Morgan Stanley: AI Materials Super Cycle Continues, Supply Shortages in Glass Fabric/Copper Foil/Optical Fiber Extended to 2028
nashnova research
Morgan Stanley says supply gaps in upstream AI materials will persist through 2028, with high-end glass cloth, HVLP4+ copper foil, and optical fiber as the three tightest bottlenecks — and argues these upstream chokepoints offer a more attractive entry than the already fully priced downstream hardware chain.
Why does Morgan Stanley say upstream matters more than downstream?
Downstream AI hardware (GPUs, server systems) already carries high expectations baked into valuations. Upstream bottleneck materials have not caught up.
The AI materials sector rose 91% over the past 12 months but has pulled back roughly 36% from its 52-week high. The 2026 forward P/E sits at about 26×, versus roughly 35× for the PCB/CCL sector.
This means → Morgan Stanley sees room for the valuation gap between upstream materials and PCB peers to narrow.
How severe is the glass-cloth shortage — and why can't it be fixed quickly?
High-performance glass cloth — a reinforcement material in PCB laminates and advanced packaging substrates — faces a triple supply constraint: limited Toyota-loom capacity, tight high-end glass yarn, and high technical barriers.
Chinese market prices have more than doubled year-to-date. Morgan Stanley estimates a ~40% supply gap in 2026, worsening in 2027 before easing to roughly 30% in 2028.
In plain terms = the specialized looms and raw yarn are controlled by a handful of producers; new capacity cannot be built fast, so the gap widens before it shrinks.
Key beneficiaries: Nittobo (capex raised from ¥80 bn to ~¥120 bn through FY2028), Arisawa Mfg., Asahi Kasei, China Jushi, and Taishan Fiberglass.
How tight is the copper-foil bottleneck?
HVLP4+ copper foil — an ultra-low-roughness electrodeposited foil used in AI server PCBs for signal integrity — is becoming one of the most severe supply bottlenecks.
AI-server HVLP copper-foil demand is projected to surge 260% YoY in 2026 to 24,000 tonnes, then another 108% in 2027 to ~50,000 tonnes. Chinese foil capacity utilization already exceeds 90%.
Morgan Stanley's model shows monthly HVLP4+ demand of ~3,270 tonnes in 2027 against available capacity of only ~2,495 tonnes — a ~31% supply shortfall.
This reflects an extremely concentrated supply base: Mitsui Mining alone holds 41% of 2026 capacity, and long qualification cycles make it hard for newcomers to fill the gap quickly.
Why is fiber-optic demand suddenly taking off?
Hyperscale AI data centers require 5–10× more optical fiber than conventional facilities; some estimates put per-rack fiber usage at 5–36× higher.
Fiber prices have hit a seven-year high. Morgan Stanley forecasts AI-driven data-center fiber demand to grow at a 20%+ CAGR from 2025 to 2030.
In plain terms = AI clusters generate far more data traffic than traditional servers, so the "cables" connecting them multiply in kind — prices and demand are surging together.
Key beneficiaries: Corning, Fujikura, Furukawa Electric, YOFC. Morgan Stanley cautions, however, that announced capacity additions could make fiber's medium-term risk-reward relatively less attractive.
How big can the copper-clad laminate market get?
The global CCL market is projected to grow at a ~20% CAGR from roughly $19 bn in 2025 to about $47 bn by 2030.
AI / data-center CCL demand alone is expected to rise from ~$4 bn to roughly $30 bn — a 7×+ increase — contributing about 90% of the industry's net growth over the period.
This means → nearly all incremental demand comes from AI. Traditional PCB demand stays relatively flat — AI is the sole engine for this market.
What is the single variable that determines how far upstream valuations can run?
Material grades keep stepping up: M8 on the Blackwell platform → M8.5 on Rubin → M9/M10 on Feynman — each generation demands higher specs and higher per-system material value.
Advanced AI PCBs take roughly 100 days end-to-end, 2.2× longer than conventional multilayer boards. An NVIDIA-class compute board requires 22 layers and 5 lamination cycles versus just 1 for a standard board.
In plain terms = Morgan Stanley believes the market has not fully priced in the growth opportunity. How far upstream materials can re-rate depends on whether each segment's capacity expansion can keep pace with AI demand growth — that is the single core variable.
市场有风险，内容仅供研究参考，不构成投资建议。