Goldman Sachs: ABF Substrate Supply Gap May Reach 51% by 2028
nashnova research
Goldman Sachs' supply-chain survey finds ABF substrate vacuum lamination equipment booked out to 2031; its model projects a supply gap reaching 51% by 2028 as AI chip packaging grows larger and the bottleneck shifts from capacity volume to large-format yield stability.
Equipment orders booked to 2031 — how tight is the squeeze?
Goldman's September survey of Taiwan's ABF substrate supply chain found that Eternal Precision — the key equipment maker — has order lead times stretching to 2031. Some customers are paying premiums for earlier slots.
This means → customers are not placing routine orders; they are reserving capacity years ahead — contract liabilities sit at roughly NT$800 million, with 10%–30% prepayments already collected.
Eternal Precision holds roughly 95% global share in mid-to-high-end ABF vacuum lamination equipment. Its August 2026 revenue hit about NT$360 million, up 91% month-on-month and 101% year-on-year.
In plain terms = there is essentially one supplier worldwide for high-end lamination gear, its backlog runs five to six years out, and customers are putting cash down just to hold a place in line.
Where does the "51% supply gap" number come from?
Goldman's model projects ABF substrate supply gaps of 14%, 34%, and 51% in H2 2026, 2027, and 2028 respectively.
This sequence rests on three assumptions — demand growth, capacity ramp, and yield — and is a model projection, not reported industry data.
On capacity: Eternal Precision's current annual output is 264 units, planned to reach 396 units by mid-2027 (roughly +50%). Goldman estimates full-utilization demand of about 1,500 units over 2027–2031, versus cumulative shipments of roughly 600 units over 2022–2025.
In plain terms = "51% gap" is the model's tightest scenario, not a stat the industry has already recorded — but equipment lead times and prepayments already point in the same direction.
Packages keep getting bigger — where exactly is the bottleneck?
JPMorgan's advanced-packaging seminar materials show AI chip package sizes advancing from roughly 85 × 85 mm toward 110 × 110 mm and even 130 × 140 mm, with roadmap discussion reaching about 9.5× reticle size by 2029.
When a GPU and multiple HBM dies — high-bandwidth memory chips — sit side by side, different materials expand at different rates under heat. Warpage worsens as size increases.
This means → the bottleneck is shifting from "how much capacity exists" to "whether large-format packages can stay flat and maintain stable interconnects" — tolerance for error in lamination, drilling, routing, and soldering is narrowing at every step.
In plain terms = chip packaging is like assembling an ever-larger jigsaw puzzle; the bigger it gets, the more heat makes it bend, and even a slight warp can break a signal path.
What is EMIB-T, and why are equipment and yield both still unproven?
EMIB-T — a technology that embeds a small silicon bridge into the substrate and routes signals and power through TSVs penetrating that bridge — is seen as more attractive for next-generation HBM4.
Eternal Precision says dedicated EMIB-T lamination equipment is planned for shipment to a Japanese customer in H1 2027. Goldman relays Unimicron's previously stated ~50% yield target and possible start of EMIB-T substrate volume shipments in 2027.
This means → equipment shipment, customer yield qualification, and stable volume production are three separate hurdles — receiving a machine does not equal mass production, and missing the yield target means product cannot ship.
Materials and drilling are chokepoints too — not just equipment?
The tightness in high-end ABF substrates is a multi-step bottleneck: lamination equipment, ABF film, low-CTE core material, and precision drilling all must qualify simultaneously.
High-grade T-type glass-fiber cloth is concentrated among very few suppliers, and leading makers are expanding cautiously. Stiffer core material helps control warpage but makes micro-via drilling harder — precision drill-bit supply and tool life become critical.
In plain terms = buying a lamination press does not guarantee output — film, core, drill bits, and equipment must all be in place, yields must ramp, and a gap in any one link stalls the whole line.
When can glass cores be used? And who delivers stable complex product first?
Glass core substrates — replacing traditional resin cores with glass for lower thermal expansion and better flatness at large sizes — are seen as the next-generation solution.
Eternal Precision sees little chance of large-scale commercialization in 2027–2028. Goldman places broad adoption after 2029. Seminar experts are more cautious still, suggesting volume production is unlikely before 2030.
This reflects a reality: between equipment scheduling and qualified substrate delivery lies a full chain of installation, material matching, customer validation, and yield ramp — whoever can stabilize complex-product delivery across this yield chain first holds the real competitive edge in this cycle.
市场有风险，内容仅供研究参考，不构成投资建议。