AI infrastructure materials supply chain Report Interpretation

Overview

This global materials report argues that the AI infrastructure build-out extends well beyond chips. Morgan Stanley expects AI servers, networking and data-center expansion to drive a multi-year increase in demand for materials embedded in PCBs, copper-clad laminates, MLCCs, optical links, semiconductor packaging and thermal management. Its strongest near-term preferences are high-end glass fabric and HVLP4+ copper foil, where capacity constraints are most acute.

Core views

Morgan Stanley’s central thesis is that AI infrastructure creates a “content” cycle rather than only a unit-growth cycle. AI systems require larger and more complex PCBs, more layers, higher-grade copper-clad laminates (CCLs), lower-loss materials and more optical connectivity. The report estimates global PCB TAM will rise from roughly US$58bn in 2025 to US$135bn in 2030, an 18% CAGR, with AI/data-center PCB demand increasing more than sixfold from about US$13bn to US$86bn and reaching roughly 64% of industry revenue. CCL TAM is projected to grow from about US$19bn to US$47bn, or roughly 20% annually; AI/data-center CCL demand rises from about US$4bn to US$30bn, a roughly 47% CAGR and more than sevenfold growth. Morgan Stanley attributes about 90% of CCL industry net growth to AI and data centers. The mechanism is rising material intensity per system. Advanced AI PCBs take approximately 100 days, or 14.6 weeks, to manufacture versus 46 days, or 6.6 weeks, for conventional multilayer boards. NVIDIA-grade computing boards are described as having 22 layers and five lamination cycles, compared with 12-16 layers and one lamination cycle for standard boards. Higher data rates and power density also drive a CCL grade progression from M6/M7 in earlier systems to M8 for Blackwell, M8.5 for selected Rubin boards, and expected M9/M10 materials for Feynman. This increases usage and value of low-loss resin, smoother HVLP copper foil and high-specification glass fabric simultaneously. High-end glass fabric is Morgan Stanley’s top-ranked material opportunity. It estimates a roughly 40% supply deficit in 2026, worsening in 2027 before easing to around 30% in 2028. Constraints include limited availability of Toyota weaving machines, low yields from Chinese-made alternatives, tight high-end yarn supply and rising platinum/rhodium costs that raise capex needs. China prices have more than doubled year to date amid very low inventories. In Japan, glass-cloth ASP rose 40% from ¥3,274/kg in 2023 to ¥4,589/kg in 2025; the high-performance glass-cloth market is forecast to expand from about ¥40bn in 2024 to about ¥150bn by 2030. The report highlights Nittobo’s approximately 90% T-glass market share and capacity-led earnings sensitivity, while Grace Fabric is shifting toward specialty cloth and localizing yarn supply. It also notes that capacity migration into high-end grades can tighten standard E-glass supply and support pricing for vertically integrated producers such as Kingboard Laminates. HVLP4+ copper foil is ranked second and described as the most acute near-term supply constraint. AI-server HVLP demand is expected to reach 24kt in 2026, up 260% year on year, and about 50kt in 2027, up a further 108%. Morgan Stanley’s model projects HVLP4+ demand of about 1,028 tonnes per month in 2026, 3,270 in 2027 and 4,335 in 2028, against capacity of about 1,080, 2,495 and 3,600 tonnes respectively. This moves the balance from a small 2026 surplus to roughly 31% undersupply in 2027 and 20% in 2028. The report links the shortage to high utilization—China’s overall copper-foil operating rate was 93.08% in July 2026—and to the qualifications required for high-speed foil. Mitsui Kinzoku, with 41% of global HVLP4 capacity in 2026, is identified as the primary pricing and volume beneficiary; Co-Tech is characterized as the fastest follower, shifting production toward advanced RTF and HVLP products. Optical fiber is ranked third: fundamentals are strong, but capacity additions lower the medium-term risk/reward relative to glass fabric and copper foil. Hyperscale AI data centers may require 5-10 times more fiber than traditional data centers, with some estimates of 5-36 times more fiber per rack. The report forecasts AI-driven data-center fiber demand growth above 20% annually in 2025-30; YOFC management expects global fiber demand above 670mn fiber-km in 2026 and data-center demand up 69% year on year. Prices have reached seven-year highs, with reported sharp increases across major fiber categories. Morgan Stanley prefers Corning in the US due to committed customer demand and planned capacity additions, while also identifying Fujikura, Furukawa Electric and YOFC as important suppliers. However, aggressive expansion by Corning, Fujikura and Furukawa could gradually normalize supply-demand conditions and pressure prices and margins over the medium term. Resins and MLCCs provide a broader structural participation channel. In CCLs, faster transmission supports migration from conventional epoxy toward PPE/OPE, BMI and cyanate-ester systems, while advanced packaging expands demand for BC/RDL materials. The BC/RDL market was approximately US$600mn in 2025 and is expected to grow about 5% annually through 2029, with RDL outgrowing BC. For MLCCs, AI server mainboards are estimated to contain 15,000-25,000 MLCCs, around ten times general-purpose servers. Morgan Stanley raised global MLCC shipment-value forecasts to US$18.48bn for 2026, US$22.16bn for 2027 and US$25.73bn for 2028. A Rubin VR200 NVL72 rack is estimated to contain roughly 570,000 MLCCs, nearly 80% above a GB300 rack, with total MLCC dollar content about US$4,320 versus US$1,530. High-capacitance 47µF+ devices could exceed 30% of Rubin units, versus less than 20% for GB300, driving an estimated 182% increase in dollar content per rack. The report expects Cloud AI MLCC TAM to reach about US$900mn by 2027 and well above US$1bn by 2030 if infrastructure demand exceeds expectations. Longer-duration opportunities include synthetic diamonds, tungsten, molybdenum, rare-earth oxides and scandium oxide. GPU thermal design power is projected to rise from around 1,200W for GB200 to around 2,300W for Rubin, creating a thermal-management case for diamond-based materials: diamond-copper composites offer roughly 600-1,000 W/(m·K) conductivity and high-purity CVD diamond 2,000-2,400 W/(m·K), versus roughly 400 W/(m·K) for copper. Tungsten benefits from advanced semiconductor deposition and higher-layer PCB drilling, but supply is geopolitically concentrated: China produced about 82% of global mined tungsten in 2024 and its tungsten exports fell 20.7% year on year after export licences were required. Molybdenum could gain as 3D NAND producers replace tungsten in selected word-line applications to reduce RC delay at higher layer counts. Morgan Stanley treats these themes as adoption-dependent, longer-term optionality rather than the clearest immediate bottlenecks. On valuation and positioning, the AI-materials group rose 91% over the preceding 12 months but had corrected roughly 36% from its May-June 2026 high as of 10 September. It traded at about 26x 2026E P/E versus roughly 35x for PCB/CCL peers. Morgan Stanley argues that the valuation gap could narrow because upstream materials benefit from the same AI infrastructure cycle and from a dual content effect—more material and higher-value material per system. It reports house estimates for cash cloud capex of about US$1.2tn in CY26, up 104% year on year, and US$1.6tn in CY27, up 38%.

