Will robots become the next super terminal after semiconductors when Unitree goes public?
On June 1, 2026, Unitree Technology will face the IPO review on the Science and Technology Innovation Board. On the surface, this is a milestone for a robotics company in the capital market; but for the semiconductor industry, it is more like a signal: following mobile phones, automobiles, and servers, robots may be emerging as a new super terminal in the chip industry.
Before 2025, robot products were perhaps small buyers in the chip market. Robots without AI support only needed to use programming algorithms as the core to control the hardware. However, with the emergence of the concept of physical AI, embodied intelligence manufacturers represented by Unitree are breaking this paradigm. Embodied intelligence is no longer satisfied with selecting from existing chip catalogs. Facing complex task scenarios and with the support of large models, embodied intelligence manufacturers are forcing the redesign of chip and electronic component solutions around rigid requirements such as motion control, millisecond - level real - time response, ultra - low power consumption, and multi - sensor fusion.
Embodied robots have a dozen to dozens of degrees of freedom throughout their bodies, and each joint requires an independent motor drive and encoder. The multi - modal perception pipeline of vision - inertial - force sensing requires data processing and command issuance to be completed within sub - millisecond levels. At the same time, as mobile devices, the power consumption limitations are far more stringent than those in cloud and in - vehicle scenarios. These requirements directly give rise to targeted demands for heterogeneous computing architectures, real - time control MCUs, high - performance analog front - ends, high - voltage drive chips, and customized SoCs.
Where are the semiconductor opportunities?
In its prospectus, Unitree clearly divides embodied intelligence into the "brain" and the "cerebellum". This division accurately defines two completely different chip technology paths.
(1) The high - computing - power "brain" camp: The battle for the platform of perception and decision - making
The "brain" focuses on cognitive intelligence, task planning, and decision - making, and is essentially the high - ground of end - side AI computing. Currently, manufacturers such as Horizon Robotics, Cambricon, and Black Sesame Technologies have started their layouts, with computing power ranging from 128 TOPS to over 1000 TOPS. Black Sesame's A2000X leads with an equivalent computing power of 1000 TOPS. However, Horizon Robotics, with the BPU architecture of its S600 (560 TOPS) and the open - source HoloBrain/HoloMotion ecosystem, has taken the lead in forming a platform effect in the "brain" market of humanoid robots and has been included in the reference designs of many robot body manufacturers. Cambricon is bringing down its cloud - based intelligent chip capabilities to provide an integrated solution for edge - side training and inference. In this "brain" battle, what is ultimately at stake is not just the paper - listed computing power, but the usability of the toolchain, the efficiency of model migration, and the depth of the scenario ecosystem.
(2) The precise control "cerebellum" camp: Real - time, reliability, and extreme energy efficiency
What also excites the semiconductor industry is the robot's "cerebellum". The cerebellum focuses on motion control, dexterous whole - body movement, and high - dynamic response. This corresponds to a large number of chip requirements in areas such as real - time control, embedded systems, motor drivers, sensors, and power management, and has extremely high requirements for reliability.
Currently, in this field, overseas giant Renesas Electronics firmly controls the industrial robot joint market with an annual shipment of 230 million motor - control MCUs. The domestic camp is breaking through in different ways: Rockchip's RK3588 has covered all robot forms from service robots to quadruped and humanoid robots. With its rich interfaces and heterogeneous computing power combination, it occupies a favorable position in the intermediate layer between the "cerebellum" and the "brain". Allwinner Technology holds the large - volume shipment market of service robots and consumer - grade robots with its cost - effective chips. Although the precise control track is not as eye - catching as high - computing - power chips, it has a large volume, a wide range, and high customer stickiness, and is very likely to give birth to a group of specialized and innovative semiconductor small giants.
(3) The full - stack integration type: The functional safety barrier from the automotive industry to the robotics industry
Xingchi Technology has taken a differentiated full - stack path and released a four - layer complete chip solution covering the "brain - cerebellum - torso - joint". Its core barrier lies in the complete migration of the ISO 26262 ASIL - D automotive - grade functional safety certification system to the robotics field. When robots enter factories, shopping malls, and households, functional safety will change from an optional feature to a necessity. Galaxy General Robot has become the first mass - production verification node for this solution. This safety gene extended from the automotive semiconductor industry is expected to have a dimensionality - reduction strike effect in the robotics track.
There is a view in the industry that sees embodied intelligence as the next direction for the development of the automotive industry. For robots to run, jump, grab, and avoid obstacles in the real world, the core challenges are "millisecond - level response" and "high - reliability motion control". From high - voltage motor drives, high - precision ADCs, to MCUs integrated with EtherCAT slave stations, and then to real - time SoCs with embedded functional safety libraries. This is very similar to the intelligent automotive industry chain. Therefore, many automotive - grade chip manufacturers have also set their sights on the embodied intelligence track. Dr. Wang Kai, the founder and CEO of Xinqing Technology, pointed out that the embodied intelligence industry will have a strong driving effect on high - computing - power chips. Behind this are some astonishing data: the Token call volume in the Chinese market soared from 100 billion times in 2024 to 140 trillion times in 2026, and the computing power demand increased by 1400 times in two years. And embodied intelligence, which requires real - time interaction with the environment at the end - side, is precisely one of the scenarios with the highest computing power demand and the strongest growth potential.
