After a 15% pullback, semiconductors may continue to outperform in 2H26

Overview

This report is JPMorgan's 2Q26 earnings preview for the semiconductor and semiconductor capital equipment industries. The core view is that AI/data center demand continues to accelerate, tight memory supply and demand is extending, and the semiconductor equipment upcycle is being lengthened by AI-driven fab construction, while cyclical areas such as industrials, automotive, analog, and MCUs are gradually recovering. The report believes that after a 13-15% pullback, the semiconductor sector still has the conditions to outperform in 2H26.

Core views

The report's core views include: first, 2Q26 results and 3Q26 guidance are likely to exceed market consensus, extending the positive earnings revision cycle; second, hyperscalers and AI labs remain compute-constrained, and upward revisions to AI infrastructure capex improve visibility across the NVDA, AVGO, and MRVL supply chain; third, custom ASICs/XPUs will continue to gain share within the AI accelerator TAM; fourth, tight DRAM, NAND, and HBM supply and demand may persist through 2028E, with the key to a re-rating of memory stocks being whether long-term agreement structures can reduce cyclicality; fifth, the WFE cycle is strengthening and lengthening due to AI-driven investment in advanced nodes, HBM, and advanced packaging; sixth, consumer electronics and PCs/smartphones are relatively weaker areas, pressured by memory BOM inflation.

Analysis framework

The report cross-validates its views using a pre-earnings-season outlook, company management commentary, industry sales data, hyperscaler capex forecasts, AI accelerator shipment mix, memory supply and demand models, WFE capex forecasts, and end-market trends by segment. Its focus is not on the valuation of any single company, but on assessing relative sector performance through demand, supply, earnings revisions, and industry bottlenecks.

Methodology notes

• earnings_previewEarnings and Guidance Upgrade Framework

  Assessing near-term sector momentum through 2Q results, 3Q guidance, EPS revisions, and forward-looking management commentary.

  The report believes the earnings upgrade cycle will continue for multiple consecutive quarters, with market attention shifting from whether the current quarter beats expectations to forward visibility for 2H26 and 2027.

• supply_demandTight Memory Supply and Demand Framework

  Assessing the pricing and earnings cycle based on supply constraints for DRAM, NAND, and HBM and growth in AI/server demand.

  Priority allocation of capacity to HBM is compressing traditional DRAM supply, while rising AI server and eSSD demand is increasing NAND consumption, extending the supply-demand tightness.

• capex_cycleAI Data Center Capex and WFE Cycle Framework

  Evaluating the semiconductor equipment cycle using hyperscaler capex, advanced-node capacity expansion, fab construction, and equipment intensity.

  The report expects WFE to maintain strong growth in CY26 and CY27 and believes customers' early reservations of equipment slots will lengthen the cycle.

• product_mixGPU and ASIC/XPU Share Migration Framework

  Assessing changes in custom AI chip share based on fragmented inference workloads and TCO advantages.

  The report expects ASICs/XPUs to account for approximately 32% of AI accelerator shipments in 2025, rising to approximately 42% in 2026E and 53% in 2027E.

Asset mapping & comparison

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

• AVGO

  Beneficiary of AI networking, custom ASIC, and high-end smartphone content exposure

  Strengths

  Clear AI revenue trajectory, approximately 70% share of switching chips, and high visibility into customer demand for custom XPUs.

  Weaknesses

  Some consumer demand remains affected by the smartphone cycle.

  Comparison

  Compared with conventional consumer semiconductors, AVGO is more dependent on structural growth in AI networking and custom chips.

  Risks

  Sustainability of AI capex, supply chain constraints, and the pace of customer budgets.

• NVDA

  Core beneficiary of AI GPUs and system platforms

  Strengths

  The Blackwell and Rubin roadmaps provide visibility into demand for CY27 and beyond, while AI infrastructure expansion supports shipments.

  Weaknesses

  Rising ASIC/XPU share may relatively reduce GPU unit share.

  Comparison

  It remains a core AI accelerated computing play, but its relative share may be affected by custom chip growth.

  Risks

  Advanced-node, HBM, advanced-packaging, and power supply bottlenecks, as well as the sustainability of AI capex after 2027.

• MRVL

  Beneficiary of AI networking and optical interconnect

  Strengths

  Optical business growth of approximately 70% year over year, benefiting from AI data center network upgrades.

  Weaknesses

  Growth is highly dependent on the pace of AI customer deployments and the network upgrade cycle.

  Comparison

  Compared with traditional communications semiconductors, it benefits more directly from the upgrade from 800G to 1.6T.

  Risks

  Customer order volatility, supply chain constraints, and elevated valuation expectations.

