Overview
On November 5, the 9th China International Import Expo opened in Shanghai. Xuhui district fielded 40 exhibitors, with the “Model Speed Space” AI cluster debuting in the AI experience zone — eight firms spanning AI chips, 3D generative models, smart hardware and industry agents. Meanwhile, Qualcomm used the Hongqiao Forum’s humanoid-robot and embodied-AI sub-forum to detail how its “compute continuum” empowers physical-AI scale-up.
Autonomous, controllable compute was a highlight: Kuaishou-spinoff Lingchuan unveiled the SL200 video-coding and AI-inference chip (three-time MSU commercial track champion); Tsinghua-born Xijian showed the world’s first successfully taped-out brain-inspired complementary vision chip “Tianmou”; and Goke planned the GK6950 edge-AI chip series for embodied-robot spatial perception.
Background & Interpretation
1. Technology Background and Evolution
AI competition is extending from cloud training to “edge-cloud” coordination and physical-world deployment. As the compute base, sovereignty and energy efficiency are the core of domestic substitution. The chips shown at CIIE — vision, video, brain-inspired and edge NPU — signal domestic AI silicon moving from point breakthroughs to scenario depth.
2. Core Drivers and Underlying Mechanism
Physical AI demands real-time perception, decisions and actuation under tight power, size and cost. Tianmou integrates “chip-protocol-model” for high-speed dynamic perception and fine recognition; GK6950 delivers 10 TOPS dense INT8 with mixed precision for real-time VLA; Qualcomm’s continuum covers everything from milliwatt wearables to robot “brains.”
3. Market Structure and Industrial Chain Impact
Domestic AI chips are moving from “usable” to “good,” building differentiated strength in video, brain-inspired vision and edge robotics. With advanced-node capacity tight at foundries like TSMC, firms with mature design and stable tape-out win terminal customers and accelerate import substitution.
Implications & Outlook
Near term, CIIE will catalyze supply-demand matching between domestic chips and robot/OEM makers. Medium term, edge and brain-inspired chips should scale in embodied AI, autonomous driving and industrial inspection. Key variables: advanced-node supply, on-device model compression, and domestic chips’ yield-cost curve in real scenarios.
Takeaways for Industry Participants
Chip firms should go deep in vertical scenarios, building “chip + protocol + model” full-stack moats. OEMs and robot makers can lock domestic compute supply via CIIE. Investors should watch hard-tech names with tape-out and mass-production capability.
This column compiles industry information and shares technical perspectives; it does not constitute investment advice.