AI PCB & CCL Supply Chain: CCL Makers Hold Stronger Pricing Power Than Board Fabricators, M9 Materials Face Earliest Shortages
N.R. Finch
The AI-server bottleneck is spilling from GPUs into board-level interconnects — AI-PCB TAM rises from ~¥152 bn in 2026 to ¥562 bn by 2028. The slower a link expands capacity, the stronger its pricing power; CCL and e-glass will keep capturing chain profits.
Why has the bottleneck moved from chips to circuit boards?
Running an AI server takes more than a GPU. PCBs — the boards that carry and connect the chips — plus CCL (copper-clad laminate, the core raw material for making PCBs), e-glass fabric, and copper foil are all must-haves.
This means → these once-peripheral materials now directly gate whether Rubin, Google TPU, and 1.6T switches ship on time.
In plain terms = the chip is the engine, but without the chassis and wiring harness, the engine cannot be installed and the car cannot run.
How big is this market, and what drives the growth?
Citi's model puts AI-PCB TAM at ~¥152 bn in 2026, rising to ~¥562 bn by 2028 — nearly a quadrupling in three years.
The growth is not just "sell more boards." Layer counts jump from 22–24 to 36–44; materials move from M7 to M8; glass fabric migrates to Low Dk Gen 2. This means → per-board value rises in lockstep — spec upgrades matter more than volume gains.
By 2028, Google TPU-related PCB procurement may surpass Nvidia's GPU chain. The "top customer" is changing.
Why does CCL have stronger pricing power than PCB fabricators?
Capacity expansion speeds run in reverse order: PCB makers expand fastest, CCL next, e-glass and specialty glass fabric slowest — constrained by looms, process know-how, and qualification cycles.
In plain terms = the slower you can expand, the scarcer you stay, the more pricing power you hold. PCB makers' price hikes largely pass through material costs; CCL, e-glass, and high-end copper foil are the links that actually set their own prices.
This reflects a profit migration toward upstream: M8/M9 materials, Low Dk e-glass, HVLP foil — an ultra-smooth copper foil that cuts signal loss — and premium CCL suppliers will keep taking a bigger share.
With so many PCB names, how do you tell who earns real profit?
Treating every PCB company as the same trade is the easiest mistake to make at this stage. Profit quality differs sharply:
Shennan Circuits (沪电股份) wins on switch and high-end telecom PCB execution; Shenghong Tech (胜宏科技) wins on AI-server PCB and overseas-customer expansion leverage; Shengyi Technology (生益科技) and Kingboard Laminates (建滔积层板) lean on upstream CCL pricing power; TUC (台光电子) and Taiyao (台燿) are Taiwan-listed high-end CCL plays; Zhen Ding (臻鼎科技) strings HDI, mSAP — a modified semi-additive process for finer PCB traces — substrate, and optical-module boards into a platform.
This means → the stock-picking key is not the label "AI PCB" but where in the chain the profit actually sits — the closer to scarce materials, the harder the pricing power.
What is the biggest risk?
If CCL expansion, e-glass supply, M9 qualification, and PCB capacity ramps all land smoothly in H2 2026 through 2027, today's pricing-power thesis faces a cooldown.
If hyperscaler capex slows, if Rubin or TPU platforms are delayed, or if customers renegotiate, board-level-interconnect stocks will be the first to see multiple compression.
In plain terms = the next phase is not about how big the TAM is — it is about whether three lines on the financials check out: can gross margins hold, are orders locked in, and does post-expansion cash flow actually convert?
Content is for reference only, not financial advice.