The rising adoption of generative AI has significantly increased the demand for advanced DDR5 DRAM and HBM technologies in data centres, as well as the need for enterprise SSDs to support AI servers.
Additionally, the market has seen the introduction of the first smartphones and PCs equipped with on-device generative AI capabilities. These devices require substantial memory and storage to handle large language models, further driving bit demand in the mobile and consumer markets.
“In response to this new wave of AI-driven growth, major players such as Samsung, SK hynix, and Micron have shifted a greater portion of their wafer capacity toward meeting the soaring demand for HBM. This shift has led to a deceleration in overall bit production and a hastened transition toward an undersupply of non-HBM products. Production cuts made in late 2022 became effective only in Q3-2023, as OEM inventory levels began to tighten, heralding a rebound in the memory market,” said Simone Bertolazzi, Ph.D., Principal Analyst, Memory at Yole Group.
This surge in demand has accelerated the memory sector’s recovery, with revenue projections showing a sharp rise in the coming years. For 2024, revenues are expected to reach $98 billion for DRAM, an 88% year-over-year increase, and $68 billion for NAND, a 74% increase. These figures are projected to continue climbing, with expectations for 2025 reaching $137 billion for DRAM and $83 billion for NAND. Long-term forecasts up to 2029, driven by strong demand from data centres, suggest further growth, potentially reaching $134 billion for DRAM and $93 billion for NAND, with compound annual growth rates (CAGR) of 17% and 16%, respectively, from 2023 to 2029.
In this context, the market research and strategy consulting company, Yole Group, has released its annual market and technology report, ‘Status of the Memory Industry 2024’. This comprehensive study provides a broad overview of the semiconductor memory industry, presenting the latest technology trends, market analysis, and insights into the supply chain and memory ecosystem.
Bertolazzi said: “The escalating need for advanced computing chips designed for AI training and inference has intensified competition among top DRAM manufacturers such as Samsung, SK hynix, and Micron. These companies are actively expanding their capacities for cutting-edge HBM3/3E technologies and are swiftly advancing the development and validation of next-generation HBM3E and HBM4 products.”
HBM technology has become vital for AI systems and holds strategic importance in the geopolitical landscape of AI technology. Due to US export restrictions, China is driven to independently develop and produce its HBM technology to meet its high demand for high-performance AI memory. These restrictions have led to significant investments and initiatives within China to manufacture these essential chips. Reports indicate that China’s first HBM2 products are maturing, currently being evaluated by select customers, and used internally for developing server processing units. Despite lagging global memory technology leaders by over six years, Chinese companies are determined to remain active in key AI technologies and aim to capitalise on the extensive local market for lower to mid-range AI accelerators. Additionally, the number of suppliers producing DRAM modules and SSDs with China-manufactured DRAM and NAND is rapidly increasing, highlighting the urgent need for locally produced semiconductor memory devices.
