Why The iShares Semiconductor ETF (SOXX) Plunged 21% in July

The iShares Semiconductor ETF (SOXX) fell 21.2% in July after a 112.8% surge in H1 2026, as AI-exposed chip positioning unwound. The selloff was intensified by Michael Burry’s bearish AI-valuation commentary, the July 16 release of Moonshot’s low-cost 2.7T open-weight model Kimi K3, and forced deleveraging from hedge fund Situational Awareness (assets reportedly ~$45B at end-June with up to ~400% leverage). Despite generally strong semiconductor earnings/outlooks, investors are split between a pullback in an ongoing AI demand cycle versus an earlier-stage “boom” turning into a deeper unwind.

Analysis

This looks more like a crowded-ownership unwind than a clean read-through on end-demand. When a sector has already re-rated aggressively, the first leg lower is usually driven by multiple compression and de-grossing, not by immediately visible revenue damage; that means the biggest near-term losers are the highest-beta, most index-owned names in the basket, especially NVDA and the ETF wrapper SOXX. The second-order risk is to suppliers and adjacent infrastructure names with the fattest positioning, while the actual fundamental damage may be smaller than the tape implies.

The contrarian point is that cheaper models are not automatically chip-negative. If AI inference gets cheaper, token consumption and application-layer experimentation usually expand faster than model vendor pricing compresses, which can still support accelerator and memory demand over a 1-3 month horizon. Where the market may be right is on duration: if cloud buyers see lower returns on incremental AI spend, capex growth can slow by 2027, but that requires evidence in commentary, not just a hot open-source release.

The cleanest watchpoint is whether the next round of earnings revisions or hyperscaler capex guides start to roll over; absent that, this is probably a positioning event with a high chance of mean reversion. The key falsifier for the bull case is a second leg down in SOXX that fails to recover on good semiconductor prints, which would suggest the market is finally discounting a real demand pause rather than a sentiment air pocket.

Key Decisions for Investors

• Tactically buy SOXX on continued weakness over the next 3-5 sessions, using a staged entry; target a 10-15% rebound over 1-3 months if forced selling exhausts, with a stop if the ETF loses another ~8% from here and fails to reclaim its prior support band.
• Prefer MU over NVDA for a short-dated rebound trade only if upcoming commentary confirms AI memory bit growth remains intact; use a 6-12 week horizon, because MU has more operating leverage to sustained AI demand while NVDA carries the larger positioning/multiple overhang. Falsify if memory pricing weakens or guidance softens.
• For a defined-risk hedge, buy NVDA put spreads into any bounce ahead of the next earnings window; this monetizes further multiple compression if the market keeps punishing crowded AI leadership, while capping premium outlay. Cover if NVDA regains its 50-day moving average on strong volume.
• No structural short yet in SOXX: wait for a second catalyst such as lower capex guides, softer backlog commentary, or broader credit spread widening before initiating. Otherwise, the better trade is to fade panic rather than press it.