• MU

  Beneficiary of tight DRAM, NAND, and HBM supply and demand

  Strengths

  Memory pricing and shipments are supported by AI servers, priority HBM allocation, and long-term agreements.

  Weaknesses

  Memory stocks have risen significantly over the past three months, and some cyclical strength has already been priced in.

  Comparison

  Compared with traditional memory cycles, this cycle may be less volatile due to improvements in SCA/LTA and HBM mix.

  Risks

  New capacity, slowing traditional demand, and long-term agreement terms falling short of expectations.

• KLAC

  Beneficiary of increased semiconductor equipment and process control intensity

  Strengths

  Advanced nodes, HBM, and AI fab expansion are increasing demand for process control.

  Weaknesses

  Sensitive to the fab capex cycle.

  Comparison

  Listed as one of the top picks in semiconductor capital equipment.

  Risks

  Downward revisions to WFE budgets, normalization of China WFE, and export controls.

• SNPS

  Beneficiary of EDA/IP and complex chip design

  Strengths

  AI accelerators, custom ASICs, 3DIC, and chiplets are increasing design complexity.

  Weaknesses

  Investors remain focused on the pace at which AI generates incremental monetization for the EDA business model.

  Comparison

  Compared with hardware cycle stocks, EDA/IP better reflects design complexity and the resilience of software/IP revenue.

  Risks

  AI workflow monetization slower than expected, customer consolidation, and weaker-than-expected synergies in system design combinations.

Key data

• Recent semiconductor sector pullback13-15%The report believes the pullback reset elevated pre-earnings-season expectations and created conditions for outperformance in 2H26.
• Top-4 U.S. hyperscaler data center capexApproximately +80% YoY in 2026E; at least +50% in 2027EJPMorgan further raised its 2026 forecast from the previous expectation of more than 70% growth.
• AI accelerator shipments2025E 10.1 million units; 2026E 16.3 million units; 2027E 23.3 million unitsThis corresponds to +62% YoY in 2026E and +43% YoY in 2027E.
• ASIC/XPU shipment shareApproximately 32% in 2025E; approximately 42% in 2026E; approximately 53% in 2027EThis reflects the expansion of hyperscalers' internally developed custom chip programs.
• HBM TAM$65bn in 2026E; $129bn in 2027E; $225bn in 2028EThe report expects the HBM supply-demand imbalance to persist through 2028E.
• WFE growth forecast+28% in CY26; +29% in CY27; +16% in CY28Supported by AI-driven advanced nodes, memory, and fab expansion.
• PC and smartphone shipment forecastsPCs approximately -9% YoY in 2026; smartphones approximately -11% YoYThe main pressures are BOM inflation caused by rising memory prices and the risk of demand destruction.
• Industry sales momentumWSTS May sales +119% YoY; +34% YoY excluding memoryThe report uses this data to support the view that industry demand momentum remains strong.

Impact & implications

If the report's view materializes, the semiconductor sector's investment theme will expand beyond pure AI compute growth to include memory, networking, equipment, EDA, and cyclical recovery in industrials and automotive. Beneficiaries include AI compute and networking leaders, memory manufacturers, equipment makers, and EDA/IP companies; relatively pressured areas are low- to mid-end consumer electronics, PCs, and smartphones. Although valuations are no longer cheap, the report believes they remain supportable given relative earnings growth.

Risks

• Demand destruction in PCs and smartphones caused by memory BOM inflation.
• Uncertainty over AI capex growth and hyperscaler monetization capabilities after 2027.
• Constraints on advanced-node wafers, HBM, advanced packaging, and power supply may limit AI hardware shipments.
• Normalization of China WFE and changes in export controls may affect equipment demand.
• A deterioration in the macro environment may weaken the recovery in industrials, automotive, and consumer electronics.
• New memory capacity or slowing traditional demand may alter the tight NAND/DRAM supply-demand landscape.

What to watch

• Whether 2Q26 results broadly exceed consensus and whether 3Q26 guidance continues to be revised upward.
• Management commentary on 2H26 and CY27 demand, orders, deliveries, and supply chain bottlenecks.
• Whether the 2026 and 2027 capex budgets of the Top-4 U.S. hyperscalers continue to be raised.
• Changes in ASIC/XPU share relative to GPUs and order visibility for AVGO, MRVL, and NVDA.
• Memory manufacturers' SCA/LTA terms, including prepayments, take-or-pay provisions, and long-term pricing mechanisms.
• HBM supply allocation, HBM4/HBM4E ramp-up, and ASP trends.
• Whether WFE customers continue to reserve equipment slots early and whether TSMC and DRAM manufacturers raise capex.
• Whether PC and smartphone demand declines more visibly due to rising memory prices.