KB Securities: "Memory Semiconductor Bottlenecks Expected to Continue"

The Game Is About to Change: Samsung Electronics and SK hynix Poised for Revaluation as "GPUs Are the Heart, Memory Is the Blood Vessel" [Weekend Money] View original image

On September 24, KB Securities stated that the core axis of artificial intelligence (AI) infrastructure investment is rapidly expanding from graphics processing units (GPUs) to central processing units (CPUs) and memory semiconductors. The firm analyzed that Samsung Electronics and SK hynix will be re-evaluated for their strategic value as essential assets in the AI infrastructure field, moving beyond their roles as simple memory suppliers.


Kim Dongwon, Head of Research at KB Securities, explained, "The recent surge in stock prices of Intel, AMD, and ARM on the New York Stock Exchange is due to growing expectations that the backbone of AI infrastructure investment is shifting beyond GPUs to include CPUs and memory." He added, "As the possibility of commercialization for AI continues to increase, with the proliferation of AI agent services such as those by Meta, hyperscalers are expected to maintain infrastructure investments for longer periods, which in turn will ensure sustained visibility for memory demand."


Kim further noted, "In particular, AI agents require much more frequent and continuous inferencing operations compared to existing generative AI." He diagnosed, "This is a factor structurally expanding memory usage across AI data centers, affecting not only CPUs and GPUs but also high-bandwidth memory (HBM), DDR5, and enterprise SSDs (eSSD)."


He continued, "The fact that AMD became the fourth U.S. semiconductor company, after Nvidia, Broadcom, and Micron, to surpass a market capitalization of USD 1 trillion indicates that the AI bottleneck is shifting to memory." He evaluated, "The increase in server CPUs will drive greater demand for high-capacity DDR5 server DRAM, while the rise in GPUs and AI accelerators will simultaneously boost HBM demand."


He also stated, "In the era of agentic AI, which requires high processing performance at the level of 1,000 tokens per second, the ability to supply as much data as quickly as possible is the key competitive edge." He explained, "To respond, Samsung Electronics is developing next-generation 'zHBM' that stacks memory on top of GPUs, with sample supply expected by the end of 2027."



Kim added, "If GPUs are the heart, CPUs are the brain, and power is the oxygen in an AI data center, then memory acts as the blood vessels that ensure the continuous circulation of data." He concluded, "With the heart of performance and efficiency moving to memory, a re-evaluation of Korea's leading semiconductor companies, Samsung Electronics and SK hynix, is anticipated."


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing