Semis are already in a regime where every headline this month has been read bearishly

Semis are already in a regime where every headline this month has been read bearishly regardless of direction, which is the real setup to understand before this week.

The SOX round-tripped from a late-June record high to a confirmed bear-market drawdown (>20% off the peak) by mid-July — a valuation unwind after a roughly 65% six-month run, not a fundamentals break: Micron’s $41.46B quarter at 84.9% gross margin, HBM booked through FY26 across 16 long-term contracts, and SanDisk’s 87% EPS beat all landed fine and got sold anyway. It then tried to bounce — three straight up days July 20-21 (+1.9%, +3.7%, +5.2%) — got rejected near 580 resistance, and broke back down July 23-24: Tesla’s earnings disaster hit sentiment broadly, then Thursday’s session saw the SOX drop 4.25% with SK Hynix -8.8%, Intel -7.9%, Micron -7.0%, while Apple (capital-light, no data-center capex) rallied 3.5% the same day. The weekly technical read as of Friday called the bounce a likely dead cat, with 540 as the next support test and 420 below that if it fails.

The mechanism that matters for this week specifically: capex guidance has been punished in both directions this earnings season. Alphabet raised its 2026 guide by $15B to $205B on July 22 and the stock fell 6% the next day, dragging the Nasdaq down over 2%. That’s the trap — a raise reads as “spending without proof of payback,” but a hold or deceleration would read as “the supercycle is topping.” Either signal from Meta or Microsoft Wednesday evening can hit chip names through the same channel: Meta’s capex commentary (does the $145B ceiling move, or does management talk about absorbing existing capacity) reads straight through to HBM/DRAM demand — Micron, SanDisk, SK Hynix. Microsoft’s Azure print against the 39-40% bar, plus any signal toward BNP’s $262B FY27 capex estimate, reads through to equipment and foundry names — ASML, Applied Materials, Lam, KLA — and to Marvell/Broadcom on custom silicon. Amazon reports Thursday alongside Apple, and AWS capex is the other swing factor those same equipment names are exposed to.

There’s also a separate thread sitting under the earnings calendar that doesn’t resolve with any single print: Chinese model competition, specifically Moonshot’s Kimi K3, prompted Deutsche Bank on July 21 to explicitly question whether the capex trajectory is justified if similar performance is achievable more cheaply. That’s an independent overhang on the whole complex, and it’s part of why Michael Burry has been adding to his Nvidia puts and SMH short rather than covering.

The honest answer is that the risk this week is close to symmetric rather than directionally obvious — I can lay out the mechanism but a confident call either way would be overconfident given how the last three weeks have gone. The two dates that settle it are Wednesday evening and Thursday, and 580 on the upside / 540 then 420 on the downside are the levels the price action has already flagged as the ones to watch.

One thing worth flagging given SOXL specifically: it’s a daily-reset 3x product, so chop like the last three weeks — sharp down, sharp bounce, sharp down again — erodes value through volatility decay even in stretches where the underlying index ends up roughly flat. That’s a cost that shows up regardless of which way this week’s prints break, and it’s worth weighing if the position is meant to be held through both earnings days rather than traded around them.