国金证券：AI驱动新存储周期 需求跃升与供给瓶颈下的历史性机遇

Core Viewpoint - The storage market exhibits significant cyclical properties driven by emerging technologies, with the current phase marking the beginning of a new storage cycle, influenced by AI demand for storage chips [1][2]. Group 1: Semiconductor Cycles - Semiconductor cycles are categorized into long (8-10 years), medium (4-6 years), and short (3-5 quarters) cycles based on supply and demand dynamics, with demand, capacity, and inventory cycles interwoven [2]. - The storage segment is the second largest in semiconductors, showing greater volatility than the overall industry, with historical performance aligned with semiconductor cycles [2]. Group 2: AI Impact on Storage Demand - The introduction of large models and mechanisms is generating substantial data storage needs, with a shift from KB to TB and even EB in storage units as multimodal model penetration increases [3]. - The cost of inference for large language models has been decreasing exponentially since the release of ChatGPT3, which is expected to drive application growth and subsequently increase storage demand [3]. Group 3: KVCache and Storage Optimization - AI is redefining data storage, with KVCache emerging as a critical optimization mechanism for enhancing inference efficiency in large models, leading to increased storage demand [4]. Group 4: Capital Expenditure Trends - Capital expenditures for DRAM and NAND Flash manufacturers are shifting focus from pure capacity expansion to technology upgrades and high-value products, with projected DRAM capital expenditure reaching $53.7 billion in 2025 and $61.3 billion in 2026, a 14% increase [5]. - NAND Flash capital expenditure is expected to be $21.1 billion in 2025, with a slight increase to $22.2 billion in 2026, a 5% growth [5].