After an 18% summer selloff in chip stocks, Citi asks what comes next

Citi expects semiconductor shares to rebound after the Philadelphia Semiconductor Index fell 18% over the summer, citing resilient data-center demand and favorable commentary expected from AMD, Marvell, Lumentum, Astera Labs and semiconductor-equipment companies. Citi raised its CPU market forecast to $237 billion by 2030 from $29 billion in 2025, implying a 52% CAGR, and identifies AMD as the primary beneficiary, with Intel also positioned to gain. The bank also lifted its 2028 wafer-fab-equipment spending outlook to a $257 billion base case and nearly $300 billion bull case, versus a prior $250 billion forecast.

Analysis

The investable read-through is not a broad semiconductor beta call; it is a rotation from AI compute beneficiaries whose expectations remain largely accelerator-linked toward the networking, optical, CPU, and capital-equipment layers needed to scale inference and general-purpose data-center capacity. AMD has the cleanest upside-to-estimate pathway if enterprise and cloud CPU refresh demand improves, while MRVL, LITE, and ALAB can benefit from higher cluster interconnect intensity even if GPU unit growth normalizes. NVDA is less directly exposed to a CPU-led revision cycle and remains vulnerable to any evidence that hyperscalers are reallocating incremental capex from training clusters toward lower-cost inference architectures.

The near-term catalyst is management commentary and subsequent sell-side estimate revisions over the next 4-8 weeks; conference optimism alone is unlikely to sustain a rerating without backlog, customer-design-win, or gross-margin evidence. Over 6-18 months, higher wafer-fab-equipment spending would favor AMAT, LRCX, KLAC, and ASML, but equipment revenue recognition lags customer capex commitments by several quarters and should not be extrapolated from a top-down capex forecast alone. Intel is a less attractive expression of CPU recovery: any demand improvement competes with its execution risk, elevated fixed-cost base, and the possibility that price competition prevents operating leverage.

Consensus may be underestimating that weak client-device demand and memory-cost inflation can delay the volume recovery required to validate a broad CPU thesis. A sustained energy-price shock or renewed rate repricing would also pressure long-duration semiconductor multiples before fundamentals change. Falsification for the bullish rotation would be hyperscaler capex guidance shifting toward flat 2027 budgets, AMD failing to raise data-center revenue outlook, or optical/networking suppliers reporting inventory-driven rather than end-demand-driven orders.

Key Decisions for Investors

• Initiate a 3-month long AMD / short INTC pair, sized 1:1 beta-adjusted. The trade isolates CPU share and margin execution: target 15-20% relative upside if AMD raises data-center expectations, with exit if AMD data-center guidance is unchanged at the next earnings report or INTC demonstrates material gross-margin recovery.
• Accumulate a basket of MRVL, LITE, and ALAB on post-conference volatility rather than chase opening strength; use a 1-3 month horizon. These names offer higher sensitivity to data-center scale-out and connectivity content, but limit position size because order visibility can be distorted by customer qualification timing and inventory builds.
• Maintain underweight NVDA versus SOXX over the next 1-2 months rather than establish an outright short. The relative thesis works if capital-spending enthusiasm broadens beyond accelerators; cover the underweight if NVDA receives incremental hyperscaler order disclosures or raises forward supply commitments.
• Place an alert on AMAT, LRCX, KLAC, and ASML for confirmed hyperscaler-to-foundry capex conversion, including customer capacity orders and raised tool delivery schedules. Do not add equipment exposure solely on the revised industry forecast; a recommendation requires evidence that incremental spend is not merely advanced-node maintenance or double ordering.
• Treat any move in AMD above conference-driven levels without upward earnings revisions as a trim opportunity. The key risk/reward asymmetry is that valuation can re-expand immediately, while proof of a durable CPU demand inflection will take at least one to two reporting cycles.