In the first half of 2026, Rockchip delivered impressive interim results. Behind this robust performance, robotics has emerged as the company’s second core growth curve following intelligent cockpits. The globally viral Microduck robotic duck adopts Rockchip’s RK3566 as its main control chip, achieving peak sales of one unit every 4 seconds upon launch. This success demonstrates the tremendous potential of edge-side SoCs in lightweight embodied intelligence hardware.
Objectively speaking, Rockchip does not separately disclose robotics segment revenue in its financial reports. Institutional estimates put the segment’s current revenue at less than 5% of total revenue; the business is still in the early scaling phase focused on ecosystem penetration and customer acquisition. At the current industrial inflection point, Rockchip’s robotics division has built a complete product portfolio and software ecosystem, supporting ample long-term growth potential. Nevertheless, it faces multiple headwinds including the industry’s commercialization cycle, international competitors and rapid technological iteration.
Product Layout
Rockchip has established a dual-track robotics computing platform covering low-to-high tiers: **main SoCs plus AI co-processors**, compatible with the full spectrum of robotic use cases. Its main control chip lineup includes mature SoCs such as RK3588, RK3576, RK3566, RK3568 and RK3562, acting as the robot’s “cerebellum” to handle real-time tasks including motion control, environmental perception, visual SLAM and path planning. The RK182X series AI co-processors serve as the robot’s “brain”, undertaking high-level AI workloads such as local inference of multimodal large models, natural language interaction and scene understanding. The synergy between the two enables closed-loop edge-side perception, decision-making and execution.
This solution has been deployed across four major robotics application categories:
1. Industrial intelligent robots: factory inspection, material handling and equipment operation;
2. Agricultural robots: sowing, harvesting and field inspection;
3. Service robots: commercial cleaning, indoor delivery and educational robots;
4. Bionic companion robots: small biped robots like the Microduck robotic duck and quadruped bionic robots.
Leading domestic robotics players including Unitree, Ubtech and Zhiyuan Robotics have deployed Rockchip RK3588 and other chip solutions in their quadruped and humanoid robot prototypes and mass-production models. Commercial cleaning robot brands equipped with the RK1828 co-processor have achieved mass rollout, enabling on-device stain recognition and autonomous path planning. Within global open-source hardware communities, RK3566 and RK3588 are seeing rapid penetration, with numerous overseas developers and open-source robotics projects adopting Rockchip platforms. The viral success of Microduck marks a milestone for overseas ecosystem expansion.
Ecosystem as Core Moat
Hardware chips form only the foundation; commercialization of embodied intelligence robots is heavily reliant on development toolchains. Rockchip’s self-developed RKNN SDK is compatible with mainstream AI frameworks such as PyTorch and TensorFlow, open to global users and serving more than 5,000 hardware manufacturers. Once customers complete algorithm migration and product development based on RKNN, switching to alternative chip platforms requires 12–18 months of re-adaptation. This high migration cost creates strong customer stickiness. Meanwhile, Rockchip continuously provides robotics reference designs, underlying BSP firmware and sensor adaptation packages, substantially lowering hardware development barriers for robotics startups and open-source developers. The Microduck project selected RK3566 precisely for its mature, lightweight and cost-effective integrated hardware-software stack. Without expensive high-end computing chips, the platform can support sophisticated motions including bipedal gait control, object grasping and fall self-recovery, proving mid-range SoCs are fully capable of powering consumer-grade embodied intelligence hardware.
Industry Megatrend
Embodied intelligence is transitioning from laboratory demos toward industrialization, unlocking long-term growth for Rockchip. Historically, the robotics sector was dominated by imported high-end computing platforms with high costs and steep development barriers, restricting the adoption of consumer-grade lightweight robots. A clear industry divergence is emerging: high-end humanoid robots pursue extreme computing power for complex industrial operations, while a large number of compact, lightweight, consumer and research-focused robots prioritize cost performance, low power consumption and rapid development — Rockchip’s core sweet spot.
Institutions forecast sustained shipment growth for service robots, bionic robots and inspection robots over the coming years, alongside expanding demand for edge-side robotics computing power and rising requirements for supply chain autonomy. Opportunities for SoCs to replace overseas solutions keep expanding. Rockchip positions robotics as a strategic growth track and continues to invest in next-generation chip development. RK3668, RK3688 and other flagship SoCs in the RK36 series are under development, together with the new-generation RK186X co-processor scheduled for launch in Q4. The upgraded RK186X delivers higher computing capacity to support larger-parameter edge multimodal models, further lifting the computing ceiling for advanced robotic scenarios.
Key Challenges
Rockchip’s robotics business also faces non-negligible challenges.
First, industry commercialization remains at an early stage. Humanoid and bipedal robots are still far from large-scale household adoption. The robotics segment currently generates limited revenue and cannot deliver massive top-line contributions comparable to intelligent cockpits in the short run. While hit products such as Microduck attract tremendous market attention, their direct chip revenue contribution per unit is modest. Their greater value lies in brand exposure and ecosystem influence, drawing global developers to build robots on Rockchip platforms.
Second, competition is intense. Overseas leading vendors retain advantages in high-end robotics computing, while multiple domestic chipmakers have entered the embodied intelligence space, fueling ongoing price wars and ecosystem rivalry.
Third, robots impose continuously evolving technical requirements for real-time control, multi-sensor fusion and edge-cloud collaboration. Sustained investment in IP, algorithms and toolchain development translates into heavy R&D cost pressure.
Fourth, robotic requirements vary drastically across verticals. Industrial, household and educational robots have divergent demands for computing power, power consumption and reliability. Balancing standardized chip platforms with customized client needs remains a long-term challenge.
Conclusion
Overall, Rockchip’s robotics business features a sound growth logic and differentiated positioning. Instead of directly competing with overseas ultra-high-computing flagship chips, it focuses on mid-tier, lightweight and cost-effective edge AI platforms, targeting niche markets including open-source communities, education & research, commercial services and industrial inspection. The breakout success of Microduck is no accident; it results from Rockchip’s years of deep cultivation in AIoT and continuous global developer ecosystem building.
Going forward, with mass deployment of the RK182X co-processor, rollout of new-generation chip platforms, and expansion of the global embodied intelligence hardware market, the robotics business is poised to deliver steady growth and become a core engine helping Rockchip mitigate cyclicality in consumer electronics and unlock a second growth curve. It is critical to recognize the pace of the sector: robotics industrialization constitutes a long-term endeavor. Near-term progress will mainly manifest in ecosystem penetration and customer accumulation, with revenue realization advancing alongside the industry’s commercialization.
