From "chasing the light" to "going upstream toward the light"! Morgan Stanley asserts: In the tsunami of computing power, fiberglass fabric and copper foil ignite a super cycle in materials.

According to The Smart Investor, Morgan Stanley has released a new research report indicating that the frenzied expansion of global AI infrastructure is rapidly spreading upstream from downstream areas such as GPUs and semiconductor foundries to key basic materials. This trend is not only reshaping the supply chains of printed circuit boards (PCB) and multi-layer ceramic capacitors (MLCC), but is also creating a new, prolonged, and highly certain "supercycle" for upstream materials.

AI Accounts for Almost All Industry Net Growth

Morgan Stanley estimates that the global PCB market size will expand at a compound annual growth rate (CAGR) of 18%, growing from approximately $58 billion in 2025 to about $135 billion in 2030. Of this, demand related to AI/data centers will increase more than sixfold, rising from around $13 billion to $86 billion, while its share of industry revenue will soar from 20%–25% to 64%. In other words, by 2030, AI and data centers will account for virtually all net increases in the industry, with other end market demand remaining essentially flat.

The slope is even steeper for copper-clad laminates (CCL): The global market size will grow from roughly $19 billion to $47 billion (CAGR around 20%), whereas demand for AI/data center CCL will skyrocket over sevenfold from about $4 billion to $30 billion (CAGR approximately 47%), contributing about 90% of the sector’s net growth. Nvidia’s platform upgrades—from Blackwell's M8 level to Rubin’s M8.5, and eventually to the anticipated Feynman M9/M10—continue to drive up the material usage and value per machine.

Manufacturing complexity is also fundamentally changing: The end-to-end lead time for a high-end AI PCB is about 100 days, which is 2.2 times longer than that of standard multilayer boards (46 days); Nvidia-grade computation boards require 22 layers and five pressings, compared to only one pressing for conventional 12–16-layer boards.

Top Three Most Scarce Materials

Number one: High-end fiberglass cloth. Due to limited supply from Toyota Industries (monthly shipments of only 300 units by end-2027), tightness in high-end fiberglass yarn, and significant technical barriers, the price of Chinese fiberglass cloth has doubled within the year while inventory remains extremely low. Morgan Stanley expects a supply gap of about 40% in 2026, which will further deteriorate in 2027 and only ease back to around 30% by 2028. The average price of Japanese fiberglass cloth has already surged from 3,274 yen/kg in 2023 to 4,589 yen/kg in 2025, an increase of 40%. Nittobo has significantly raised its cumulative capital expenditure for FY2025–2028 from 80 billion yen to about 120 billion yen.

Number two: HVLP4+ ultra-low profile copper foil. According to CBC Metals, the demand for HVLP copper foil in AI servers will grow by 260% year-on-year to 24,000 tons in 2026 and increase another 108% to approximately 50,000 tons in 2027. The operating rate in the Chinese copper foil industry has already surpassed 93%. Morgan Stanley’s model shows a supply-demand gap of about 31% in 2027 (monthly demand at 3,270 tons versus supply capacity of only 2,495 tons), with a remaining gap of about 20% in 2028. Supply is highly concentrated, with Mitsui Mining & Smelting alone controlling 41% of global capacity in 2026, making it the main beneficiary in both volume and price.

Number three: Optical fiber, with prices at a seven-year high. The usage of optical fiber in ultra-large-scale AI data centers is 5–10 times that of traditional data centers; some estimates put usage per rack at 5–36 times higher. From 2025 to 2030, AI-driven data center optical fiber demand is expected to have a CAGR of more than 20%, with data center-related optical fiber demand growing 69% year-on-year in 2026. Corning has committed to boosting US fiber capacity by 50% and optical connectivity capacity by 10 times. Lumen alone has secured 10% of Corning’s global capacity for 2025–2026.

Valuation: Correction Presents a Second Opportunity

The AI materials segment has surged 91% over the past 12 months, though it still lags behind the PCB/CCL segment’s 147% return in the same period; since reaching 52-week highs in May–June 2026, the sector has seen an average pullback of about 36%. The current sector’s 2026 expected P/E ratio is about 26 times, compared to roughly 35 times for PCB/CCL—leaving room for the valuation gap to narrow. Morgan Stanley emphasizes that "the next phase is about content, not quantity": Their forecast for a 2030 CCL market size of around $47 billion significantly exceeds the consensus of $30–35 billion, reflecting the ongoing underestimation of per-machine material value. Capex remains robust—Morgan Stanley expects cash cloud capital expenditure to reach approximately $1.2 trillion in 2026 (year-on-year +104%) and about $1.6 trillion in 2027 (+38%).

Portfolio and Long Tail

Morgan Stanley’s top picks are: Nittobo, Grace Fabric, Mitsui Mining & Smelting, Co-Tech, Corning, Furukawa Electric, Mitsubishi Gas Chemical, and Denka.

Resins (PPE/OPE, BMI, cyanate ester) are structural beneficiaries but face less scarcity. For MLCC, Murata saw an orders-to-shipments ratio of 1.47 from April to June, the highest since 2004, with an operating rate of about 95%.

Longer-term options include: Synthetic diamonds (Rubin’s power output reaches 2300W—traditional copper cooling is insufficient; diamond-copper composite materials offer thermal conductivity of 600–1000 W/m·K, with high-purity CVD diamonds reaching 2000–2400), the substitution of tungsten and molybdenum (3D NAND shifting from tungsten to molybdenum), and rare earths and scandium oxide (SOFC power solutions for data centers).

Risk Disclosure

Aggressive mid-term expansion of optical fiber capacity could normalize supply-demand and suppress prices; Toyota Industries’ bottleneck is unlikely to be resolved soon, restricting fiberglass cloth expansion; price increases for MLCC dielectric powders will still require tough negotiations with downstream customers, potentially squeezing component margins.

Conclusion

Morgan Stanley’s core viewpoint is that much of the downstream AI hardware valuation has already priced in most of the narrative, while upstream materials—especially high-end fiberglass cloth, HVLP copper foil, and optical fiber—will remain in short supply at least until 2028. That’s where the largest valuation gap in this supercycle lies.