Analysis
A brutal overnight session in Seoul spread into a global chip-stock selloff this week, with more than $1 trillion wiped from chip-related names as the rout jumped from Asia to Europe to the US. The Kospi slumped 10.8% as Samsung Electronics and SK Hynix fell 13.4% and 14.7% respectively, driven in part by a wave of retail-leverage positions unwinding at once. By the time US markets opened, Micron and AMD were down 10-11%, Intel had fallen roughly 8%, and even Nvidia -- usually the most resilient name in the group -- slipped about 1%.
Three factors are doing most of the explanatory work, according to analysts covering the move. First, reports of a breakthrough in Chinese chipmaking tools have raised fears that Beijing is closing the technology gap faster than expected, threatening the pricing power Western and Korean memory makers have enjoyed during the AI-driven demand surge. Second, broker research warning that memory-chip prices are now more likely to peak in 2027 than continue climbing gave momentum traders a reason to take profits after a year of outsized gains. Third, leveraged chip-sector ETFs -- vehicles that amplify daily moves in either direction -- appear to have magnified both the initial decline and its acceleration through the session.
The selloff is notable for how quickly it moved beyond memory-specific names. Micron and SK Hynix are pure memory plays and an obvious read-through, but AMD and Intel's declines show the market treating this as a broader AI-hardware repricing rather than an isolated memory story. That distinction matters: memory chips are commodity inputs whose pricing cycles are somewhat separable from AI logic demand, but a selloff that drags AMD and Intel down in lockstep suggests investors are re-rating AI-hardware capital intensity broadly, not just picking on one sub-sector.
“Three factors are doing most of the explanatory work, according to analysts covering the move.”
The timing compounds the pressure on this week's Big Tech earnings. Microsoft and Meta report Wednesday, Apple and Amazon on Thursday, and all four companies' AI infrastructure spending assumes continued favorable chip economics. A memory-price downgrade to 2027 changes the input-cost assumptions baked into every hyperscaler capex model, right as investors are already scrutinizing that spending more closely following Apple's capital-discipline-driven overtake of Nvidia for the world's-most-valuable-company title.
For venture investors, the read-through runs directly into portfolio company cost structures: any startup whose unit economics assume falling GPU and memory costs through 2027 just got a real-time reminder that chip pricing is cyclical and politically contested, not a smooth downward curve. Inference-cost-reduction startups -- the Etched and Multiverse Computing types betting on cheaper, faster AI compute -- may find their pitch more urgent, not less, if memory costs stop cooperating.
What to watch: whether the selloff stabilizes before Wednesday's earnings or bleeds into the broader Nasdaq, any official Chinese confirmation or denial of the chipmaking-tool breakthrough driving the fear, and whether Micron's own upcoming earnings commentary validates or contradicts the 2027 price-peak thesis.
