SK Hynix Rises as Samsung’s HBM4 Narrows Pricing Edge

SK Hynix shares rose about 1.9% to 1.706 million won in Seoul on Tuesday after a month of losses. The stock had fallen roughly 9% over the past month, during which foreign investors sold about 5.98 trillion won of SK Hynix shares.

HBM export prices reached a record $76.13 per unit and conventional DRAM export prices rose 24.3% in July. While market prices are strong, analysts and investors are focusing on which suppliers capture the profits from rising memory demand.

SK Hynix established an early position supplying high-bandwidth memory (HBM) for AI accelerators. Samsung is accelerating its rollout of HBM4 and is becoming a wider alternative for customers. LS Securities estimates Samsung’s HBM4 share of its HBM shipments rose from about 5% in the first quarter to roughly 35% in the second, and that Samsung’s blended HBM yield improved by more than five percentage points.

Jung Woo-sung, an analyst at LS Securities, described the change as ‘does not mean growth in the HBM market is slowing,’ and called it a normalisation of competition among suppliers. Large cloud and AI customers typically qualify multiple suppliers to reduce supply risk and increase negotiating options.

LS Securities reduced its SK Hynix target price by 27.3% to 2.4 million won from 3.3 million won while keeping a buy rating. The broker also cut its forecast for HBM operating margins next year to about 60% from roughly 80% and noted that HBM represents more than twice as much of SK Hynix’s estimated intrinsic value as it does for Samsung under its model.

LS Securities warned that while supply should remain tight, rising memory costs are taking a larger share of Big Tech server spending and could limit how far customers accept higher prices. UBS projects SK Hynix will hold about 48% of HBM bit shipments this year but expects Samsung to have roughly 41% versus 39% for SK Hynix next year, indicating a more even split among suppliers.

Bank of America estimates global DRAM revenue growth could exceed 80% in 2027 if AI-related demand continues to expand and manufacturers restrain additions to conventional memory capacity.