The bottleneck in AI PCB upgrades is shifting from PCB manufacturing to CCL and electronic fiberglass cloth

Overview

Based on a joint analyst call hosted by Citigroup, this report discusses changing bottlenecks in the AI PCB supply chain from PCB and CCL to upstream electronic fiberglass cloth and copper foil. The core conclusion is that AI server architecture upgrades are increasing PCB layer counts, CCL material grades, and copper foil and fiberglass cloth specifications, making demand more material-intensive; compared with the PCB manufacturing segment, upstream CCL and electronic fiberglass cloth supply is more constrained by capacity expansion cycles, customer qualification, and equipment bottlenecks, and therefore has stronger pricing leverage and earnings elasticity.

Core views

The report believes PCB capacity is expanding the fastest, and AI-related price increases are mainly reflected in material cost pass-through, leaving relatively limited room for margin expansion; CCL capacity growth is expected to exceed 20% this year, and as AI projects ramp up in the third quarter, some PCB makers may face CCL allocation issues; electronic fiberglass cloth supply is the slowest to respond, and traditional E-glass and standard electronic cloth are being squeezed as capacity shifts toward higher-end Low Dk products, with the monthly price increase trend likely to continue. China Jushi is viewed as the preferred beneficiary in China’s materials chain due to its capacity, pricing, product mix upgrades, and potential for special electronic cloth certification.

Analysis framework

The report conducts top-down and supply-chain cross-validation from AI server end demand, PCB specification upgrades, the CCL supply-demand gap, copper foil and fiberglass cloth material bottlenecks, capacity expansion cycles, price transmission, and stock-specific beneficiary pathways. It focuses on comparing the pace of capacity growth, pricing mechanisms, supply constraints, and earnings elasticity across different segments.

Methodology notes

• Supply Chain AnalysisBottleneck Shift Framework

  Shifting from PCB manufacturing upstream to CCL and electronic fiberglass cloth

  By comparing the expansion speed, technical barriers, customer qualification cycles, and equipment constraints of PCB, CCL, and electronic fiberglass cloth, the report concludes that scarcity and pricing power in the AI PCB upgrade cycle are shifting to more upstream material segments.

• Price and Profit AnalysisPrice Transmission and Earnings Elasticity

  The difference between cost pass-through and direct price hikes

  PCB price increases are more about passing through AI-related material costs, with limited margin improvement; CCL and electronic fiberglass cloth, due to tighter supply, see more direct price hikes and stronger earnings elasticity.

• Capacity Cycle AnalysisCapacity Expansion Cycle and Supply Response

  Constraints from capex, equipment, and qualification

  Greenfield PCB expansion takes about 13 to 15 months from groundbreaking to meaningful revenue recognition; greenfield CCL projects take about 18 months in China and about two years in Southeast Asia; new electronic fiberglass cloth capacity takes about 1.5 years and is limited by loom supply and customer qualification, causing upstream supply response to lag demand upgrades.

Asset mapping & comparison

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

• China Jushi (600176.SS)

  Top China materials beneficiary pick, benefiting from rising electronic fiberglass cloth prices, capacity expansion, product mix upgrades, and potential special AI electronic fiberglass cloth certification.

  Strengths

  Large-scale electronic fiberglass cloth capacity, with incremental additions from the Huai’an project; rising thin cloth mix and relatively high gross margin; if special electronic fiberglass cloth passes customer certification, it may drive valuation rerating.

  Weaknesses

  Special electronic fiberglass cloth is still at the customer sampling stage, and the timing and share of scaled supply remain uncertain.

  Comparison

  Compared with the PCB manufacturing segment, China Jushi is positioned further upstream in electronic fiberglass cloth materials, where supply is tighter and price elasticity is higher.

  Risks

  AI server ramp-up below expectations, slower-than-expected certification of special electronic fiberglass cloth, and weaker-than-expected gains in electronic fiberglass cloth prices.

• EMC / Elite Material (2383.TW)

  Preferred name in the CCL segment, benefiting from tight supply, rising prices, and product mix upgrades.

  Strengths

  Limited CCL supply growth may strengthen pricing power after AI projects ramp up; product mix upgrades toward M7/M8 and above bring earnings elasticity.

  Weaknesses

  AI-related M8 price increases are relatively moderate, and long-term customer relationships still need to be balanced.

  Comparison

  Among CCL suppliers, the report has a higher preference for EMC, which benefits more directly from price increases than PCB makers.

  Risks

  Faster-than-expected qualification of new suppliers, customer bargaining pressure limiting price increases, and delayed AI demand timing.

• TUC (6274.TWO)

  CCL beneficiary, benefiting from tight supply and rising prices.

  Strengths

  Output is expected to increase about 20% to 30%, yet supply may still remain insufficient to meet demand; product mix is shifting toward higher-grade materials.

  Weaknesses

  Capacity expansion ability is limited and must be released in line with downstream AI project timing.

  Comparison

  Compared with the PCB segment, the CCL segment where TUC operates has slower supply growth and stronger pricing power.

  Risks

  CCL supply release faster than expected, downstream PCB expansion or AI orders below expectations.

• WUS Printed Circuit (002463.SZ)

  Preferred A-share PCB name, benefiting from technical capability, stable AI customer share, and relatively proactive capex.

  Strengths

  Strong high-end PCB technical capability, stable AI customer share, and active capacity expansion.

  Weaknesses

  Price increases in the PCB segment are mainly material cost pass-throughs, and margin upside is more limited than for upstream materials.

  Comparison

  Compared with CCL and electronic fiberglass cloth, PCB capacity expands the fastest and is less scarce.

  Risks

  Delayed AI demand, insufficient material cost pass-through, and intensifying competition due to industry capacity expansion.

