A fresh global materials industry report released by Morgan Stanley on September 13th puts forward a decisive thesis: the AI sector is igniting an entirely new materials supercycle. Unlike the market's previous fixation on terminal hardware like GPUs, optical modules, and HBM, the core dividend of the AI industry has now shifted from growth in equipment volume to a dramatic increase in the value of high-end materials per unit, with multiple upstream material sectors facing substantive supply-demand gaps and ample room for valuation re-rating.
The report specifically highlights high-end fiberglass cloth, HVLP4+ low-roughness copper foil, and data center optical fiber as the three core bottleneck materials with the highest certainty and tightest supply-demand dynamics. Even with significant gains in the AI materials sector this year, considering valuations and industry growth rates, this cycle is not yet complete.
Core Logic: The AI Supercycle Is No Longer About "Shipping Volume" but "Material Content"
Over the past two years, market speculation on the AI industry chain has centered on the shipment scale of computing hardware. However, with the iterative upgrading of AI servers, the industry's underlying logic has completely shifted: the material usage, material grade, and process complexity of advanced AI equipment have doubled, with the value per unit far exceeding that of traditional servers. The most intuitive example is in PCB and CCL substrates: ordinary commercial servers require only 12-16 layers of PCB, while NVIDIA's high-end AI computing boards reach 22 layers, requiring multiple lamination processes, larger dimensions, more complex structures, and comprehensively upgraded material grades.
Industry incremental data is highly explosive: First, the global CCL market is expected to grow from $19 billion in 2025 to $47 billion by 2030, a five-year compound annual growth rate of approximately 20%; among this, AI and data center CCL is projected to surge from $4 billion to $30 billion, an increase of over 7 times in five years, becoming the absolute core increment of the industry. Second, the global PCB market is expected to grow from $58 billion in 2025 to $135 billion by 2030; AI data center PCB is set to grow from $13 billion to $86 billion, an increase of over 6 times, with AI scenario demand accounting for 64% by 2030.
In short, over the next five years, the vast majority of new demand in the PCB and CCL industries will be driven by AI computing power iteration, with the material premium per unit of computing power becoming the core support for this supercycle.
The Three Core Bottleneck Materials: Sustained Supply-Demand Tightness, Pricing Power, and Volume-Price Dual Growth Dividends
High-end fiberglass cloth, ranked as the top priority track in Morgan Stanley's AI materials picks, faces continuously expanding rigid gaps. High-frequency, high-speed AI PCBs place extremely high demands on the dielectric properties, uniformity, and yarn purity of fiberglass cloth, which is completely distinct from traditional consumer and industrial control board materials. Meanwhile, the supply side faces long-term structural hard constraints that cannot be rapidly expanded in the short term. The supply chokepoint is clear: high-end fiberglass production heavily relies on Toyota's high-end looms, but the brand's annual shipments are extremely limited, with monthly shipments of only 300 units expected in 2027; domestic alternative looms currently have insufficient yield rates and have not yet gained recognition from leading high-end manufacturers. Combined with tight capacity in high-end yarn and high technical barriers, the industry's expansion cycle has been significantly prolonged. The supply-demand gap is worsening year by year: the supply gap for high-end fiberglass cloth will reach 40% in 2026 and expand further in 2027, with the industry's tightness expected to gradually ease only by 2028. The year 2027 will be the period of tightest supply-demand and greatest earnings elasticity.
Key beneficiaries include Nittobo, Honghe Technology, Asahi Kasei, China Jushi, Sinoma Science & Technology, and Kingboard Laminates. Among them, Nittobo is significantly increasing capital expenditure, raising its cumulative investment from ¥80 billion to ¥120 billion for 2025-2028, fully positioning itself in high-end capacity.
HVLP low-roughness copper foil is a core material for high-frequency, high-speed signal transmission, effectively reducing signal loss to meet the ultra-high transmission rate demands of AI servers, making it an essential companion for high-end AI boards. Demand is experiencing explosive growth: AI server HVLP copper foil demand in 2026 is expected to surge 260% year-on-year to 24,000 tons, then double again by 108% in 2027 to surpass 50,000 tons. Supply is severely lagging behind demand expansion, and a clear supply gap for HVLP4+ copper foil will emerge in 2027, bringing the industry a triple dividend of scarcity, pricing power, and earnings elasticity. Key beneficiaries include Mitsui Mining & Smelting (with a 41% global market share in 2026, deeply benefiting from the industry gap), Taiwan Union Technology, Furukawa Electric, and quality domestic copper foil manufacturers.
Traditional data centers have limited optical fiber usage, but AI computing clusters rely on massive GPU high-speed interconnects, with exponentially growing data exchange between racks and rooms, directly reshaping the demand structure of the fiber industry. Data shows that optical fiber usage in hyperscale AI data centers is 5-10 times that of traditional data centers, with the maximum difference per rack reaching 36 times. From 2025 to 2030, AI-driven data center optical fiber demand is expected to grow at an annualized rate of over 20%. The industry has already delivered high prosperity: total global optical fiber demand will surpass 670 million core-kilometers in 2026, with data center demand surging 69% year-on-year, and current fiber prices have climbed to seven-year highs, with continuous earnings recovery for leading companies. A risk warning applies here: the barrier to capacity expansion in the fiber track is relatively low, and major manufacturers such as Corning, Fujikura, and Furukawa are all accelerating expansion, which could suppress price upside as capacity releases. Therefore, this track requires careful selection of leaders to avoid oversupply risk. Key beneficiaries include Corning, Fujikura, Furukawa Electric, and Yangtze Optical Fibre and Cable.
