Overview
Per Shenzhen Special Zone Daily on October 20, D-Robotics (Di Gua Robots) drew attention with its flagship Sunrise S600 chip: a palm-sized module packing 560 TOPS of compute that fuses large-model inference, multimodal perception and real-time motion control—providing the substrate for embodied-AI robots to perform complex tasks continuously. The company positions itself as the upstream compute-platform supplier of the embodied-AI chain: it builds no robots itself, but aims for “all robots to grow out of our products.”
The Sunrise S600 is purpose-built for humanoids, carrying the core Brain Processing Unit (BPU) that accelerates neural-network operators with stability and precision, supporting larger models and smoother VLA (vision-language-action) execution. As humanoid makers accelerate mass production this year, D-Robotics’ orders grew “exponentially”: it now works with 20+ Chinese embodied-AI firms and explores production with UBTECH, Zhongqing and Leju; its chips ship to over 60 countries and supply Sweden’s Husqvarna.
Background & Interpretation
1. Technical Background and Evolution
Embodied-AI competition is shifting from “body showmanship” to “brain and compute” positioning. Past robots leaned on cloud inference or general GPUs, where latency and power hurt real-time control; on-device inference chips close the perceive-decide-act loop locally. The clustered debut of Sunrise S600, T-Head E908A and Everest 200 signals the dawn of self-sufficient embodied compute.
2. Core Drivers and Underlying Mechanisms
The mechanism is “software-hardware co-design”: the BPU deeply optimizes neural operators, and with compilers, frameworks and toolchains, large models are pruned and quantized onto the edge while preserving precision and cutting latency/power. D-Robotics’ cluster of 60+ customers within a 5 km radius—“idea to prototype in days”—is its moat for rapid scenario iteration; real-world data refines models.
3. Market Structure and Industrial Chain Impact
For body makers, upstream platforms lower the chip self-development barrier and speed mass production; for Nvidia and peers, China’s on-device robot chips add substitution pressure. The pattern may echo autos—most bodies adopt public compute platforms, with full-stack self-design emerging only in mature scenarios. Developers should watch toolchain adaptation and open-model edge deployment.
Implications & Outlook
Short-to-medium term, humanoid mass production will ignite on-device embodied-compute demand, with 560-TOPS-class chips leading volume. Watch VLA edge-inference frame-rate/precision limits, domestic-chip yield and cost, and how fast non-Shenzhen supply chains replicate.
Takeaways for Industry Participants
Robot-body firms can use mature compute platforms to scale fast and focus on scenarios rather than heavy chip R&D. Chip and AI firms should polish co-design and developer toolchains, not just flops. Investors may find upstream “pick-and-shovel” players (compute, sensors, actuators) realize gains before bodies. Integrators must rebuild adaptation layers for multiple architectures.
This column compiles industry information and shares technical perspectives; it does not constitute investment advice.