AI系列跟踪专题报告：国产算力高景气持续

Investment Rating - The industry investment rating is "Outperform the Market," indicating that the industry index is expected to perform better than the benchmark index over the next 6-12 months [11]. Core Insights - The report highlights the sustained high demand for domestic computing power driven by the ongoing restrictions on advanced chip imports from the US, which accelerates the domestic computing power substitution process. Domestic cloud vendors are increasing capital expenditures and gradually releasing industrial demand, while the iteration of domestic AI models and applications is boosting computing power demand [1][3]. Summary by Sections Investment Recommendations - It is recommended to prioritize attention on the construction and application of domestic AI computing power network infrastructure, including operators such as China Mobile, China Telecom, and China Unicom, as well as server and switch equipment manufacturers like ZTE, Unisoc, Inspur, Ruijie Networks, and Shengke Communication. Additionally, focus on optical modules and optical devices from companies like NewEase, Zhongji Xuchuang, Yuanjie Technology, Huagong Technology, Guangxun Technology, Shijia Photonics, and Huafeng Technology [3]. Industry Trends - The report notes that Huawei's Ascend 910C has begun mass shipments, marking a new phase in the commercialization of domestic computing power. A recent tender announcement indicated that a smart computing center project plans to use 4,500 Ascend 910C-2 servers, with an expected capacity of 20,000 P computing power. The Ascend 910C features a single-chip computing power of 320 TFLOPS (FP16), making it suitable for AI tasks such as natural language processing and computer vision [1][3]. - Domestic cloud vendors and operators are increasing capital expenditures on computing power, with Alibaba planning to invest 380 billion RMB in cloud construction and AI hardware infrastructure over the next three years, averaging over 120 billion RMB annually. Tencent's capital expenditure in Q1 2025 reached 27.5 billion RMB, a year-on-year increase of 91%, with a focus on resources for large model training and inference [1][3]. - The demand for computing power is expected to grow due to breakthroughs in application-side inference technology, which significantly lowers barriers to entry. The report cites Alphabet's inference volume reaching approximately 634 trillion tokens in Q1 2025, a 50-fold increase from the previous year [1][3].