US Semiconductors Report Interpretation

Overview

This US semiconductor industry update raises BofA's long-term sales and equipment-spending outlook. The central argument is that AI infrastructure demand is increasingly constrained by physical inputs such as chips, memory, power and manufacturing capacity, supporting a larger and more capital-intensive semiconductor cycle through CY30.

Core views

BofA raises its CY30 semiconductor total-addressable-market estimate to $3.2tn from $2.7tn previously, increasing the implied CY26-30 CAGR to 18% from 14%. The revised outlook is led principally by memory and data-center growth, with an improved recovery outlook for automotive and industrial markets adding incremental support. The report estimates total semiconductor sales of $1.677tn in CY26, up 113.2% year on year, and $3.204tn by CY30. Core semiconductors excluding memory are projected to reach $1.351tn in CY30, implying a 16% CAGR versus the prior $1.089tn and 11% CAGR forecast. The strongest end-market growth is expected in compute and storage, which BofA models up 50% year on year in CY26 on continued server strength. Wired communications is forecast to grow 29% on data-center infrastructure buildout, industrial 32% as demand and inventory conditions improve, and automotive 12% despite weak vehicle-unit trends because semiconductor content is rising. Conversely, wireless communications, smartphones and consumer markets remain constrained by unit headwinds, with CY26 forecasts of -8%, -9.2% and -7%, respectively. Memory is the largest driver of the upgrade. BofA forecasts memory sales to rise nearly 327% year on year in CY26, including DRAM growth of 328% and NAND growth of 341%. It argues that memory shortages and price inflation are a critical lever for industry upside, while AI-server demand provides an additional source of support. Logic is also expected to grow strongly, up 41% in CY26 on AI-accelerator demand, while microprocessors are forecast to rise 32% on hyperscaler consumption, partly offset by weak PC demand. BofA raises its CY26 WFE forecast by 8% to $156bn from $144bn, implying 33% year-on-year growth. CY27 WFE rises to $210bn from $190bn and CY28 to $271.9bn from $250bn. The largest revisions are in memory equipment: CY26 DRAM WFE rises 15% to $45bn and NAND WFE to $15bn from $13.3bn, while CY27 DRAM WFE is raised to $52bn and NAND to $20.5bn. The new CY30 WFE forecast is $360bn, versus about $300bn previously, implying a 25% CY25-30 CAGR; its component forecasts are $114bn for DRAM, $30bn for NAND and $215bn for foundry/logic. The report attributes the longer-term equipment expansion to advanced logic below 3nm, new nodes such as A14, HBM and advanced DRAM, NAND greenfield capacity, and additional cleanroom capacity. It expects non-memory equipment to remain about 60% of total WFE through CY30, but leading-edge WFE to grow faster: leading-edge logic WFE is projected to reach $154bn by CY30, a 31% five-year CAGR. China WFE is estimated at $45bn, or 29% of total WFE, in CY26; although it is expected to grow at a 13% CAGR through CY30, its share is forecast to decline toward 20% as non-China spending grows faster. EUV WFE is forecast to grow at a 28% CAGR versus 25% for ex-EUV equipment. BofA argues that WFE intensity as a percentage of semiconductor sales is less useful in this cycle because memory average selling-price gains mechanically depress the ratio. It models overall WFE intensity at about 11% by CY30, below typical mid-teens levels, but expects it eventually to recover as unit growth becomes a larger driver. Instead, it emphasizes WFE per 12-inch wafer start, which removes pricing effects and measures the equipment required for each additional unit of output. Across six semiconductor upcycles since CY00, WFE outgrew total semiconductor sales in all but two, and median WFE intensity increased by about 300 basis points. The WFE-per-wafer analysis supports the forecast because manufacturing complexity is rising. In foundry and logic, 2nm gate-all-around ramps, low initial yields at emerging foundries, greenfield capacity additions, High-NA EUV and backside-power adoption are expected to lift equipment spending through CY26-28. In DRAM, HBM's three-to-four-wafer trade ratio relative to conventional DRAM, higher layer counts, advanced packaging, hybrid bonding, larger core dies, deeper EUV insertion and possible 4F-squared DRAM are expected to raise WFE per wafer. For NAND, the expected spending drivers are migrations toward 300-layer and 400-layer devices, triple-deck architectures and deeper high-aspect-ratio etch; possible greenfield activity in 2H27/CY28 and demand for high-capacity QLC inference storage represent upside potential. BofA's wafer-based calculation implies WFE of about $210bn in CY27, $278bn in CY28 and $370bn in CY30, directionally supporting its updated $360bn CY30 estimate. On current AI demand, BofA reports no signs of a slowdown in customer orders, long-term agreements, capacity commitments or semiconductor pricing. DRAM and NAND pricing was flattish week on week, while NVIDIA B200 GPU rental pricing was $5.72 per hour, having risen consistently over the prior two months and sitting less than 10% below the approximately $6.10 March peak. The report says 2027 is largely fully booked or contracted across compute, networking and memory vendors, and expects 2028 supply to remain tight as CPU demand, XPU attachment, optics-based scale-up and accelerator activity expand. Despite the SOX rising 67% year to date before a roughly 17% correction, BofA considers semiconductor valuation attractive relative to growth: the group trades at 18x next-twelve-month P/E, below the S&P 500's 19x, despite projected 139% year-on-year EPS growth. Still, it remains tactically cautious until midterm elections pass and macro concerns moderate, expecting the sector potentially to remain range-bound near term. It identifies NVIDIA and AMD in compute, Marvell in networking, and Analog Devices and onsemi in analog as relatively resilient; if momentum resumes, it expects Micron, Lam Research, Applied Materials and Intel could lead.

