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, with the selloff driven by bearish AI-related positioning and forced liquidation. Short-seller Michael Burry increased his shorts, and sentiment worsened after China’s Moonshot released the 2.7T-parameter open-weights Kimi K3, raising fears of price competition for U.S. labs. The AI hedge fund Situational Awareness reportedly used up to ~400% leverage and faced margin calls, triggering forced selling that amplified the drawdown. Despite the selloff, the article notes recent semiconductor earnings/outlooks showed no weakness in AI chip demand.

Analysis

This looks more like a forced de-grossing episode than a clean reassessment of AI capex. When the entire basket is owned through similar factor exposures, the first leg lower typically overshoots the change in fundamentals by 2-4 weeks, and the highest-beta names get hit regardless of balance-sheet quality. That said, the levered unwind leaves a useful tell: if leadership quickly reasserts in NVDA while the ETF lags, the market is signaling dispersion, not a secular top.

The key second-order effect is on spending priority, not total spend. Cheaper model competition can pressure app-layer economics, but it can also increase inference usage and keep GPU demand intact; the real risk is that hyperscalers slow the cadence of incremental orders while they digest existing commitments. That makes MU more cyclical than NVDA over the next 1-3 months, while exchanges/market-structure names like NDAQ can see a temporary volume tailwind from the volatility.

Contrarian view: the market is conflating better software efficiency with lower hardware demand. Historically, cheaper compute expands TAM faster than it reduces unit demand, but leverage-driven liquidations can still push SOXX below fair value for several weeks. The thesis breaks if hyperscaler capex commentary softens or if NVDA/MU guide below the current demand trajectory; otherwise this is more likely a position-cleanup trade than a regime change over 6-18 months.

Key Decisions for Investors

• Go long NVDA on continued weakness or post-earnings confirmation over the next 1-3 months; use a tight risk limit keyed to any guide-down in next-quarter AI revenue or gross margin.
• Put on a beta-adjusted pair: long NVDA / short SOXX for 4-8 weeks to isolate quality leadership from ETF de-grossing; target ~1x-1.5x relative alpha if forced selling persists.
• Do not chase an outright short in SOXX after the selloff; only revisit on a failed relief rally if hyperscaler capex commentary turns weaker over the next earnings cycle.
• For a lower-conviction expression, sell out-of-the-money SOXX put spreads 1-2 months out only if implied vol stays elevated versus realized; the trade thesis is that the downside is increasingly flow-driven, not fundamental.