The Zhitong Finance App learned that Guoxin Securities released a research report stating that there are several major technology trends in HDD, SSD, NAND, DRAM, and HBM in the storage market, and there are many major players. AI big model reasoning is driving rapid growth in storage demand. According to the bank's calculation results, AI inference's demand for DRAM and NAND will be 23.0 EB and 593.5 EB respectively in 2026. Short-term supply is in short supply. Storage prices are expected to continue to rise, and the storage cycle is on the rise.
Guoxin Securities's main views are as follows:
storage system
System composition and classification. 1) System composition: AI storage systems are mainly divided into network-side storage and local storage. Among them, network-side storage mainly stores cold data, mainly HDD and SSD; local storage mainly stores heat data and temperature data, mainly HBM, DRAM, and local SSD storage. 2) Storage classification: Storage can be mainly divided into “volatile” memory and “non-volatile” storage. Among them, memory mainly includes DRAM and HBM, and its advantages are high speed; storage mainly includes SSD and HDD, which have the advantages of large capacity and low cost.
Market and technology trends
HDD, SSD, NAND, DRAM, HBM 1) HDD: HDD is a hard drive, which mainly increases HDD capacity by increasing surface density. Recently, HAMR (heat-assisted magnetic recording) technology can greatly increase single disk capacity. The market pattern shows dual oligopolies, with Seagate Technology and Western Digital as the main participants; 2) SSD: SSD is a solid state drive, a storage device using NAND Flash as the medium. The number of NAND stacks continues to grow, and the unit architecture gradually changes from SLC to MLC and QLC. The main market participants are Samsung, Hynix (including Solidigm), Flash, and Medi., armor Xia; 3) DRAM: DRAM has advantages such as low power consumption, high integration, and low cost. It is gradually being iterated from DRAM to SDRAM and DDR. Currently, leading manufacturers have begun DDR6 research and development. The downstream demand is mainly in the fields of mobile phones, PCs, and servers. The main participants include Samsung, Hynix, and Micron; 4) HBM: multi-layer DRAM chip stacking to achieve vertical interconnection through TSV, which in turn has higher storage density and greater bandwidth. Currently, it is mainly used in the AI field, mainly for Hynix, Micron; 5) NAND: The number of bits expressed in storage units continues to grow, and the number of stacked layers continues to grow. Downstream demand is mainly for SSDs and mobile phones. The main participants are Samsung, Hynix, Kioxia, Micron, and SanDisk.
Demand calculation
AI training and inference are driving growth in storage demand. AI big model reasoning is driving rapid growth in storage demand. According to the bank's calculation results, AI inference's demand for DRAM and NAND will be 23.0 EB and 593.5 EB respectively in 2026. Short-term supply is in short supply. Storage prices are expected to continue to rise, and the storage cycle is on the rise.
Company sorting
The business focus of global storage companies: Samsung Electronics and Hynix all have a high market share in DRAM, HBM, NAND, SSD, etc.; followed by Micron, which has a comprehensive product matrix, but the market share is slightly lower than Samsung and Hynix; SanDisk and Kixia focus on the NAND and SSD fields, and Western Digital and Seagate Technology focus on the HDD field.
Risk warning: The expansion of DRAM and NAND production by manufacturers will cause the risk of falling product prices; the risk that the capital expenditure of major Internet companies falls short of expectations; the risk that the number of active users of AI applications will not grow as much as expected; optimization of AI model solutions to further reduce the risk of storage demand, etc.