Citi: AI PCB Bottleneck Shifting Upstream, E-Glass Fabric Has Strongest Pricing Leverage
0xBroomberg
Citi maps the AI PCB supply chain across three layers and concludes that pricing power is concentrating upstream toward e-glass fabric — the slower a segment can add capacity, the stronger its ability to raise prices.
What exactly is AI upgrading inside the PCB supply chain?
AI servers don't just need more boards — they drive simultaneous upgrades in layer count, laminate grade, copper-foil spec, and glass-fabric class.
From Nvidia's GB300 to Rubin, PCB layers rise from roughly 24 to near 32. Google's TPU V8 pushes harder — potentially jumping from 22–24 layers to 36–44.
This means → each board consumes multiples more high-end material, sending volume-and-price pressure up the chain layer by layer.
Who can add capacity fastest — and why does that decide pricing power?
PCB capacity leads — greenfield takes roughly 13–15 months, and makers are expanding aggressively. But most price increases are simple cost pass-throughs; margin upside is limited.
CCL (copper-clad laminate — the mid-layer that bonds copper foil to glass fabric) grows only about 20% this year. Key suppliers EMC, TUC, and Shengyi (生益科技) added no major new lines; Panasonic and Doosan won't deliver meaningful new supply until next year.
E-glass fabric responds slowest: a 100 kt project costs roughly RMB 3.7 bn, high-end looms come mainly from Toyota, current monthly loom output is only about 200 units, and that won't reach 300 until end-2027.
In plain terms = the further upstream you go, the more expensive, longer, and scarcer the expansion — so pricing leverage runs exactly opposite to capacity speed.
How far have e-glass fabric prices already moved?
Conventional e-glass fabric saw month-on-month price hikes from January through June. The June increase widened from RMB 0.3–0.5/m to roughly RMB 0.7/m.
The driver is capacity substitution: one Toyota loom produces over 200,000 m/year of standard thick fabric, but output drops to roughly one quarter when switched to AI-grade specialty fabric.
This means → every extra meter of AI fabric costs three meters of conventional output — conventional prices are rising not from a demand boom but because AI orders are crowding out capacity.
Does CCL follow the same pricing logic as e-glass fabric?
Not entirely. CCL hikes are already visible: M7 and below rose at least 20%; low-end M4 grades are up 30–40%.
Yet AI-linked high-end M8 laminate rose only about 10–20% — suppliers are protecting long-term customer relationships and resisting short-term markups.
Citi analyst Jack Chen sees a possible second round of hikes in H2 if AI customers prove willing to pay a premium to lock in supply.
Why does Citi name China Jushi as its top pick?
China Jushi's (中国巨石) current e-glass capacity is roughly 960 million meters. A Huai'an 100 kt project in two phases adds about 390 million m, bringing 2025 total capacity to roughly 1.35 billion m.
Citi estimates that at the current 7628-fabric price of about RMB 7.4/m, unit net profit could approach RMB 5/m, implying full-year earnings of roughly RMB 8.8 bn — about 17.7× P/E on the June 9 share price.
Specialty AI fabric is still at the customer-sampling stage; management targets roughly 15% long-term market share during the 15th Five-Year Plan period.
This reflects Citi's chain of logic: if H2 prices keep climbing and 7628 fabric breaks its 2021 peak, consensus earnings may be revised up — whether the current valuation already prices in that re-rating is the key test.
Content is for reference only, not financial advice.