Layer 4 · 3037 — Unimicron Technology
Thesis
Taiwanese ABF substrate duopolist with Ibiden — the bottleneck that gates every Blackwell, every Rubin, every AWS Trainium 2 package. Q1 26 printed a profit-margin inflection that validates the structural-tightness thesis (NI +451% YoY, op margin 14% vs 3% YoY). Tier B not Tier A only because the stock is +263% trailing year and a -2.3% intraday on May 18 hints that the marginal buyer is finally tiring — the cleanest expression of the substrate bottleneck, but the entry has to be patient.
What it does + financial print
Unimicron sits at Layer 4 — the advanced packaging / substrate layer that sits underneath every leading-edge accelerator package. ABF (Ajinomoto Build-up Film) substrates are the printed-circuit-board-on-steroids that carry the GPU die and the HBM stacks on a CoWoS interposer; Ibiden and Unimicron are the only two suppliers who can hit the layer count (20+ build-up layers), the line/space (sub-15μm), and the yield required for AI accelerator packages. Samsung Electro-Mechanics is the credible third source but is still ramping. Capacity is sold out through 2026 and the lead times have stretched from six months to nine-plus.
Q1 FY26 printed NT$37.4B revenue (+25% YoY), EPS NT$3.00 vs NT$0.60 YoY (a massive beat — consensus had ~NT$1.80), net income +451% YoY. Operating margin expanded to 14% from 3% — that's the operating-leverage signature of a capacity-constrained business taking price. Guidance described capacity as "sold through" — management code for "we are pricing what we can ship." Stock at NT$802 on May 18, ~5% off the recent high.
Bull case
ABF substrate tightness persists through 2027 because the only meaningful capacity adds (Ibiden's Ogaki North, Unimicron's Taoyuan expansions, SEMCO's Sejong line) all come online into a demand curve that grows faster — every Blackwell→Rubin→Vera-Rubin Ultra generation increases substrate area per package. Unimicron at sub-25x forward earnings is materially cheap if you believe the cycle has another 18-24 months. The optionality on top: CoPoS / glass substrate platform sets up the next-decade tail with TSMC as the named partner.
Gap / bear case
+263% TTM is the gap — this is a stock that has already priced the entire near-term substrate-tightness narrative. The -2.3% intra-day on May 18 isn't a thesis-break but it is a positioning warning: when a Taiwanese substrate name with this much momentum starts trading heavy intraday, the next test is whether NT$750 holds. If it doesn't, the rerating from "structural tightness" back to "cyclical winner" can take 20-30% off the multiple before anything in the fundamentals changes. Secondary risk: if ABF substrate yields improve faster than expected at the third-source (SEMCO), the duopoly premium compresses.
Trigger to upgrade / downgrade
Upgrade to Tier A on (a) a clean break and hold above NT$830 with confirmation of Q2 revenue continuing >+20% YoY, AND (b) explicit AI/HPC substrate revenue disclosure at >50% of mix. Downgrade on loss of NT$720 close (50-day proxy) — that level marks the line between healthy consolidation and trend break.
The trade
- Entry zone: NT$740-780 on pullback
- Stop: NT$700 close (50-day MA)
- Position size: 1.5% NLV (already owned 10K sh; treat as add-on at zone)
- Catalyst date: Q2 print July 29 2026
- Conviction: 7/10