Chip stocks close out their best first-half and quarter on record. What investors should do now

The VanEck Semiconductor ETF (SMH) is up 75.5% in the first half of 2026 (65% in Q2), after a volatile week where it fell 7.3% as investors rotated out of chips. Analysts argue pullbacks look driven by end-of-quarter rebalancing rather than a fundamental break, pointing to strong AI-chip memory fundamentals and a Broadcom–Google TPU long-term agreement through 2031. Retail flows also remain supportive (Micron drew ~$30.8M in inflows on Monday; retail semiconductor inflows >$103M over two sessions), reinforcing the view that the next leg in semis may continue despite risks from potential component supply mismatch and macro/yield pressures.

Analysis

The opportunity set is now more about where AI economics accrue than about the direction of AI spending itself. Contracted/custom-silicon exposure and memory scarcity look like the cleanest ways to stay long the theme, while high-beta names without revenue visibility are increasingly just momentum vehicles. Second-order, tighter HBM and component supply raises build costs for hyperscalers and server OEMs, which can eventually slow unit growth unless they shift more spend toward efficiency-enhancing silicon and away from brute-force GPU expansion.

Near term, the pullback looks flow-driven rather than thesis-breaking, so the key signal is whether semis keep respecting their short-term trend support on weak tape. Over the next 1-3 months, earnings and capex commentary will matter more than price action: any hint of 2027 capex digestion or FCF pressure should hit the less defensible AI beneficiaries first. A sustained rise in yields is the main macro overhang because semis are priced like long-duration assets; that can compress multiples even if demand stays intact.

The contrarian point is that retail enthusiasm is probably concentrating into the most crowded expression of the trade rather than the best one. The best risk/reward may be in quality AI infrastructure with contractual visibility and in the platform owner that internalizes compute economics, not in the names most exposed to order timing and narrative momentum. If memory supply loosens earlier than expected or hyperscalers start cutting forward guidance, the whole group can re-rate quickly, but absent that, the current debate is more about rotation than end-demand failure.

Key Decisions for Investors

• Long AVGO / short MRVL as a 1-3 month relative-value pair: AVGO has the cleaner revenue visibility and better multiple support; MRVL is more exposed to sentiment swings if AI order growth pauses. Falsify if MRVL starts printing stronger backlog conversion or AVGO loses premium on guide.
• Buy GOOGL on sector weakness for a 6-12 month hold: internal TPU economics and supply assurance should improve AI compute efficiency and cloud margin durability. Risk is that regulatory overhang or a capex step-down overwhelms the strategic benefit.
• Do not chase MU after the retail flow spike; wait for a retracement to trend support before adding, or use a defined-risk call spread only if HBM pricing/tightness persists. This is the most crowded expression of the theme and is vulnerable to any inventory scare.
• Tactically underweight/short INTC into strength versus SMH for 1-3 months: the rally has more squeeze than visible earnings inflection, so it is the weakest risk-adjusted way to express a semi view. Cover if management delivers a credible margin and cash-flow reset.