• Delton Tech (001389.SZ)

  High-elasticity A-share PCB name with overseas AI PCB exposure.

  Strengths

  As a second-tier PCB supplier, its overseas AI PCB business provides higher beta.

  Weaknesses

  Compared with leading manufacturers, its customer mix, technical barriers, and capacity stability may require further validation.

  Comparison

  More elastic than large PCB leaders, but potentially with weaker certainty.

  Risks

  AI order wins below expectations, or slower-than-expected validation or ramp-up progress with overseas customers.

• Gold Circuit Electronics (2368.TW)

  Preferred Taiwan PCB name, benefiting from AWS ASIC and potential Google TPU demand.

  Strengths

  Strong Trainium 2/3 demand, copper foil upgrading from Gen 2 to Gen 4, and capacity expansion in Thailand, Taiwan, and China supporting ASIC demand.

  Weaknesses

  Trainium 4 contribution may remain limited before 2028.

  Comparison

  Benefits more directly from ASIC server demand than typical PCB makers, but still operates in PCB manufacturing, with less profit elasticity than upstream materials.

  Risks

  Delayed AWS ASIC ramp-up, weaker-than-expected realization of Google TPU demand, and capacity ramp risks.

• Tripod Technology (3044.TW)

  A relatively inexpensive Taiwan PCB beneficiary of CPU servers.

  Strengths

  Benefiting from recovery in general server demand, with industry checks showing general server shipments up about 20% year over year.

  Weaknesses

  Material upgrades for CPU servers mainly move from M6 to M7, and PCB layer count improvements are less significant than for AI accelerators.

  Comparison

  Valuation is relatively inexpensive, but the intensity of material specification upgrades is lower than in the GPU and TPU chains.

  Risks

  Weakening general server demand, limited material upgrade magnitude, and insufficient price pass-through.

Key data

• Report Date2026-06-10Citigroup Viewpoint report, 13 pages.
• PCB Specification UpgradeGB300 around 24 layers to Rubin close to 32 layers; Google TPU V8 around 36 to 44 layersNvidia Rubin and Google TPU upgrades are driving higher PCB layer counts and upgraded copper foil and fiberglass cloth specifications.
• CCL Supply GrowthIndustry supply growth this year may be only slightly above 20%Major suppliers have limited new capacity additions, which may lag downstream PCB expansion.
• CCL Price ChangesM7 and below up at least 20%; low-end M4 up about 30% to 40%; AI-related M8 up about 10% to 20%CCL makers are optimizing product mix and shifting capacity toward M7/M8 and higher-grade products.
• PCB Capacity Expansion CycleAbout 13 to 15 monthsFrom land preparation to meaningful revenue recognition, including construction, equipment installation, customer audits, and ramp-up.
• CCL Capacity Expansion CycleChina greenfield projects about 18 months; Southeast Asia projects about two yearsSpace at existing plants is limited, and most CCL makers have already planned 2028 greenfield expansions.
• Electronic Fiberglass Cloth Expansion CycleAbout 1.5 yearsAffected by high capex, high technical requirements, long qualification cycles, and loom supply constraints.
• Investment in Electronic Fiberglass Cloth ProjectsA 100kt electronic yarn/electronic fiberglass cloth project requires about RMB 3.7 billionInvestment intensity is about three to four times that of coarse yarn capacity.
• Toyota High-End Loom SupplyCurrent monthly production is about 200 units, possibly rising to 300 units by end-2027Loom supply is one of the key bottlenecks in expanding electronic fiberglass cloth capacity.
• China Jushi Electronic Fiberglass Cloth CapacityCurrently about 960 million meters, with two phases of the Huai’an 100kt project adding about 390 million metersTotal capacity may rise to about 1.35 billion meters this year, with another Huai’an project adding about 320 million meters by end-2027.
• China Jushi Thin Cloth MixAbout 15% rising to 30%Thin cloth gross margin can reach 40% or above, helping improve earnings quality.
• China Jushi Earnings SensitivityWhen 7628 electronic fiberglass cloth is priced at about RMB 7.4/meter, unit net profit may approach RMB 5/meterAt current prices, full-year earnings may reach about RMB 8.8 billion, implying about 17.7x PE based on the latest June 9 share price.

Impact & implications

The investment implication is that AI PCB upgrades are not just about PCB order growth, but about a repricing cycle driven by upstream material specification upgrades and supply constraints. PCB makers can pass through AI material costs, but have limited earnings leverage; CCL and electronic fiberglass cloth, with tighter supply, slower capacity expansion, and more direct price increases, may achieve greater profit elasticity. For portfolios, the report recommends watching CCL price hikes, continued upside in electronic fiberglass cloth prices, pre-reserved capacity by AI customers, and progress in China Jushi’s special AI electronic fiberglass cloth certification or volume supply.

Risks

• AI server ramp-up is slower than expected.
• Deployment of Nvidia Rubin, Google TPU, or LPU is delayed.
• Quartz cloth adoption is slower than expected.
• Qualification of new CCL or fiberglass cloth suppliers is faster than expected, weakening the pricing power of incumbent suppliers.
• General server demand is weaker than expected.
• Cost pass-through ability in non-AI PCB applications is limited.
• China Jushi’s special electronic fiberglass cloth certification is slower than expected.

What to watch

• Whether CCL sees another round of price increases in 2H 2026.
• Whether Google TPU, Nvidia Rubin, and LPU demand is stronger than expected.
• Whether PCB makers begin to face CCL allocation or supply assurance issues.
• Whether monthly price increases in electronic fiberglass cloth continue, especially for standard E-glass and Low CTE cloth.
• Whether China Jushi’s special AI electronic fiberglass cloth can obtain customer certification or achieve volume supply within the year.
• Whether AI customers continue locking in PCB and material capacity for 2027 to 2028 in advance.