Core Insights - Huawei has introduced a new computing solution based on China's available chip manufacturing technology to address the significant demand for AI computing power in the country [1] - The rapid growth of AI applications has led to an explosive increase in computing power demand, with a notable gap in high-end computing resources in China [1][2] - Huawei's strategy emphasizes system capabilities over single-chip advantages, proposing a "super node + cluster" approach to enhance computing power through resource collaboration [2] Group 1: Computing Power Demand - The demand for computing power is driven by the exponential growth of AI model parameters and the increasing requirements for real-time and stable applications [1] - China's AI industry is developing quickly, but there are constraints such as insufficient supply, high costs, and an underdeveloped ecosystem, leading to a computing power gap [1][2] Group 2: Innovative Solutions - Huawei's approach includes using multiple chips to create a powerful computing system, leveraging the concept that "many hands make light work" [1] - The "super node" consists of multiple cabinets and cards, functioning as a single computer for learning and reasoning, while clusters connect multiple super nodes to form a large-scale computing group [2] Group 3: Ecosystem Development - Huawei's open-source strategy aims to attract more developers to participate in ecosystem building, creating a positive feedback loop for product optimization [3] - By encouraging the development of complementary products based on Huawei's open protocols, a complete autonomous AI ecosystem can be gradually established [3] - The integration of architectural innovation and ecosystem construction positions China to potentially lead in the global AI computing power competition [3]

Investment Rating - The report maintains a "Recommended" rating for the industry [6] Core Insights - Huawei's recent announcements at the 2025 Full Connect Conference highlight the launch of the Ascend AI chip roadmap, indicating a strategic push to compete with NVIDIA in the high-end AI market [3][4] - The Ascend chip series will see annual upgrades, with significant performance enhancements in computing power, interconnect bandwidth, and memory capacity, positioning domestic AI computing chips among the world's best [4] - The emergence of supernodes as a new standard for AI infrastructure emphasizes the importance of internal interconnect capabilities, with Huawei's Atlas 950 SuperPoD expected to outperform NVIDIA's upcoming products in multiple performance metrics [5][7] Summary by Sections Chip Development - The Ascend 950 series will achieve a single-chip computing power of 1 PFLOPS (FP8) and 2 PFLOPS (FP4), with subsequent models doubling this performance [4] - The interconnect bandwidth of the 950 series will increase by 2.5 times compared to the current Ascend 910C, reaching 2 TB/s [4] Supernode Infrastructure - Supernodes are becoming the dominant product form in AI infrastructure, with Huawei's Atlas 950 SuperPoD supporting up to 15,488 Ascend cards [5] - The Atlas 950 Supernode's interconnect bandwidth is projected to reach 16.3 PB/s, significantly surpassing competitors [5][7] Investment Recommendations - The report suggests focusing on companies involved in chip production, supernode technology, and related infrastructure, including companies like SMIC, Tongfu Microelectronics, and others [7]