再造一个HBM！

Core Insights - The storage market is experiencing a rare supply-demand imbalance, driven by increased capital expenditure from cloud giants and rising AI demand, leading to memory shortages and significant price increases for both DRAM and NAND flash [2][3] - High Bandwidth Flash (HBF) is emerging as a new technology concept that could elevate NAND flash to a core asset in the AI era, similar to how High Bandwidth Memory (HBM) has transformed DRAM [2][3][4] Group 1: HBF Technology Development - HBF is a NAND version of high bandwidth memory technology, connecting 16-layer NAND stacks to logic chips and then to GPUs via multi-channel high bandwidth connections [3][4] - The architecture of HBF aims to balance performance and cost, matching HBM's bandwidth while providing 8 to 16 times the storage capacity at a similar cost [4][5] - The HBF ecosystem is beginning to take shape, with various industry players showing interest and starting to develop this technology [4][5] Group 2: Industry Players and Strategies - SK Hynix has launched the "AIN Family" product lineup, which includes HBF technology, aiming to meet the rapidly growing demand for NAND storage products in the AI inference market [5][6] - SanDisk is aggressively pushing HBF technology, planning to provide samples by mid-2026 and formal products by early 2027, showcasing strong confidence in its technological prospects [10][12] - Kioxia is taking a pragmatic approach by exploring multiple technical paths for HBF, which allows it to leverage partnerships while developing differentiated solutions [13][15] Group 3: Competitive Landscape - Samsung is cautiously entering the HBF space, focusing on early concept designs without a clear timeline for product release, reflecting a more conservative strategy compared to its competitors [16][17] - The competition among SK Hynix, SanDisk, Kioxia, and Samsung will shape the emerging HBF market, with each company adopting different strategies to establish their positions [19] - The transition of NAND from a low-cost storage solution to a high-value core component is underway, driven by AI demands and the evolution of memory architecture [19]