Layer 5 · 3661 — Alchip Technologies
Thesis
Alchip is the pure-play independent AI ASIC design house — the closest thing the public market has to a listed bet on "everyone wants their own chip." Q1 EPS NT$17.23 beat by 16.4%; 3nm production ramp is locked for Q2; AI/HPC is 83% of revenue. It's Tier B because AWS Trainium concentration is the elephant in the room and the N2 ramp is already priced in at NT$5,310.
What it does + financial print
Alchip's stack position is identical to GUC's at Layer 5 — ASIC design services bridging hyperscaler-owned IP and TSMC manufacturing. The differentiator is independence. GUC is a TSMC subsidiary; Alchip can theoretically migrate to Samsung Foundry or Intel 18A if the customer demands it. In practice, almost all of Alchip's leading-edge work runs through TSMC, but the optionality matters at the margin.
Q1 EPS came in at NT$17.23 versus a NT$14.80 consensus — a 16.4% surprise. Management confirmed 3nm AI accelerator production ramps in Q2 and flagged that N2 projects will tape out by end of CY26. The AWS Trainium relationship remains the single largest customer concentration in the book — somewhere north of 40% of revenue runs through that single program.
Bull case
What gets this to Tier A is customer diversification away from AWS Trainium plus a clean N2 tape-out without slippage. Alchip has been signing additional hyperscaler design wins quietly through the year; any disclosed second mega-customer (Meta, Microsoft, ByteDance) would re-rate the multiple cleanly.
Gap / bear case
AWS Trainium concentration is real, and the May 2026 commentary out of AWS on the slower-than-expected Trainium 2 ramp is exactly the kind of headline that hurts Alchip first. NT$5,310 prices in a smooth N2 transition; any tape-out slip would be felt.
Trigger to upgrade / downgrade
Upgrade trigger: disclosed second hyperscaler at >15% of revenue or clean N2 first-silicon news. Downgrade trigger: any AWS Trainium 3 timeline pushout from re:Invent or analyst-day disclosures.
The trade
- Entry zone: NT$4,900-5,150 - Stop: NT$4,500 (2×ATR; prior swing low) - Position size: 1% of NLV - Catalyst date: AWS re:Invent late November; Q2 print early August - Conviction: 6/10