Shenwan Hongyuan Securities·2025-11-22 12:43Group 1 - Google's release of Gemini 3 Pro and Nano Banana Pro significantly enhances multi-modal understanding and production capabilities, moving beyond simple image generation to more complex outputs [5][7][20] - The Chinese AI industry is expected to evolve from a competitive landscape to a more structured development path by 2026, with opportunities in computing power, models, and applications [5][6][20] - Huawei's Flex:ai technology represents a breakthrough in AI infrastructure, improving computing resource utilization by 30% through advanced scheduling and management of heterogeneous computing resources [5][6][76] Group 2 - The performance gap between Chinese and American large models is narrowing, with domestic models like DeepSeek and Qwen3 showing competitive capabilities in language and reasoning tasks [35][38][42] - The trend of "super nodes" in computing power is becoming clearer, with significant advancements in domestic AI chip performance and architecture, enhancing the overall competitiveness of Chinese solutions [20][25][33] - The software industry in China is entering a prime period for AI application, leveraging industry-specific know-how that large models cannot fully replace, thus creating a unique competitive advantage [63][66][70] Group 3 - The introduction of mid-training in model development signifies a shift towards a more refined and systematic approach, enhancing the capabilities of large models through targeted training [56][60] - The emergence of physical AI, which combines physical laws with data-driven decision-making, is expected to revolutionize various industries, particularly in areas like autonomous driving and digital twins [51][52] - Huawei's Flex:ai is positioned to compete with NVIDIA's Run:ai, offering a more versatile solution for managing diverse AI workloads across different hardware platforms [79][81]