Nomura — Asia AI PCB (printed... Nomura — Asia AI PCB (printed circuit board)/… · @2030FY_X · TwKit

@2030FY_X

Nomura — Asia AI PCB (printed... Nomura — Asia AI PCB (printed circuit board)/IC substrate and CCL (copper-clad laminate): 2026 TPCA show observations — further supply tightening, technological innovation advancing in parallel I. Core view • Takeaways from the Taipei IMPACT-IAAC and TPCA show: the AI compute hardware bottleneck is shifting down the entire chain from wafers and advanced packaging to “substrates—PCB—CCL—electronic glass cloth—copper foil—key equipment.” Supply constraints are not a short-term spike but a structural shortage from 2H 2026 to 2028. • Overall ABF (Ajinomoto Build-up Film) substrate nominal capacity is in surplus, but the shortage of high-end ABF substrates (large size, high layer count, fine line width/spacing, embedded passive/active devices) continues to widen; lengthening lead times for materials and equipment are the core source of pricing power. II. Packaging substrate and materials bottlenecks • Low-CTE (coefficient of thermal expansion) T-glass (low-dielectric glass cloth) remains in tight supply: Nittobo plans to more than double its “thick T-glass” capacity by FY28E versus FY26, but constrained by loom bottlenecks, its long-term share is expected to continue declining. The core loom supply comes from Toyota Industries, with lead times extended from about 3 years last year to 5 years (to 2031–2032); Tsudakoma equipment has not yet fully met industry standards. • The price increase cycle for electronic-grade E-glass (conventional electronic glass cloth) extends beyond 1H 2027; the narrowing spread between E-glass and Low-Dk1 glass cloth may trigger capacity switching, which would instead push up Low-Dk1 prices. • Lead times for key equipment have lengthened across the board: CO2 laser drilling machines from 12–14 months to 3 years; exposure equipment from 1 year to 3–5 years; ABF vacuum laminators from less than 6 months to 2.5 years. • Substrate technology roadmap: size moving from 90–100 mm toward 120–150 mm and even 230 mm; layer counts evolving from 10+2+10 to 18+n+18; line width/spacing iterating from 8/8 μm to 5/5 μm and 3/3 μm; embedding IPD (integrated passive devices), DTC (deep trench capacitor), MLCC and multi-layer core (MLC) becoming system-level integration directions. • Glass core TGV (through-glass via) is returning to the industry’s focus, but SeWaRe (delamination micro-cracking) and reliability remain the main obstacles before commercialization; aluminum nitride ceramic core has high thermal conductivity advantages, but Dk (dielectric constant) is relatively high. Commercialization of glass core still requires time. III. PCB/HDI and CCL evolution • AI platform material upgrade path: in 2026–2027, the mainstream will migrate from M8K1/K2 (paired with M4/M6) to M9K1/K2; Rubin Ultra compute boards tend toward an HDI+5 structure; high-layer HDI migration is accelerating for switch boards and compute boards. • Optical modules are driving mSAP (modified semi-additive process) HDI upgrades: 800G uses 10–12 layers, 1.6T uses 12–16 layers, and NPO (near-package optics) requires 16–18 layers; embedded silicon capacitors (Si-cap) improve high-frequency performance. Zhen Ding (ZDT) has active IC embedding experience and a clear first-mover advantage. • Copper foil supply inflection point: HVLP3/4 (ultra-low-profile copper foil) is in tight supply driven by AI platforms; leading players exiting low-end HTE/RTF capacity has caused simultaneous tightening in copper foil for consumer-grade and general-server applications. Domestic suppliers such as Defu and Tongguan are entering the substitution sequence. • Resin route divergence: the hydrocarbon resin (HCR) camp—EMC, Doosan, and Nan Ya; the PTFE (polytetrafluoroethylene) camp—SYTECH and TUC are positioned at M9/M10 and above. PTFE offers superior electrical performance, but high-temperature lamination, drilling, desmear, and interfacial bonding yield are difficult; the hybrid approach (PTFE+HCR) is a transitional solution, and the final outcome will be hard to determine before mid-2027. IV. Price and Earnings Pass-Through • Mid- to low-end CCL is implementing monthly price hikes driven by E-glass price increases; high-end CCL, where supply and demand are tighter, is expected to have 10%–20% room for price increases from Q4 2026, further optimizing the gross margin structure. • If demand for AI CPU/customized platforms beats expectations, shortages will be worse than the base-case assumption, and upward revisions to pricing and earnings forecasts will be the sector’s core catalyst. V. Investment Conclusion and Ratings • IC substrates: Reiterate Buy on Unimicron, benefiting from the high-end ABF shortage, EMIB-T, and leadership in large-size and embedded technologies. • PCB/HDI: Reiterate Buy on ZDT, Compeq, SCC, VGT, and WUS; the HDI migration benefits leading board makers with high-layer HDI capabilities. • CCL: Reiterate Buy on EMC and SYTECH; EMC is advantaged by its capacity scale and positioning in HDI materials, while SYTECH benefits from rising PTFE and AI CCL penetration. • Risks: longer equipment lead times, glass cloth/copper foil capacity expansion below expectations, PTFE process yield below expectations, slower customer qualification, and AI capex volatility. VI. Industry Mapping Conclusion This is not a “cycle peak” but “bottlenecks shifting to later stages + material repricing.” What is truly scarce is effective capacity that can pass major-customer qualification, not nominal capacity footprint. Slow-moving supply variables (looms, T-glass, HVLP4, laminators, HDI yield) will dominate profit allocation in 2027–2028.

曝光 193 · 评论 0 · 点赞 0 · 书签 0