Goldman Sachs TPCA Survey: TUC Enters ABF Substrate CCL; Customer-Funded Capacity Lock-In Spreads to Materials Chain
nashnova research
Goldman Sachs' TPCA show visit uncovered a first: end-customers are directly funding capacity expansion at CCL and copper-foil makers — a sign they expect material supply to get tighter, not looser, over the next several years.
Why are customers paying out of pocket to help materials makers expand?
Goldman visited four companies and spoke with over ten supply-chain firms at the Taipei TPCA show on October 21. The key finding: end-customers have begun directly funding capacity expansion at materials makers — via long-term agreements, expansion subsidies, prepayments, and incentive fees.
This means → customers are putting real money down to lock capacity, signaling they expect supply to be tighter in the years ahead, not looser.
This arrangement is spreading from ABF substrates — a type of high-end chip-packaging board — down into CCL (copper-clad laminate, the core raw material of PCBs) and copper foil.
How does customer funding improve materials makers' margins?
Goldman sees two profit-improvement paths: first, depreciation drops — CCL makers' depreciation-to-revenue ratio falls from about 1.9% in 2025 to about 1.2% by 2028; second, subsidies and prepayments are gradually recognized as income.
In plain terms = when a customer pays for your factory, your depreciation burden shrinks, and the subsidy itself turns into revenue on your income statement — a double benefit.
Co-Tech (金居) will begin recognizing customer subsidies as early as 2027. Goldman has raised earnings estimates for the affected companies and sees room for further re-rating as more capacity gets pre-booked.
When does M10 CCL ship — and why is the PTFE route stuck?
TUC's (台光電) M10 CCL is expected to ship in H2 2027 for a specific AI customer's GPU rack design, with meaningful volume in 2028 as more GPU and ASIC programs adopt it. Ventec's (台燿) M10 ships from 2028 for 3.2T Ethernet switches.
The price jump is striking: TUC's and Ventec's 2026 average selling prices are roughly $63 and $46 per sheet, while M10-grade CCL is priced at about $800 per sheet — even a small share of shipments will lift the blended ASP materially.
PTFE — a low-loss, high-frequency material — based PCB yields sit at just 10–15%, versus 80–90% for M9-grade CCL using Q-glass. This means → Nvidia's Rubin Ultra switch-tray PCB upgrade will likely use M9/M10 with Q-glass or hydrocarbon resin, not PTFE. PTFE mass production is unlikely before late 2028 at the earliest.
Can TUC really capture half of global ABF-substrate CCL capacity?
ABF-substrate CCL sells for about $1,500 per sheet, versus just $130 and $180 for M7- and M8-grade CCL. Goldman estimates ABF-substrate CCL gross margins currently exceed 50% and could converge toward the industry average of 70%+ over time.
TUC entered the ABF-substrate CCL market in early 2026. Monthly revenue from the highest-end TPU substrate CCL has reached NT$150–200 million — still under 1% of TUC's September revenue. Goldman expects TUC to start gaining share from H2 2026, reaching 20–30% by 2028.
The capacity ramp is equally aggressive: TUC indicated that by end-2027, monthly ABF-substrate CCL capacity could reach 1.5 million sheets — four to five times the current 300,000 — which Goldman estimates would represent 50% of global capacity.
How big can the ABF-substrate CCL market actually get?
Goldman sizes the addressable market at $0.8–1.2 billion in 2026, growing to $2.0–2.5 billion in 2027 and $4.0–4.5 billion in 2028.
There is an under-appreciated demand multiplier: under 2028 substrate designs, a quarter-sheet of ABF-substrate CCL may yield only 2–3 high-end ABF substrates, versus 6–8 under current specs — the CCL area consumed per substrate more than doubles.
In plain terms = substrates are getting larger and more complex, so fewer of them can be cut from the same sheet of CCL. Even if substrate shipment volumes stay flat, CCL demand accelerates.
Will CPO undercut demand for high-end CCL?
Goldman sees limited near-term substitution risk from co-packaged optics (CPO — mounting optical modules directly next to the chip to shorten the electrical-to-optical conversion path). The industry is unlikely to adopt CPO in GPU server switch trays before 2028 at the earliest.
Even if all switch boards were then downgraded to M4 or M7-grade CCL, the impact on the 2028 CCL market would be less than 10%.
As long as pluggable optical modules remain the mainstream scale-out approach, switch-board layer counts and design complexity will keep rising. Goldman forecasts the optical-module mSAP — modified semi-additive process, a finer PCB circuit-fabrication method — market to grow at an 87% CAGR from 2026 to 2028, from about $2 billion to about $7 billion. Zhen Ding (臻鼎) plans to expand mSAP capacity five-fold within two years.
市场有风险，内容仅供研究参考，不构成投资建议。