To The Moon

U.S. semiconductor and memory stocks posted notable losses on Monday.

SanDisk ‑6.5%, SK Hynix ‑4.9%, Micron ‑5.8%, Seagate ‑6.5%, Western Digital ‑5.2%. NVIDIA fell 2.9%, marking its seventh consecutive trading day of declines — its longest losing streak since 2022. Super Micro Computer and Intel both dropped over 3%.

Major semiconductor ETFs tracked the sector lower: $半导体指数ETF-HOLDRs(SMH)$ fell 2.4% and $iShares费城交易所半导体ETF(SOXX)$ declined 2.7%.

On the macro front, Monday’s backdrop was relatively supportive. Crude oil pulled back: NYMEX October WTI crude fell $2.05 (‑2.35%) to settle at $85.01 per barrel. October Brent crude dropped $2.22 (‑2.35%) to $92.17 per barrel. The 10‑year U.S. Treasury yield stood at roughly 4.70%, edging down from Friday’s level.

In theory, calmer macro conditions should offer tech stocks some relief. Instead, memory names got hammered and semiconductors weakened broadly. What drove this move? The sell‑off was largely sector‑specific rather than macro‑driven. Three key catalysts stand out:

1. Apple may source Chinese memory chips, pressuring US memory stocks

   On August 24, multiple foreign media outlets cited sources reporting the US government is considering approving Apple’s purchase of memory chips from Chinese manufacturers to ease its supply constraints. Apple has already tested DRAM and NAND chips from two Chinese memory firms. If approved, these chips would likely be used for Apple devices sold within the Chinese market. This marks a notable shift from previous hard‑line US stances.

   Investors worry that Chinese memory chips entering Apple’s supply chain would erode market share for Micron, SK Hynix and peers, lowering earnings expectations. Pre‑emptive risk‑off selling triggered a broad memory‑sector sell‑off.

2. The memory‑wall problem worsens; lagging HBM becomes an AI bottleneck

   At the Hot Chips 2026 conference at Stanford University on August 23, Micron researcher Sreeramaneni pointed out that the memory‑wall bottleneck is widening rather than easing. He noted AI compute power triples every two years while memory bandwidth grows by less than two‑fold. No matter how fast compute chips run, narrow data pipelines will ultimately throttle performance. This near‑term unresolved memory‑wall issue weighs on sector valuations.

3. Profit‑taking ahead of NVIDIA earnings

   With NVIDIA’s earnings report due this Wednesday, investors locked in profits across high‑valuation AI hardware. Fears that forward guidance could miss lofty consensus expectations triggered position unwinding in high‑beta segments such as memory and optical communications — typical risk‑reduction behavior ahead of big earnings prints.

Would you buy the dip in semiconductors, or stay on the sidelines? [Grin]

$英伟达(NVDA)$$英特尔(INTC)$$美国超微公司(AMD)$$闪迪(SNDK)$$美光科技(MU)$$SK海力士(SKHY)$$希捷科技(STX)$$西部数据(WDC)$