Samsung, SK Hynix Resist U.S. Chip Rout

Samsung Electronics rose 0.7% and SK Hynix advanced 1.82% in Seoul on Tuesday, a partial recovery after a sharp semiconductor selloff in the U.S. that knocked the PHLX Semiconductor Index down about 5% on Monday amid concern over slower frontier-AI model development.

Both Korean memory firms had already taken heavy losses the previous session: Samsung fell 4.05% and SK Hynix dropped 6.35%, contributing to a 3.26% fall in the KOSPI, so Tuesday’s gains reflected some regained ground rather than a full reversal.

Traders said markets in Korea largely anticipated the Wall Street move and investors were reluctant to sell the same risks again without clear signs that orders, server deployments or memory consumption were weakening. Individual investors sold more than 13 trillion won of Samsung and SK Hynix shares between Sept. 1 and 11, while foreign investors trimmed exposure and corporate treasury-share purchases absorbed much of the selling pressure.

Analysts outline a direct downside scenario in which slower AI development reduces hyperscaler capital expenditure, leading to fewer accelerators and lower memory demand. At the same time, supply constraints are affecting market dynamics. Ben Barringer, global head of technology research at Quilter Cheviot, argued that available inference capacity remains limited and that demand continues to exceed supply, suggesting a modest slowdown in rollouts may not heavily damage revenues for memory suppliers.

Industry data reflect strong recent memory-market growth. Research firm TrendForce reported a 59.5% quarter-on-quarter jump in DRAM industry revenue in the second quarter to about $154.73 billion, driven by demand for high-bandwidth memory (HBM) and high-capacity server DRAM. TrendForce forecasts server DRAM shortages and elevated pricing into 2027 as production shifts to HBM. Some Chinese AI-chip suppliers have raised prices by roughly 20% to 50% on certain products amid HBM scarcity.

Kim Dong-won, head of research at KB Securities, predicted that next year the center of profit in AI infrastructure will move from graphics processing units to memory, a shift that helps explain why Korean memory stocks can diverge from U.S. GPU-focused names. For Samsung and SK Hynix, earnings are increasingly tied to memory pricing, HBM availability and server demand rather than to the timing of a single large model release.

Risk remains if hyperscalers reassess returns on their AI investments. Christopher Wood, global head of equity strategy at Jefferies, warned that if large AI capital outlays fail to generate adequate returns, spending cuts would eventually reduce demand across GPUs, servers and memory. Market participants said they will monitor order flows and server-level consumption for signs that infrastructure spending is starting to decline.