Analysis framework

BofA updates top-down semiconductor sales forecasts by end market and device type, cross-checks them against bottoms-up company estimates, and then translates projected memory and logic wafer volumes into WFE demand. It emphasizes WFE per 12-inch wafer start rather than equipment intensity as a percentage of sales because memory price inflation can distort the latter measure.

Methodology notes

• Industry AnalysisSupply-demand framework

  Semiconductor demand and supply analysis by end market, device type, capacity commitments and pricing.

  The report links AI-server demand, memory availability, pricing, capacity additions and end-market recovery to projected semiconductor sales and equipment spending.

• Industry AnalysisVolume-price decomposition

  Separating unit-driven growth from average-selling-price-driven growth.

  BofA explains that memory ASP inflation can make WFE intensity as a share of sales look artificially low, so it distinguishes pricing effects from underlying production and equipment requirements.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  Translating semiconductor wafer demand and manufacturing complexity into equipment spending.

  The report uses projected logic and memory wafer starts, together with WFE-per-wafer assumptions, to derive aggregate WFE demand through CY30.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• NVIDIA (NVDA)

  Compute and AI-infrastructure beneficiary; identified as relatively resilient in a potentially range-bound sector.

  Strengths

  Leading share in AI compute and networking markets.

  Weaknesses

  Lumpy global AI projects and cyclical gaming exposure.

  Comparison

  BofA groups NVIDIA with AMD as a compute resilience candidate.

  Risks

  Gaming weakness, AI competition, China shipment restrictions, lumpy enterprise and data-center demand, power constraints and government scrutiny.

• Advanced Micro Devices (AMD)

  Compute beneficiary and relative-resilience candidate.

  Strengths

  AI CPU/GPU share-gain potential and more than 50% annual EPS CAGR potential cited in its price-objective rationale.

  Weaknesses

  Slower growth in PC, embedded and console markets.

  Comparison

  Grouped with NVIDIA in compute.

  Risks

  MI400 rack-scale execution, timing and magnitude of Middle East AI projects, lumpy spending, reliance on one outsourced manufacturing partner and console-cycle maturity.

• Marvell Technology (MRVL)

  Networking beneficiary and relative-resilience candidate.

  Strengths

  AI connectivity, switching, compute and custom-ASIC opportunities.

  Weaknesses

  Exposure to cyclical storage, enterprise-networking and carrier markets.

  Comparison

  Identified as the report's networking resilience candidate.

  Risks

  Reduced visibility on major custom ASIC projects and competition from other merchant vendors and incumbent ASIC suppliers.

• Analog Devices (ADI)

  Analog beneficiary and relative-resilience candidate.

  Strengths

  Best-in-class profitability and secular growth across AI, industrial, automotive and communications.

  Weaknesses

  Cyclical exposure to automotive and industrial demand.

  Comparison

  Grouped with onsemi as an analog resilience candidate.

  Risks

  Economic downturn, unrealized Maxim cost synergies, lower-cost competition and exposure to US-China tensions through China automotive.

• onsemi (ON)

  Analog and automotive/industrial recovery beneficiary.