Analysis framework

The report combines bottom-up supply-demand modeling, technology-roadmap analysis and company-level supply-chain work. It estimates AI hardware volumes and material content per rack, board and switch; compares those requirements with qualified manufacturing capacity; then links shortages, product-grade migration, utilization and pricing to companies across the upstream supply chain. It also compares valuations of AI-material companies with PCB/CCL peers.

Methodology notes

• Industry AnalysisSupply-demand framework

  Bottom-up supply-demand analysis for constrained materials such as high-end glass fabric, HVLP4+ copper foil and optical fiber.

  The report compares projected AI-driven demand with effective qualified capacity to identify shortages, pricing power and likely earnings sensitivity.

• Industry AnalysisVolume-price decomposition

  Material-content and product-mix analysis.

  Morgan Stanley separates growth from higher AI-system volumes from growth in material content, higher specifications and richer product mix per system.

• Valuation methodsP/E and PEG Valuation

  Forward P/E comparison between AI-material suppliers and PCB/CCL peers.

  The report compares approximately 26x 2026E P/E for the AI-materials group with approximately 35x for PCB/CCL peers to frame the potential valuation gap.

• Competition & strategyValue chain analysis

  AI infrastructure materials supply-chain mapping.

  The analysis follows the chain from upstream yarn, foil and resins through CCLs and PCBs to servers, switches, data centers and end applications.

Asset mapping & comparison

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

• Nittobo (3110.T)

  Preferred exposure to high-end glass fabric and T-glass shortages.

  Strengths

  Approximately 90% T-glass market share; dominant position in advanced semiconductor-substrate materials.

  Weaknesses

  Share-price weakness reflected uncertainty over additional price rises and capacity plans.

  Comparison

  Ranked among the report’s preferred glass-fabric exposures.

  Risks

  Competitor qualifications and future capacity additions could reduce scarcity.

• Grace Fabric Technology (603256.SS)

  Preferred high-end electronic-glass-fabric beneficiary.

  Strengths

  Capacity shift to specialty grades, low-DK and low-CTE certification, and planned high-performance yarn and cloth investment.

  Weaknesses

  Expansion requires substantial planned capex of about RMB5.9bn across active projects.

  Comparison

  One of few domestic scaled suppliers of high-end specialty electronic cloth.

  Risks

  Execution of expansion and continued supply tightness are important.

• Mitsui Kinzoku (5706.T)

  Primary HVLP4+ copper-foil beneficiary.

  Strengths

  41% of global HVLP4 capacity in 2026 and AI-related copper-foil profit growth forecast at about 80% CAGR over F3/26-28.

  Weaknesses

  Near-term quarterly sales-plan revisions create noise.

  Comparison

  Clear global capacity leader versus qualified followers.