CCL Substrates: Continuous AI Iteration Upgrades, Full-Chain Material Binding to Upgrade Dividends
The platform iteration of NVIDIA AI servers continues to drive CCL substrate grade upgrades: from the M6/M7 grades of the Ampere/Hopper era, to the M8 grade of Blackwell, the M8.5 of the Rubin platform, and future Feynman architecture moving to ultra-high grades of M9/M10. The core of these grade upgrades is comprehensive rigid binding of low-loss resins, HVLP copper foil, and low-dielectric fiberglass cloth. In AI high-frequency, high-speed transmission scenarios, ordinary materials suffer excessive signal loss and cannot support stable computing operations, making high-end CCL a hard requirement. At the same time, AI board production difficulty has greatly increased: a single high-end AI PCB has a production cycle of about 100 days, 2.2 times that of traditional multilayer PCBs (46 days), with fiberglass, CCL substrates, and PCB manufacturing each consuming significant time. A shortage in any material link will constrain overall capacity release, further strengthening upstream material barriers.
MLCC Redefined by AI: Shifting from a Consumer Electronics Staple to High-Growth Computing Track
In traditional market perception, the core demand for MLCC comes from consumer and automotive electronics, but AI servers have completely rewritten the industry's demand logic. In terms of usage volume: ordinary servers carry only around 2,000 MLCCs, while AI servers can use 15,000-25,000 units; the GB300 NVL72 rack houses 320,000 MLCCs, and the new generation Rubin rack usage surges to 570,000 units, an increase of nearly 80% per rack. In terms of structure: AI's high-power consumption characteristics are forcing product upgrades, with the share of high-capacity MLCCs above 47μF rising from under 20% in GB300 to over 30% in Rubin. The industry thus enjoys not only volume growth but also the dual dividend of product structure upgrades and higher unit prices. Morgan Stanley estimates that the value of MLCC demand contributed by AI servers could approach $1 billion by 2027. Moreover, high-end, high-capacity MLCC production consumes significant capacity and poses high manufacturing difficulty, so even if AI demand accounts for a modest share of overall industry revenue, it will continue to reshape global capacity allocation and support cyclical prosperity.
Electronic Resins: Steady Upgrades with Less Scarcity Than Core Materials
Adapting to AI high-frequency scenarios, traditional epoxy resins are gradually iterating toward high-end systems such as PPE, OPE, BMI, and cyanate esters, with the industry broadly benefiting from AI-driven sales growth and product structure upgrades. However, compared to fiberglass cloth and HVLP copper foil, the current supply-demand landscape for electronic resins is relatively loose, with weaker scarcity and pricing power. This is a stable-growth track whose explosiveness does not match the three core bottleneck materials. Key companies to watch include Mitsubishi Gas Chemical, DIC, Nippon Kayaku, Kumiai Chemical, and Denka.
Long-Term Hidden Opportunity: New Material Substitution Dividend Driven by AI's High Power Consumption
Beyond the tracks already delivering returns, the continued surge in AI computing power consumption is giving rise to two major long-term material opportunities. First, synthetic diamond: GPU power consumption is climbing from GB200's 1200W to Rubin's 2300W, and the heat dissipation capability of traditional copper materials has peaked. Diamond and diamond-copper composite materials offer far superior thermal conductivity to copper and are expected to be broadly applied in AI chip thermal management scenarios in the future. Second, tungsten and molybdenum new materials: with the iteration of advanced semiconductor processes and 3D NAND architectures, tungsten and molybdenum, leveraging their excellent process adaptability, are expected to see material substitution demand, representing a long-term positioning track.
Valuation Review: The AI Materials Rally Is Far From Over with Ample Re-Rating Room
Data shows that the AI materials sector has risen 91% cumulatively over the past year, below the 147% gain of the PCB/CCL sector. Currently, the AI materials sector trades at an expected 2026 P/E ratio of only 26 times, significantly lower than the PCB/CCL sector's 35 times valuation, and the sector as a whole remains 36% below its 52-week high, leaving ample valuation recovery room. In summary, the core logic of this supercycle is that the AI industry has bid farewell to mere "computing power quantity expansion" and has officially entered a phase of continuously rising high-end material value per unit. In the short term, supply-demand tightness in high-end fiberglass cloth, HVLP copper foil, and optical fiber offers the strongest certainty; in the medium term, CCL, MLCC, and electronic resins continue to deliver incremental gains through upgrades; and in the long term, new materials such as diamond and tungsten-molybdenum open up growth ceilings. The entire AI materials industry chain possesses core logic for sustained re-rating.