  Strengths

  Potential EPS and free-cash-flow recovery, silicon-carbide and image-sensing content and share gains, and restructuring benefits.

  Weaknesses

  High automotive and industrial exposure.

  Comparison

  Grouped with ADI as an analog resilience candidate.

  Risks

  Macro and trade risks, delayed automotive/industrial recovery, 300mm fab ramp difficulties and elevated capital spending.

• Micron Technology (MU)

  Memory-cycle beneficiary that could lead if sector momentum resumes.

  Strengths

  Exposure to the memory upcycle and AI HBM business.

  Weaknesses

  Sensitivity to memory pricing and end-market demand.

  Comparison

  BofA separates traditional cyclical memory value from AI HBM value in its sum-of-the-parts framework.

  Risks

  Larger-than-expected memory ASP declines, China competition, share loss and weaker data-center, smartphone or PC demand.

• Lam Research (LRCX)

  WFE-cycle beneficiary that could lead if momentum resumes.

  Strengths

  Etch/deposition leadership, rising process intensity, WFE-cycle exposure and free-cash-flow generation.

  Weaknesses

  Near-term cost inflation and tariff concerns.

  Comparison

  Grouped with Applied Materials and Intel as potential leaders if sector momentum resumes.

  Risks

  Slower capital spending, delayed memory capacity additions, share loss, integration risk, macro headwinds, customer consolidation and China exposure.

• Applied Materials (AMAT)

  WFE-cycle beneficiary that could lead if momentum resumes.

  Strengths

  Potential WFE outgrowth in 2026/27.

  Weaknesses

  More balanced growth profile and lower profitability than certain large peers.

  Comparison

  Expected to benefit alongside Lam Research from a stronger WFE cycle.

  Risks

  US government probe, slower capital spending, delayed memory capacity additions, segment share loss, integration risk and macro headwinds.

• Intel (INTC)

  Potential leader if semiconductor momentum resumes, with server CPU and external-foundry optionality.

  Strengths

  External-foundry wafer and packaging opportunities and potential server-CPU growth.

  Weaknesses

  Near- and medium-term manufacturing ramp uncertainty and mature-PC exposure.

  Comparison

  Grouped with equipment and memory names as a potential momentum leader.

  Risks

  Lower-than-expected 18A/14A yields and ramps, lack of major external-foundry customers, weak PC trends and CPU share loss.

Key data

• CY30 semiconductor TAM$3.2tnRaised from $2.7tn; implies 18% CY26-30 CAGR versus 14% previously.
• CY26 semiconductor sales growth+113.2% YoYForecast total semiconductor sales of $1.677tn.
• CY26 memory sales growth+326.9% YoYDRAM is forecast up 328.2% and NAND up 341.4%.
• CY26 WFE forecast$155.9bnRaised from $144.4bn; +33.4% YoY.
• CY30 WFE forecast$359.8bnRaised from $291.7bn previously; implies a 23% CY26-30 CAGR.
• NVIDIA B200 rental price$5.72/hourLess than 10% below the approximately $6.10 March 2026 peak.
• Semiconductor valuation18x NTM P/EBelow the S&P 500's 19x despite projected 139% YoY EPS growth.

Impact & implications

The report's revised forecast shifts the industry outlook toward a longer, more equipment-intensive AI-led expansion, with memory and leading-edge logic as the primary drivers. It argues that the expected growth in wafer complexity and capacity should benefit compute, networking, analog, memory and semiconductor-capital-equipment exposure, although near-term macro uncertainty may restrain sector performance.

Risks

• The report remains tactically cautious because macro concerns may keep the sector range-bound until midterm elections pass.
• A slowdown in AI infrastructure spending, customer orders, capacity commitments or memory demand would challenge the forecast.
• Memory price declines could reduce the projected sales and WFE expansion.
• Delays in memory capacity additions, slower capital spending, tariffs, China exposure and competition are cited as risks for equipment and semiconductor companies.

What to watch

• Memory pricing, particularly DRAM and NAND, as a key lever for industry growth upside.
• Customer orders, long-term agreements, capacity commitments and GPU rental prices for evidence on AI infrastructure demand.
• The pace of HBM, advanced DRAM, leading-edge logic, EUV and High-NA EUV adoption.
• Cleanroom and greenfield capacity additions, including potential NAND greenfield activity from 2H27/CY28.
• Macro conditions and the sector's performance after the midterm elections.