数据中心供配电设备行业跟踪：韩国半导体出口同比大幅增长，DRAM价格持续上涨

Investment Rating - The report rates the industry as "Outperform" [2] Core Insights - The data center industry has become a core incremental application scenario for the power equipment sector, directly driving demand growth and technological iteration [5] - The capital expenditure of major cloud vendors is a key indicator of demand for power distribution equipment, with significant growth observed in both overseas and domestic markets [2][12] - The supply chain remains robust, with companies like NVIDIA and TSMC reporting strong revenue growth, indicating a healthy demand for GPUs and semiconductors [18][23] - AI applications are expanding, with an increase in model updates and API call volumes, which are crucial for assessing the sustainability of capital expenditure cycles in data centers [30][35] Summary by Sections 1. Industry Overview - The semiconductor export from South Korea has significantly increased, with DRAM prices continuing to rise [1] 2. Capital Expenditure Trends - Overseas cloud vendors' capital expenditure reached $99.617 billion in Q3 2025, a year-on-year increase of 80.39% [7] - Domestic cloud vendors like Alibaba and Tencent showed a slowdown in capital expenditure, with Alibaba spending 31.5 billion yuan, up 80.10% year-on-year but down 18.55% quarter-on-quarter [12] - The investment in data centers is expected to remain high, with major cloud vendors planning substantial capital expenditures for AI infrastructure [11][13] 3. Supply Chain Analysis - NVIDIA's revenue for Q3 2025 was $57.006 billion, with data center products contributing $51.215 billion, marking a historical peak [18] - TSMC reported a revenue of 335 billion New Taiwan dollars in December 2025, a year-on-year increase of 20.4% [23] - The price of DRAM surged from $27.14 to $71.25 within two months, reflecting strong server demand [26] 4. AI Application Development - The number of AI models is steadily increasing, with significant updates and a growing number of API calls, indicating a robust application landscape [30] - The price of tokens for high-performing AI models has decreased by over 50% following the release of new models [48] - The weekly token call volume for AI applications was 7.50 trillion, showing a slight decline of 1.96% [35]