  Risks

  Large future capacity additions could change the supply balance.

• Co-Tech Development (8358.TWO)

  Fast-following supplier of advanced copper foil for AI boards and high-end switches.

  Strengths

  Redirecting capacity toward RG and HVLP; HVLP4+ capacity expected to scale materially through 2028.

  Weaknesses

  HVLP remains a smaller share of revenue in 2026.

  Comparison

  Ranked behind Mitsui Kinzoku in the report’s copper-foil hierarchy.

  Risks

  Qualification, ramp and capacity-delivery execution.

• Corning (GLW.N)

  Preferred US exposure to AI-driven optical-fiber demand.

  Strengths

  Large customer commitments, planned >50% increase in US fiber capacity and 10x US optical-connectivity capacity expansion.

  Weaknesses

  Committed contracts may limit near-term spot-pricing upside.

  Comparison

  Preferred US optical-fiber exposure versus other suppliers.

  Risks

  Global capacity additions may normalize fiber pricing and margins over the medium term.

• Furukawa Electric (5801.T)

  Beneficiary across copper foil and optical fiber/cable.

  Strengths

  Qualified foil supplier and announced roughly ¥100bn fiber/cable expansion across four countries.

  Weaknesses

  Full-scale fiber expansion is weighted toward later periods.

  Comparison

  A key Japanese supplier in both relevant supply chains.

  Risks

  Later capacity additions could contribute to industry normalization.

• Mitsubishi Gas Chemical (4182.T)

  Preferred electronic-resin exposure.

  Strengths

  Broad portfolio including BT laminate, low-dielectric OPE and cyanate materials for high-speed CCLs.

  Weaknesses

  Resin has less acute near-term scarcity than glass fabric or copper foil.

  Comparison

  Included among preferred exposures for clean earnings sensitivity to AI-material demand.

  Risks

  Demand growth depends on continued CCL grade migration.

• Denka (4061.T)

  Preferred supplier of resin modifiers and low-dielectric silica.

  Strengths

  SNECTON modifiers and low-dielectric silica address higher-performance CCL requirements.

  Weaknesses

  Resin-related opportunity has less scarcity-driven pricing upside.

  Comparison

  Included in the report’s preferred stock exposures.

  Risks

  Product adoption and the pace of next-generation CCL upgrades.

Key data

• Global PCB TAM~US$58bn in 2025 to ~US$135bn in 203018% CAGR; driven almost entirely by AI and data-center infrastructure.
• AI/data-center PCB demand~US$13bn in 2025 to ~US$86bn in 2030More than sixfold growth; estimated to reach ~64% of industry revenue.
• AI/data-center CCL demand~US$4bn in 2025 to ~US$30bn in 2030~47% CAGR and more than sevenfold growth.
• High-end glass fabric deficit~40% in 2026Expected to worsen in 2027 and ease to ~30% in 2028.
• HVLP4+ copper foil balance~31% undersupply in 2027Demand of ~3,270t/month versus ~2,495t/month capacity.
• AI-driven optical-fiber demand>20% CAGR from 2025 to 2030Hyperscale AI data centers may require 5-10x more fiber than traditional data centers.
• AI materials valuation~26x 2026E P/EVersus ~35x for PCB/CCL peers.
• AI materials group performance+91% over 12 months; ~36% below 52-week highAs of 10 September 2026.

Impact & implications

Morgan Stanley argues that the best AI-material exposures are not simply suppliers with AI links, but suppliers whose qualified capacity is difficult to expand while AI-system complexity raises their material content. It sees the clearest near-term combination of scarcity, pricing power and earnings leverage in high-end glass fabric and HVLP4+ copper foil; optical fiber remains supported near term but faces a more material normalization risk as capacity ramps.

Risks

• Optical-fiber capacity additions by Corning, Fujikura and Furukawa could normalize supply-demand conditions and pressure prices and margins over the medium term.
• Limited Toyota weaving-machine supply and low yields from Chinese alternatives constrain glass-fabric capacity expansion.
• MLCC powder prices have remained inelastic; any increase requires negotiation with downstream customers and may compress component-level margins.
• China’s dominant tungsten production and export-licence regime create geopolitical supply risk.
• Longer-duration opportunities in synthetic diamonds, tungsten and molybdenum depend on technology adoption translating into earnings.

What to watch

• Whether high-end glass-fabric shortages persist through 2027-28, including weaving-machine deliveries, yarn availability, inventory and price trends.
• HVLP4+ copper-foil demand, qualified capacity additions and the projected 2027 supply deficit.
• The pace of CCL grade migration from M8 toward M8.5 and M9/M10 in future AI platforms.
• AI data-center fiber demand, pricing and the timing of global fiber and cable capacity additions.
• MLCC mix improvement, high-capacitance device adoption, utilization and book-to-bill trends.
• Evidence that diamond-based thermal materials and molybdenum word-line applications move from development to wider commercial adoption.