Summary of Key Points
Leading domestic robotics companies (such as Zhiyuan, Teshi Zhihang, and Ubot) are transitioning from relying on NVIDIA chips to using domestically produced computing power chips. This shift is driven by the cost pressures associated with mass production of robots and the gradual maturity of domestic computing power solutions in edge-based applications. Robots no longer only require individual chips but also depend on a cohesive "cloud-edge-device" computing network for support. Although domestic chips face challenges in terms of ecosystem integration, the long-term prospects are optimistic due to the explosive growth of the robotics industry.
1. From NVIDIA Chips to Domestic Chips: A Requirement Driven by Mass Production
In the early stages of their development, most robotics companies used NVIDIA chips. However, as mass production becomes a critical phase, the situation has changed—foreign chips have become too expensive. Zhou Jian, the chairman of Ubot, stated explicitly, "The industry has suffered greatly from the dominance of foreign chips, which account for one-third of the total cost (BOM) of robots." To control costs, domestic chips offer a more cost-effective alternative. For example, Zhiyuan is using domestic chips in areas such as cerebellar control and joint microcontrollers (MCUs). Its subsidiary, Lingqiao Shou, has chosen domestic MCUs for its dexterous hands to reduce costs without compromising on performance.
2. Multiple Robotics Companies Exploring Domestic Chips with Various Applications
- Zhiyuan: Collaborating with several domestic computing power companies, Zhiyuan is using chips for cerebellar control, joint MCUs, and communication components. Its subsidiary Lingqiao Shou uses domestic general-purpose MCUs in its dexterous hands and is also working on customizations.
- Teshi Zhihang: The new generation of wheeled dual-arm robot A3 will be equipped with a domestic edge-side chip with 560 TOPS (trillions of operations per second) of computing power, capable of handling both large model inference and real-time control requirements for industrial and automotive mass production.
- Ubot: Ubot has formed a joint venture with Muxi Inc. to develop intelligent chips specifically for end-use applications, breaking down barriers between domestic computing power and actual use cases.
- HaiGuang Information: Its chips are used in intelligent inspection robots that can navigate autonomously in data centers and issue alerts for anomalies, already being implemented in real-world scenarios.
3. Robots Need More Than Just Chips: A Comprehensive "Cloud-Edge-Device" Computing Network
Robots are not isolated devices; they require continuous perception, reasoning, and decision-making, which all rely on a robust computing network:
- Cloud: Similar to an oil refinery, it handles large-scale complex computations (such as training large models).
- Edge: Like a gas station, it provides low-latency processing close to the end-users (for real-time robot responses).
- Device: The robot itself uses chips for local control.
This computing network is becoming increasingly domesticized: domestic servers (from companies like HaiGuang and Kunpeng) account for over 90% of purchases by Unicom. SenseTime has launched the "Galaxy Plan" to build domestic intelligent computing clusters, while Tencent Cloud aims to deploy a large amount of domestic computing power and has invested in local chip manufacturers. Yuntian Changxiang already provides domestic computing services for robots at the edge level. However, there is still room for improvement in the efficiency of domestic chips for brain-related tasks, although they perform well in motion control.
4. Challenges Faced by Domestic Chips
- Ecosystem Integration: Switching to domestic chips requires developers to relearn toolchains, and migrating algorithms can be resource-intensive (especially for teams without prior edge-side optimization experience).
- Limited Early Orders: Only 20,000 humanoid robots were shipped globally last year, resulting in limited orders for chip manufacturers during the initial mass production phase.
5. A Promising Future for Domestic Chips
- Industry Potential: The robotics industry is still in its early stages but has great potential to become a major sector (with more opportunities than autonomous driving).
- Accelerated Progress through Collaboration: Partnerships, such as that between Muxi and Ubot, can provide valuable data and market insights to accelerate chip development.
- Historical Lessons: China's industries in automotive and photovoltaics have gone from being followers to leaders; the same can happen with chips. As Wang Hao, the CEO of a chip company, said, "In the Chinese market, it's hard to identify insurmountable barriers."
6. The Opportunity Window for Domestic Computing Power in the Age of Intelligent Agents
Robots are set to become data hubs, generating data with every interaction and perception, which requires computing power. As AI models become more compact and algorithms improve, computing power is moving from the cloud to the edge. 2026 could mark the beginning of a new era for edge-based computing. Domestic computing power can meet the cost challenges of mass production while optimizing products through practical use cases. Over the next five years, it will not only replace NVIDIA chips but also reshape the entire AI hardware ecosystem. With some time, domestic computing power has the potential to secure a significant position in the robotics industry.