Bernstein: ABF Substrates Super-Cycle Initiated, Full Shortage Expected by 2027
Taylor Wilson
Bernstein sharply raises ABF substrate demand forecasts, projecting industry-wide shortage from 2027; substrates account for just ~1% of AI server cost, leaving room to raise prices with little pushback.
What are ABF substrates, and why do they matter now?
ABF substrates — the intermediate layer connecting a chip to the circuit board — are required in every server CPU and GPU, yet account for only ~1% of total AI server cost.
This means → they are cheap but irreplaceable. A substrate shortage stops the entire server from being built.
Bernstein sees the industry at the start of a super-cycle: demand CAGR of 22% from 2025 to 2028, versus supply CAGR of just 12%.
In plain terms = demand is growing nearly twice as fast as supply, and the gap only widens from here.
Why is demand surging this fast?
The core driver is server CPUs, which account for roughly 40% of ABF substrate demand.
Fueled by the rise of agentic AI, Bernstein projects 30 million server CPU shipments by 2030 — roughly 4× the 2026 level — representing a $137 billion market (Nvidia's own estimate is $200 billion).
Specs are scaling in lockstep: Intel's Diamond Rapids substrate is ~36% larger than its predecessor; Amazon's Graviton series has grown from 16 cores to 192 cores over five generations — bigger, more complex chips need bigger substrates.
GPU and ASIC accelerators add further pressure. Bernstein maintains that substrate value per chip doubles every generation, projecting a 33% CAGR in GPU/ASIC substrate area demand through 2028.
Why can't supply keep up?
Industry supply CAGR is just 12%, well below the 22% demand growth. The most aggressive expanders are Ibiden and Unimicron (欣兴电子).
Ibiden has committed ¥500 billion in capex over three years: ¥280 billion at its Ono plant for top GPU clients, with FY2027 capex alone at ¥210 billion — over 40% of the three-year total.
Unimicron guides 2026 capex at NT$34 billion, targeting a 40% year-on-year capacity increase by year-end, with further expansions through 2027-2028.
This reflects a fundamental mismatch: even the most aggressive builders cannot ramp capacity fast enough to match AI-driven demand growth.
How will the shortage unfold?
Bernstein maps the shortage in three phases:
Phase 1 (now): Upstream materials — especially T-glass — are already bottlenecked, with key suppliers running at 90%+ utilization.
Phase 2 (2026): Localized tightness hits the high-end segment; some suppliers have already begun raising prices on select products.
Phase 3 (2027-2028): Tightness spreads industry-wide. Utilization rates climb from 80-85% in 2025 to above 100% by 2027, giving suppliers broader pricing power.
What does this mean for investors?
The entire ABF substrate industry is roughly $10 billion in size — just ~1% of total AI server cost.
This means → downstream customers have very low sensitivity to substrate price increases. Paying an extra 1% to keep the entire production line running is an easy trade-off.
Put simply = substrate makers sit in a sweet spot — "inexpensive but indispensable." Low resistance to price hikes plus high profit leverage is the textbook signature of a super-cycle.
Content is for reference only, not financial advice.