed hundreds of millions of yuan in Series A and Series A+ financing. Its founder and CEO, Lai Jie, has been deeply involved in the fields of artificial intelligence and robotics for 17 years. He was once the first employee and core architect of Tencent's Robot Lab.
Fang Ke, co-founder and CFO of Stardust Intelligence, told Economic Observer in an interview that these robots will gradually undertake key tasks such as material distribution, turnover box handling, loading and unloading, and empty box recycling on the production line. Those dull, repetitive and even potentially unsafe work processes will gradually be handed over to robots for handling.
He believes: "Robots must enter real application scenarios as soon as possible." The deployment of a thousand units is by no means merely a market order; it is an important starting point for data return, continuous algorithm optimization, and iterative upgrading of the robot body.
The thousand-unit test has evolved from manual assembly to large-scale mass production
From the current industry situation, the deployment of humanoid robots by the vast majority of enterprises is still at the pilot stage of dozens or hundreds of units, while orders of thousands of units represent a leap from laboratory products to industrial mass production.
The Shenzhen Artificial Intelligence Industry Association predicts that by 2025, the shipment volume of humanoid robots in China is expected to exceed 20,000 units. As of September, only a few leading enterprises such as Yushu Technology, Zhiyuan Robotics and Songyan Power have officially announced that they have received orders of over a thousand units this year. Stardust Intelligence's entry into the "thousand-unit club" to some extent indicates the maturity of its technology and products.
Currently, the research and development version robot S1 of Stardust Intelligent is priced at approximately 500,000 yuan per unit, including a remote operation platform and a software development kit (SDK). The delivery model for this order is S1 and will be adjusted according to the scenario.
The Astribot S1, which was released in 2024, is currently the representative product of Stardust Smart. In public demonstrations, this robot has been able to complete a variety of complex tasks such as cooking and making tea, ironing and cleaning, playing music and dancing, and competing in cup stacking. The product has been put into use in multiple scenarios at home and abroad, including universities, enterprises, and data centers.
Fang Ke told Economic Observer that the order will start delivery in the fourth quarter of this year. The first batch of application scenarios will focus on industrial manufacturing, warehousing and logistics. In the future, it is also planned to expand to industries such as computer communication consumer electronics (3C), new energy, semiconductors, construction machinery and biomedicine with the help of Xian Gong Intelligent's customer network.
To complete the delivery of a thousand units, the primary challenge that Stardust Intelligent faces is the transformation of its production model. The manufacturing of humanoid robots still mainly relies on manual assembly. Orders of thousands of units require enterprises to establish a standardized production system. For this reason, Stardust Intelligence has restructured its supply chain system over the past six months.
In terms of product design, they adopted the principle of "usability first", developed a teleoperation system suitable for ordinary users, and pre-set modular SDK interfaces to facilitate subsequent functional expansion and customized development by customers.
For the robot body transmission solution, Stardust Intelligent has chosen a relatively niche but more humanoid rope-driven transmission system. Unlike traditional rigid motor drives, this solution mimics the contraction and relaxation principles of human muscles, enabling faster response speeds, more precise operational flexibility, and safer human-machine interaction performance.
This technical solution has been verified in complex tasks. For instance, it has demonstrated outstanding flexibility and execution ability in long-sequence tasks (such as making tea and coffee), high-dynamic operations (such as competitive cup stacking and basketball shooting), and high-precision force control tasks (such as playing the yangqin).
However, the rope-driven technology route also comes with considerable engineering challenges. The Stardust Intelligence team has introduced mechanical modeling technology that combines rigid and flexible materials, and continuously optimized the transmission mechanism and manufacturing process to address technical challenges such as friction loss, precise control of rope tension, and system integration.
Fang Ke said that the current robot products of Stardust Intelligent have passed the 24-hour continuous operation test and can work stably for 3.2 to 6.2 years in multiple scenarios, and support modular parts replacement. A new generation of products for industrial applications is under development and is expected to be released in the second half of this year, with further optimization in cost control.
Industrial collaboration and in-depth cooperation with Xiangong Intelligence
The collaboration between Xingchen Intelligence and Xiangong Intelligence is not merely a simple purchase order, but rather an experiment of industrial synergy.
