Highlights

• WAIC 2026 was held in Shanghai under the theme "Intelligent Partners, Creating the Future Together". The event featured over 1,100 exhibitors, approximately 3,000 exhibits, more than 300 global product launches and over 300 robots, covering the full AI stack from chips, computing infrastructure and foundation models to agents, embodied AI and smart devices. This year's focus shifted towards full-stack AI deployment and real-world commercialisation. Maintain MARKET WEIGHT..
   

WHAT’S NEW

• We believe WAIC 2026 signals that China's AI industry is entering a new phase, with competition shifting beyond model scaling toward efficiency, AI agents, embodied AI and AI infrastructure. 
  
  a)    LLMs. Competition is moving from parameter scaling to memory efficiency, token utilisation and inference optimisation, with AI agents emerging as the primary commercialization vehicle for foundation models. 
  
  b)    Embodied AI. The industry is evolving from Vision-Language-Action (VLA) models toward World Models, enabling robots to better understand, predict and interact with physical environments. 
  
  c)    AI infrastructure. Competition is shifting from individual chips to training-inference integrated AI accelerators and super-node clusters. Scale-up networking remains the fastest-evolving AIDC technology, while CPO is a long-term opportunity despite yield and engineering challenges. 
  
  d)    Next AI frontier shifts toward AI-driven research and physical AI commercialisation. Discussions increasingly focused on AI evolving from a productivity tool to a research partner, capable of accelerating scientific discovery and potentially designing next-generation model architectures. At the same time, embodied AI is shifting its focus from demonstration prototypes to real-world commercial deployment, emphasising scalable industrial and service applications. 
  
  d)    Long-term bottlenecks. China's AI competitiveness will increasingly depend on advanced chips, compute infrastructure, proprietary data and ecosystem collaboration, with long-term value shifting beyond model performance.
   

• China's LLM competition has evolved from cost efficiency (DeepSeek) to frontier intelligence (Zhipu GLM) and now pricing power, led by Moonshot's Kimi K3. Launched on 17 July, the 2.8T-parameter Kimi K3 sets a new benchmark in coding and agentic AI, with pricing of US$2.3 per 1M blended tokens, above Qwen3.7 Max (US$1.4), GLM-5.2 (US$0.9), MiniMax M3 (US$0.22) and DeepSeek V4 Pro (US$0.18). Looking into 2H26, we expect competition to centre on coding, AI agents and proprietary data flywheels, with 2–5T-parameter model launches from Zhipu, Alibaba and MiniMax. While lower-end API pricing (US$0.1–0.2 per 1m tokens) is likely to remain under pressure, we believe pricing power, cost efficiency and financial strength will increasingly differentiate long-term winners. Beyond foundation models, agent platforms and video generation remain the most attractive monetisation opportunities.