BofA raises US semiconductor industry outlook to 18% annual growth

Bank of America raised its 2026-2030 U.S. semiconductor-industry growth forecast to an 18% CAGR from 14%, projecting a $3.2 trillion total addressable market by 2030 versus $2.7 trillion previously. It increased wafer-fab-equipment forecasts to $156 billion in 2026 and $210 billion in 2027, driven primarily by DRAM and AI/data-center demand. Continued customer orders, stable DRAM and NAND pricing, and Nvidia B200 GPU rental rates only 10% below March highs support the bullish outlook, while the Philadelphia Semiconductor Index is already up 67% year to date.

Analysis

The incremental earnings leverage is likely greater in the memory-capex chain than in NVDA, whose valuation already embeds sustained accelerator demand. Higher DRAM intensity per AI server expands the bottleneck from compute to high-bandwidth memory and advanced packaging: SK Hynix, Micron (MU), Samsung Electronics, ASMPT, and Advantest should see more direct estimate upside if capacity commitments convert into tool orders. Equipment beneficiaries are not homogeneous; Lam Research (LRCX) and KLA (KLAC) have stronger memory-process exposure than ASML, while Applied Materials (AMAT) benefits more broadly but faces a higher China/export-control sensitivity.

The key near-term issue is whether reported GPU rental rates translate into durable utilization after power, networking, and depreciation costs. Stable pricing supports hyperscaler ROI, but it does not prove incremental end-demand; a wave of financed GPU-cloud supply could pressure rentals before chip orders slow. Over the next 1-3 months, memory contract-price revisions, HBM qualification yields, and foundry/equipment order commentary matter more than top-down TAM upgrades. A sharp decline in rental pricing, inventory days rising at MU/SK Hynix, or a material reduction in hyperscaler capex guidance would falsify the capacity-cycle thesis.

Consensus appears to be treating AI capex as a single NVDA trade. The more attractive 6-18 month expression is the second derivative: memory and test equipment can experience earnings revisions as HBM supply remains constrained, while systems integrators and GPU-cloud operators face margin compression from component and power costs. The principal tail risk is a synchronized memory capacity response in 2027-28, which historically converts tight supply into severe pricing volatility; avoid paying peak-cycle multiples for undifferentiated DRAM exposure.

Key Decisions for Investors

• Initiate a 3-6 month pair: long LRCX and KLAC / short SMH or a matched NVDA basket. The thesis is that memory-tool estimate revisions have further to run than broad AI-beta; reassess if 2027 WFE order commentary fails to improve by the next two earnings cycles.
• Build a staged long MU position over the next 4-8 weeks rather than chase a single-session move. Use a 12-18 month horizon and cap risk with a stop/review trigger if DRAM and NAND contract pricing turns down for two consecutive months or management guides gross margin below consensus.
• Avoid adding outright NVDA exposure solely on industry TAM revisions; retain existing core exposure only if accelerator lead times and hyperscaler capex plans remain intact. A sustained GPU-rental-rate decline below roughly $5/hour would be an early warning that downstream supply is outrunning monetizable demand.
• Monitor AMAT and ASML as conditional adds, not immediate recommendations: require evidence that incremental memory spending is not offset by additional China restrictions or delayed logic-node investments. For AMAT, export-control changes and China revenue mix are the dominant downside variables.