Xian Gong Intelligence was founded in 2020, with its headquarters in Shanghai. It has now developed into a globally leading supplier of robot controllers. According to its prospectus, citing data from CIC, Xiangong Intelligence ranked first globally with a 23.6% market share in 2024 and has held the title of global shipment champion for two consecutive years. Its customer base has soared from 380 in 2022 to 832 in 2024. Its business footprint spans 65 countries and regions, and its service areas cover over 20 sub-sectors including 3C electronics and automotive manufacturing.
With the help of Xian Gong Intelligence's industrial ecosystem network, Xingchen Intelligence's humanoid robots will have the opportunity to quickly enter diverse industrial scenarios.
From a business perspective, this cooperation also reflects the bets of both sides on the business models of emerging technologies. In May 2025, Xiangong Intelligence submitted its prospectus to the Hong Kong Stock Exchange for the first time, planning to list on the main board of the Hong Kong Stock Exchange under the 18C chapter model.
According to the prospectus data, Xiangong Intelligence's revenue in 2024 reached 340 million yuan, with a gross profit margin approaching 46%, and the proportion of R&D investment to revenue exceeded 20%. Despite maintaining a compound annual revenue growth rate of 35.7%, with revenue increasing from 180 million yuan in 2022 to 340 million yuan in 2024 and gross profit margin remaining stable at over 45%, Xiangong Intelligence has yet to turn a profit. From 2022 to 2024, Xiangong Intelligence suffered net losses of 30 million yuan, 50 million yuan and 40 million yuan respectively. Its commercialization path is still being optimized.
Fang Ke said that the expectations of partners such as Xiangong Intelligence for this cooperation are multi-dimensional: there is both the immediate need for cost reduction and efficiency improvement and the strategic consideration for long-term algorithm iteration and generalization capability enhancement. The core of this cooperation is not merely a simple hardware connection, but a deep integration with the intelligent technology system of Stardust - a complete solution composed of the robot body, the remote operation data collection platform, and the AI algorithm model.
He believes that the remote operation system of Stardust Intelligence is a core capability that is easily overlooked by the outside world but is actually extremely crucial. Teleoperation technology is not only a high-quality data source for training AI models, but also an important technical guarantee for handling sudden abnormal events and ensuring service continuity and stability in actual delivery scenarios.
Three-stage teleoperation strategy
The current technological paths in the humanoid robot industry are highly differentiated. The pure AI route represented by Tesla is committed to achieving end-to-end autonomous decision-making through large models. The traditional approach represented by Boston Dynamics focuses on the dynamic optimization of the mechanical body. Emerging enterprises represented by Figure and 1X, on the other hand, are exploring a hybrid model of AI and human-machine collaboration.
Against this backdrop, teleoperation technology is becoming the focus of industry attention. Unlike the early simple concept of remote control, modern teleoperation systems have evolved into an integrated technical platform that combines data collection, AI training, and real-time intervention. The core logic of this technical path lies in: filling the current capability boundaries of AI in complex scenarios through human-machine collaboration, while accumulating training data for future full autonomy.
According to the technological evolution plan of Stardust Intelligence, the teleoperation system will play a core role in three key stages:
Phase One: Short-range remote operation data collection period. Through interactive equipment such as VR headsets and force feedback devices, high-quality multimodal heterogeneous data is collected to provide rich learning samples for AI model training. The focus of this stage is to establish a complete data foundation, covering operation modes, environmental variables and task sequences in various industrial scenarios.
Phase Two: AI+ Remote Operation collaborative Deployment period. The AI system takes the lead in handling standardized and repetitive tasks, while the remote operation system specifically covers corner cases and abnormal situations, forming an "AI+ human" collaborative operation mode. This division of labor mechanism not only ensures the stability of delivery but also continuously optimizes the generalization ability of AI models, driving the data flywheel and forming a virtuous cycle.
Phase Three: Remote Remote Operation as a safety net period. Even after AI achieves high generalization, remote operation will still exist as a long-term technical guarantee mechanism, similar to the role of a "remote safety officer" in autonomous driving systems, providing the last line of defense for safety in extremely complex scenarios to ensure the continuity and reliability of services.
Fang Ke said that this remote operation system is not only a core component of the technical architecture of Stardust Intelligent products, but also a long-term commercial delivery strategy. This order of a thousand units is a crucial starting point to verify whether the entire robot industry chain can truly run smoothly.
(Author: Zhou Yue) Source: Economic Observer Online