Asian semiconductor industry Report Interpretation

Overview

This WSTS-based industry update examines July semiconductor demand through revenue, unit shipments and average selling prices. J.P. Morgan sees memory accelerating on AI-driven allocation and supply tightness, while logic remains healthy but slows modestly because of weaker consumer-electronics demand.

Core views

Overall semiconductor revenue increased 129% year on year in July, down from 143% in June after ten consecutive months of acceleration. The report attributes the still-strong expansion to structurally tight memory supply, accelerating memory-unit shipments as supply is allocated to AI compute, and continued AI demand across accelerators, AI CPUs, networking and AI-peripheral chips. The moderation primarily reflects weaker consumer-electronics demand. Memory was the strongest component: revenue grew 435% year on year versus 399% in June, unit shipments rose 79% versus 50%, and ASPs increased 199% versus 233%. J.P. Morgan says long-term agreements, with 50-70% volume contribution and 20-25% prepayment, are locking in elevated pricing through 2027-28 while giving suppliers more flexible pricing and better demand visibility. The report interprets reductions in SOCAMM capacity per CPU, from 1.5TB to 768GB, and in Rubin Ultra HBM4E configurations, from 16-high to 8-high/12-high, as supply-management responses rather than product de-specification. It expects DRAM wafer allocation to HBM to reach 32% of capacity mix by end-2028E and forecasts AI CPU unit TAM growth at a 155% CAGR. Tight conditions and strong agentic-AI CPU demand create upside risk to its assumption of 42% year-on-year HBM ASP growth in 2027E; HBM price negotiations for 2027E were described as being in later-stage review. NAND is also said to be tracking above expectations, supported by KV-cache offloading demand and emerging solutions. Logic semiconductor revenue rose 35% year on year, versus 38% in June, marking 35 consecutive months of year-on-year growth. Logic unit growth accelerated to 13% from 7%, supported by AI compute and pockets of power-management strength, but consumer electronics remained an offset. At the leading edge, the report sees utilization above 100% at N5/N3, driven by AI accelerators and agentic-AI CPUs, and expects this to continue into 2028. It cites TSMC's indication that AI-demand visibility reaches 2029-30 and expects TSMC to retain more than 95% share in N2/A16. J.P. Morgan does not expect major front-end project wins for Intel because it views Intel 18A as comparable with TSMC N3E; it expects TSMC's N2 family to reach a US$50-60 billion revenue run rate by 2028, when Intel 14A enters mass production. Pricing trends diverge by node and product. Overall semiconductor ASP growth slowed to 97% year on year from 124%; logic ASP growth slowed to 19% from 30%, while memory ASP growth slowed to 199% from 233%. Advanced-node mix improvement at TSMC partly offsets mature-node consumer-electronics weakness in logic. At mature foundries, the report expects continued pricing strength in 8-inch capacity and AI-peripheral applications, but sees 12-inch price increases limited largely to tighter 40nm and 65nm nodes because broader 12-inch pricing remains constrained by weak consumer-electronics demand. Looking ahead, J.P. Morgan expects the AI-driven semiconductor upcycle to remain supported by hyperscaler capital expenditure and widening compute demand. Combined 2Q26 cloud revenue at AWS, Microsoft and GCP was US$106 billion, up 43% year on year versus 35% in the prior quarter. It also identifies neoclouds, including SpaceX, as an additional source of compute demand alongside cloud-service-provider buying intentions. The report reiterates TSMC, ASE, Unimicron, Aspeed, Hon Precision, SK hynix, Chroma, Advantest and Tokyo Electron as its top picks.

Analysis framework

The report uses July WSTS data to separate semiconductor performance into revenue, unit shipments and ASPs, then compares memory and logic trends with the prior month. It links these measures to supply tightness, long-term agreement terms, AI-compute demand, foundry utilization, node-level pricing and hyperscaler cloud-revenue momentum.

Methodology notes

• Industry AnalysisVolume-price decomposition

  Revenue analysis split into unit shipments and average selling prices for memory and logic semiconductors.

  The report uses changes in units and ASPs to explain why memory revenue accelerated and why logic growth and pricing moderated.

• Industry AnalysisSupply-demand framework

  Assessment of memory tightness, capacity allocation, foundry utilization and demand from AI compute and consumer electronics.

  The report connects constrained supply and AI-led demand to elevated memory pricing, while weaker consumer demand limits pricing at some mature logic nodes.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  Transmission from hyperscaler and AI-compute spending into memory, foundry, accelerator, CPU, networking and peripheral-chip demand.

  The report uses AI infrastructure demand and cloud-revenue growth to support its view of demand across semiconductor supply-chain segments.

Asset mapping & comparison

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

• TSMC (2330.TW)

  Reiterated top pick and expected beneficiary of AI-driven leading-edge foundry demand.

  Strengths

  Expected to retain more than 95% market share in N2/A16; utilization at N5/N3 is seen above 100%.

  Comparison

  The report views Intel 18A as comparable with TSMC N3E and does not expect major front-end project wins for Intel.

  Risks

  Broader mature-node pricing remains constrained by weak consumer-electronics demand.

• SK hynix (000660.KS)

  Reiterated top pick linked to HBM and memory-market tightness.

  Strengths

  Memory conditions, long-term agreements and AI CPU demand support elevated HBM pricing.

• ASE Technology Holding (3711.TW), ASPEED Technology, Unimicron, Hon Precision, Chroma ATE, Advantest (6857.T), Tokyo Electron (8035.T)

  Reiterated top picks positioned within the AI-related semiconductor supply chain.

  Strengths

  The report sees continuing AI compute, accelerator, CPU, networking and peripheral demand.

  Risks

  Exposure to weaker consumer-electronics demand varies across semiconductor segments.

Key data

• Overall semiconductor revenue growth129% YoY in JulyVersus 143% YoY in June.
• Memory revenue growth435% YoY in JulyVersus 399% YoY in June.
• Logic semiconductor revenue growth35% YoY in JulyVersus 38% YoY in June; the 35th consecutive month of year-on-year growth.
• Memory unit growth79% YoY in JulyVersus 50% YoY in June.
• Memory ASP growth199% YoY in JulyVersus 233% YoY in June.
• Projected HBM capacity allocation32% of DRAM wafer capacity mix by end-2028EJ.P. Morgan expectation.
• Combined hyperscaler cloud revenueUS$106bn in 2Q26AWS, Microsoft and GCP combined; up 43% YoY versus 35% in the prior quarter.
• TSMC N2 family revenue run rateUS$50-60bn by 2028J.P. Morgan expectation.

Impact & implications

The report sees AI-compute demand extending the semiconductor upcycle through memory, leading-edge logic and selected AI-peripheral applications. It expects memory suppliers to benefit from tight supply and evolving long-term agreements, while mature-node logic remains more exposed to consumer-electronics weakness.

Risks

• Weak consumer-electronics demand is moderating logic growth and constraining broader 12-inch mature-node pricing.
• The report identifies Intel competition as a key investor focus for leading-edge foundry markets.

What to watch

• Memory long-term agreement terms, 2027E HBM pricing negotiations and the pace of DRAM wafer allocation to HBM.
• AI CPU, accelerator, networking and AI-peripheral demand, including hyperscaler and neocloud compute spending.
• Leading-edge foundry utilization and the evolution of 8-inch versus 12-inch mature-node pricing.
• Consumer-electronics demand and its effect on logic units and ASPs.