For automotive - grade chip manufacturers, embodied intelligence is undoubtedly a new factor stimulating business growth. The global automotive chip market is about $50 billion, while the AIoT + end - side intelligence market has reached $200 billion, 4 - 5 times the size of the former. As the ultimate form of end - side intelligence, the chip market for robots has extremely high imagination.
The robotics industry: Don't overestimate in the short term, don't underestimate in the long term
Unitree's prospectus reveals the most realistic side. The company's main raw materials include mechanical parts, electronic components, electrical materials, etc. Among them, electronic components cover capacitors, resistors, inductors, PCBs, transistors, crystal oscillators, chips, antennas, etc. From January to September 2025, the procurement amount of electronic components was 130 million yuan, accounting for 25.10% of the total raw material procurement; while in the whole year of 2024, it was only 46.2681 million yuan, accounting for 23.89%.
This data conveys two key messages:
Firstly, the short - term absolute volume is still relatively small. Even though Unitree is a global leading enterprise in embodied intelligence, its annualized chip - related procurement scale is currently only in the order of a few hundred million yuan, far from the tens of billions or even hundreds of billions of yuan in chip procurement of mature terminals such as mobile phones and automobiles.
Secondly, the rising growth rate and proportion show a strong trend of increasing chip value content. In the BOM of the whole machine, the proportion of electronic components is steadily increasing, and the growth rate far exceeds that of the whole - machine shipment volume.
In the long run, the requirements of robots for chip complexity, real - time performance, energy efficiency, and reliability will give rise to a group of brand - new semiconductor sub - markets. When robots gradually enter factories, shopping malls, and households, their demand for MCUs, analog chips, sensor interface chips, and functional safety chips may replicate the path of automotive electrification and even create a more diverse semiconductor demand matrix. In the short term, it is not advisable to overestimate the performance explosion brought about by the shipment volume, but in the long term, the potential of robots to define new chip categories must not be underestimated.
Chip manufacturers have also regarded the embodied intelligence market as an important growth scenario. On May 28, Qichacha information showed that Xixuan Chuangzhi Technology (Wuxi) Co., Ltd. was established with a registered capital of 100 million yuan. Its business scope clearly includes integrated circuit design, integrated circuit sales, and intelligent robot R & D. After penetrating the equity, the shareholders of this company include GPU chip company Muxi Co., Ltd., leading humanoid robot company Ubtech, and Fenglong Co., Ltd. Muxi previously clearly listed "embodied intelligence" as a key frontier direction in its annual report. However, its existing products (Xisi N series, Xiyun C series) are mainly for cloud - based training and inference and are not suitable for the low - power consumption scenario of the robot end - side. The joint venture with Ubtech is the first step for Muxi to migrate its GPU architecture capabilities to end - side lightweight inference chips. With the support of the whole - machine customer, the risk of the R & D route is greatly reduced.
For semiconductor practitioners and distribution channels, the most direct meaning of this news is: the demand for embodied intelligence chips is shifting from "purchasing standard products" to "customized dedicated SoCs". This round of the joint - venture model between whole - machine manufacturers and chip companies is likely to be replicated by other leading robot companies such as Unitree and Zhiyuan. Domestic end - side embodied chips may enter a new stage of intense competition.
Conclusion
The significance of Unitree's IPO review goes far beyond the possible emergence of a star robotics company in the A - share market. It marks that the embodied intelligence industry chain has begun to be systematically examined and priced by the capital market. For the semiconductor industry, what really deserves attention is not whether robots can dance, but the emerging set of brand - new chip demand structures behind them. This structure will profoundly affect the entire semiconductor chain from IP, design, manufacturing to packaging and testing.
The policy level also provides certainty. Li Chao, the deputy director of the Policy Research Office of the National Development and Reform Commission, clearly stated at a press conference on May 22 that in the next step, the construction of key infrastructure for embodied intelligence will be used as a starting point to accelerate the construction of training infrastructure for embodied intelligence, better support the training of "brain and cerebellum" models, and at the same time, accelerate the construction of pilot bases, improve the software and hardware ecosystem, and accelerate technological innovation for application implementation. The national will to make robots "enter factories, shopping malls, and households" is laying the most solid foundation for the rise of this new chip terminal.
Whether robots can become the next super terminal in the semiconductor industry, the answer may already be written at the intersection of the industry and policy.
This article is from the WeChat official account “Semiconductor Industry Insights” (ID: ICViews). Author: Liu Qian. Republished by 36Kr with permission.