Chip stocks that notched record rallies in second quarter start Q3 with a dud

Chip stocks reversed sharply after a blockbuster Q2: Micron fell 11% and lost $138B in market value, while Intel (-9%) and AMD (-7%) dragged the VanEck Semiconductor ETF (SMH) down more than 5%. The selloff was tied to fears AI infrastructure supply may be catching up to demand after a report that Meta could rent out excess computing capacity, even as Meta shares gained over 9% following its lackluster Q2. Despite the drop, Micron’s latest results were strong, with revenue more than quadrupling and gross margin rising to 84.9% from 39% a year earlier.

Analysis

This looks less like a fundamentals reset and more like the market de-rating the "every dollar of AI capex is automatically accretive" assumption. The first-order winners remain the hyperscalers with the best balance of capex intensity and monetization leverage; if compute can be rented or reallocated, META can improve ROI faster than peers, while the marginal dollar of spend becomes more discretionary. That shifts relative performance away from equipment and toward platforms that can turn AI capacity into enterprise revenue without carrying the full depreciation burden.

The more vulnerable names are the semis whose valuation depends on a clean multi-quarter capex upcycle: AMAT and LRCX are exposed to order timing, not just end-demand, so any pause in fab tool bookings can hit estimates before it shows up in revenue. INTC and AMD are also at risk because the market’s "more CPUs/memory" thesis is now being challenged; if the incremental AI stack proves more GPU-centric and utilization-driven, their share of wallet can disappoint even if overall AI demand stays healthy. NVDA is comparatively insulated in the near term, but not because it is immune — because it remains the bottleneck supplier until utilization evidence clearly bends.

The contrarian view is that this is a positioning unwind, not a capex peak. Renting excess capacity can be read as disciplined monetization of idle compute, which actually supports hyperscaler returns and may extend, not shorten, the AI investment cycle. What would falsify the bearish read is a second-derivative slowdown in Meta/peer capex guidance, a fall in cloud instance pricing, or two straight quarters of downward order revisions from equipment vendors; absent that, the move may be too early to fade aggressively.

Key Decisions for Investors

• Pair trade: long META / short SMH for 1-3 months. Thesis is that hyperscalers with monetization flexibility outperform the infrastructure basket as AI ROI becomes the gating variable. Risk: renewed capex acceleration or stronger-than-expected semiconductor order data.
• Short AMAT or LRCX on any bounce into the next 2-6 weeks. These names are most sensitive to capex digestion and backlog multiple compression; stop if foundry/equipment commentary shows no order deceleration into the next earnings cycle.
• Reduce exposure to INTC and AMD relative to NVDA for the next quarter. If the market is shifting from "more CPUs" to "more utilization," these are the most exposed valuation narratives; reverse the view if CPU/memory ASPs and AI attach rates re-accelerate.
• Avoid chasing outright short NVDA. Relative to the rest of the group, it still has the best bottleneck economics; use it only as a hedge within a broader short basket, not as the primary bearish expression.