Samsung Expands AI Memory Push Beyond HBM

Shim Ho-jun, Corporate Vice President and Head of the Application Engineering Team at Samsung Electronics’ Memory Business, delivers a presentation at the OCP Global Summit 2026 in San Jose, California, on Oct. 12. (Courtesy of Samsung Electronics) Samsung Electronics is broadening its artificial intelligence (AI) memory strategy beyond high-bandwidth memory (HBM) to encompass memory expansion and data storage, as increasingly complex AI models drive demand for greater processing capacity and more efficient data management. Rather than focusing solely on memory speed, the company aims to improve efficiency across the entire AI computing system through an integrated approach to memory and storage. Shim Ho-jun, Corporate Vice President and Head of the Application Engineering Team at Samsung Electronics’ Memory Business, outlined the strategy during a keynote presentation at the Open Compute Project (OCP) Global Summit 2026 in San Jose, California, on Oct. 12 (local time). The strategy is built around three pillars: speed, memory expansion and persistence. On the speed front, Samsung is advancing its HBM portfolio alongside customized memory solutions designed to meet the evolving requirements of AI computing. Its next-generation technologies include HBM5, zHBM and processing-in-memory (PIM), which are intended to improve data throughput and computational efficiency. Samsung has also introduced HBM4E and low-power memory technologies as part of its broader AI infrastructure portfolio. Its PIM approach brings certain computing functions closer to or within memory, reducing the need to move data between processors and memory devices. Memory expansion represents the second pillar, with Compute Express Link (CXL) playing a central role. Samsung is using its CXL-based CMM-D memory module to expand available memory capacity and improve resource allocation across AI systems. At the summit, Samsung and AMD showcased their collaboration on CXL-based memory expansion, demonstrating how the technology can help address capacity and bandwidth constraints as AI workloads become more demanding. Samsung also highlighted memory pooling, which allows computing systems to draw on shared memory resources rather than relying exclusively on memory installed within individual servers. The third pillar, persistence, focuses on enterprise-grade solid-state drives (SSDs) and high-capacity NAND flash technology. These products are intended to support the reliable storage and management of large volumes of AI data while improving efficiency across the broader computing infrastructure. The company is also exploring how CXL-based memory expansion and high-density SSDs can work together to support more scalable AI inference, including the management and offloading of key-value (KV) cache data. To incorporate customer requirements early in product development, Samsung continues to validate memory and storage technologies in practical AI environments and expand collaboration with global technology partners. The strategy builds on Samsung’s substantial position in the global memory industry. According to market research figures cited in the industry, the company held a 39% share of the global DRAM market and a 29.3% share of the NAND flash market in the second quarter of 2026, ranking first in both segments. Its annual DRAM production capacity is estimated at approximately 8.175 million 300mm-equivalent wafers, exceeding that of competitors SK hynix and Micron. Samsung’s broader AI memory push comes amid a sharp recovery in semiconductor earnings. In preliminary results announced on Oct. 8, the company estimated third-quarter revenue of 195 trillion won and operating profit of 107.4 trillion won, becoming the first South Korean company to project quarterly operating profit above 100 trillion won. The estimated operating profit represents a 782.5% increase from a year earlier, with an operating margin of approximately 55.1%, underscoring the earnings momentum generated by strong demand for AI-related memory products.