Search published article titles and excerpts here, or choose a wider Google search.
Analysis
Goldman Sachs prime desk data show hedge funds net sold AI stocks for five straight sessions, reversing all net buying since 2025, while memory stocks lost over $700B. Yet Goldman kept Buy ratings on Samsung and SK hynix.
Analysis
The SOX fell 8 straight sessions and the KOSPI lost over 20%, yet TSMC raised its outlook and KOSPI forward EPS was revised up. Why prices and earnings diverged — and the five signals that would turn this credit repricing into a real downcycle.
Analysis
TSMC net profit rose 68.7% to a sixth straight record, yet the KOSPI fell 7.57% with a circuit breaker on October 15, 2026. The market now prices financing sustainability, not AI demand.
Analysis
The SOX fell 11.41% in a week after eight straight down days, yet SMH constituents' earnings roughly doubled in a year and the trailing P/E fell from 38x to 32x. TSMC and ASML already answered with record prints. Four reports — Lam on Oct 21 to Intel on Oct 29 — now settle the debate.
Analysis
Oct 17, 2026 channel checks report Broadcom, Google, Microsoft and Nvidia wins for Intel 14A, with Apple and others evaluating. None is Intel-confirmed. PowerDirect, hedge economics, and the Oct 29 test.
Analysis
Samsung will reportedly build NVHBM base dies on TSMC 3nm. But the requirement applies to all three memory makers, and turnkey survives in standard HBM. The real shift is who owns the standard.
Daily
Samsung reportedly adopting TSMC 3nm for Nvidia NVHBM base dies, TSMC's $46.7B quarter and 58% Q4 growth guidance, ASML's record quarter, and the earnings test ahead for chip stocks — 8 stories to read on October 19.
Analysis
The HBM race's real bottleneck is not bonding - it is test. Stack yield is a multiplication problem: 99% die yield becomes 88.6% at 12-high. How HBM4's 2,048-bit, 16-high era makes inspection time the constraint.
Analysis
The AI bottleneck moved from chips to electricity: 945 TWh of data centre demand by 2030, a 2,285 GW US interconnection queue, 144-week transformer lead times - and the Korean equipment makers collecting the premium.
Analysis
DeepX is integrating its DX-M1 NPU into Intel's OpenVINO as a selectable processor, with initial results due early next year. Talks with AMD are open but still at the direction-setting stage. The structural read: for edge NPUs, the gate is software ecosystem friction, not TOPS.
Analysis
ASML says even 30% capacity increases in both 2027 and 2028 will not meet customer demand. SK hynix pulled its Yongin cleanroom opening forward three months and Samsung accelerated P5 Fab 2. In this cycle, expansion schedules are decided by secured EUV slots, not announcements.
Analysis
A Stockholm data center will deploy 1,800 FuriosaAI RNGD chips in early 2027 and 8,800+ at full build-out — alongside some 7,000 GPUs, not instead of them. Why Europe picked a Korean inference chip comes down to 180W power economics.
VLSI Korea