从资本开支规划与模型、应用迭代看算力需求

Investment Rating - The industry investment rating is "Positive (Maintain)" [5] Core Viewpoints - Despite a lack of major catalysts in the computing power sector recently, the iteration speed of AI models remains rapid, with multimodal and agent applications expected to consume more computing power, thereby sustaining the industry's prosperity [3] - Domestic policies favoring self-controlled computing power and data security requirements are leading to increased attention and procurement of domestic AI chips and integrated machines [3] Summary by Sections Capital Expenditure and Demand - Major internet companies and telecom operators are significantly increasing their capital expenditure (Capex) in computing power, with Tencent's 2024 Capex exceeding 76.7 billion yuan, a 221% year-on-year increase, and accounting for 12% of its revenue [8] - China Telecom plans to invest 21.96 billion yuan in cloud and computing power, a 22% increase year-on-year, while China Unicom plans to invest 18 billion yuan, up 28% [8] Model and Application Iteration - The launch of OpenAI's image generation application has led to overwhelming demand, causing the company to limit usage to alleviate GPU overload [8] - The complexity of applications, such as the agent application Manus, indicates a high computing power cost, with tasks consuming significant resources [8] Policy and Market Growth - Shanghai aims for its intelligent computing cloud industry to exceed 200 billion yuan by 2027, with a target of achieving 200 EFLOPS in computing power, where self-controlled computing power should account for over 70% [8] - Other cities like Shenzhen are also setting specific goals for the intelligent computing sector [8] Investment Recommendations - For domestic computing power, recommended stocks include Haiguang Information (688041, Buy), Cambrian-U (688256, Not Rated), and Zhongke Shuguang (603019, Buy) [3] - In the computing service and IDC sector, recommended stocks include Hainan Huatie (603300, Buy) and others [3]