摩根大通：存储芯片，“饥饿游戏”开启，一场为期四年的上行周期

Core Viewpoint - The "memory hunger" trend driven by AI computing is pushing the entire industry into a structural growth phase, leading to an unprecedented four-year pricing upcycle in the DRAM market from 2024 to 2027, with the global storage market expected to reach nearly $300 billion by 2027 [1][2]. Group 1: Market Dynamics - The core driver of the current cycle is the immense demand for high-performance memory due to AI computing, which has rapidly expanded from high-bandwidth memory (HBM) to traditional DRAM and NAND flash [2][3]. - Suppliers may struggle to meet the entire demand in the next 12 months, supporting sustained price increases [2]. - The NAND flash market is also experiencing a strong rebound, partly due to a severe shortage of hard disk drives (HDD), pushing customers towards enterprise solid-state drives (eSSD) [2][9]. Group 2: DRAM Market Outlook - The DRAM market is entering an unprecedented four-year pricing upcycle from 2024 to 2027, breaking the previous pattern of volatility [3]. - By 2027, HBM is expected to account for 43% of the total value in the DRAM market, stabilizing traditional DRAM price fluctuations and enhancing overall market profitability [3]. - Even with potential price drops for HBM3E products in 2026 due to increased supply, the next-generation HBM4 is anticipated to maintain a premium of about 35%, keeping the average selling price (ASP) stable [3]. Group 3: AI-Driven Demand - The rise of AI inference, AI services, and edge AI applications necessitates upgrades in memory configurations for general servers to achieve lower latency and better power efficiency [8]. - By 2027, AI-related applications are projected to account for 53% of the DRAM market's total addressable market (TAM) [8]. Group 4: NAND Market Recovery - The NAND flash market is experiencing a strong pricing recovery after two years of underinvestment, driven by a surge in demand for eSSD [9]. - The shortage of traditional HDDs, with delivery cycles extending up to 52 weeks, is prompting a significant shift towards eSSD in nearline storage markets [9]. - The structural importance of NAND is increasing as AI models transition from training to inference applications, raising demands for data read speeds and latency [10].