Samsung Electronics and SK Hynix rose in Seoul on Tuesday, resisting the semiconductor selloff that swept Wall Street overnight as investors reassessed whether calls to slow artificial intelligence development imply weaker chip demand.
Samsung gained 0.7% and SK Hynix advanced 1.82% by 11:20 a.m., while the KOSPI was little changed.
The moves followed a 5% slide in the PHLX Semiconductor Index on Monday, when AI-linked chipmakers were hit by fears that a slower pace of frontier-model development could curb infrastructure spending.
The divergence is less dramatic than it appears, as Korean memory stocks had already absorbed much of the shock a session earlier.
Korea priced the AI scare first
Samsung fell 4.05% on Monday and SK Hynix dropped 6.35%, helping pull the KOSPI down 3.26%.
Seoul therefore reacted to the AI slowdown debate before Wall Street delivered its own semiconductor selloff.
Tuesday’s rebound suggests investors were unwilling to sell the same concern twice without evidence that orders, server deployments or memory consumption are weakening.
However, it is not a broad vote of confidence, as individual investors sold more than 13 trillion won of Samsung and SK Hynix shares between September 1 and 11, while foreign investors also reduced exposure.
Corporate treasury-share purchases absorbed much of that selling.
Memory shortages complicate the bearish case
The simplest bearish chain is that slower AI development leads to lower capital expenditure, fewer accelerators and ultimately less memory.
But current supply conditions make that relationship less straightforward.
Ben Barringer, global head of technology research at Quilter Cheviot, told CNBC that inference capacity remains constrained even if training and model rollouts slow.
“Demand still far outstrips supply,” he said, arguing that a modest slowdown may not materially damage company revenues.
TrendForce reported this month that DRAM industry revenue jumped 59.5% quarter-on-quarter in the second quarter to nearly $154.73 billion.
It said AI servers are driving demand for HBM and high-capacity server memory while supplier inventories remain historically low.
The research firm also expects server DRAM shortages and elevated pricing to extend into 2027 as capacity shifts towards HBM.
That scarcity is already affecting chip prices. Chinese AI-chip suppliers have raised prices by roughly 20% to 50% on some products as HBM shortages increase costs.
Korea is watching where AI profits move
The bigger Korean investment case is that AI spending may increasingly reward memory suppliers rather than only GPU designers.
“Next year, the center of profit will shift from graphics processing units to memory,” Kim Dong-won, head of research at KB Securities, told the Seoul Economic Daily.
That view helps explain why Samsung and SK Hynix can trade differently from Nvidia and other US semiconductor names.
Their earnings are increasingly leveraged to memory pricing, HBM scarcity and server demand rather than solely to the speed of the next frontier-model release.
The risk is that this distinction eventually disappears.
Christopher Wood, global head of equity strategy at Jefferies, told the Korea Times that the biggest danger is markets deciding companies will not be able to “monetise this capex.”
If hyperscalers conclude their huge AI investments cannot generate adequate returns, spending cuts would eventually hit GPUs, servers and memory alike.
The post Samsung and SK Hynix defy a 6% US chip rout: what is Korea seeing differently appeared first on Invezz
