AP Memory Technology Corp: Too early to be concerned about Si-Cap competition; OW

Core Conclusion

AP Memory (6531.TW) remains Overweight with an NT$1,555 price target (78% upside from NT$876). The market has overreacted to the SEMCO Si-Cap contract announcement, mispricing three factors: a structural non-TSM silicon capacitor supply shortage through 2028, the near-term VHM catalyst, and the extended qualification timeline required for SEMCO to gain meaningful share.

What the Market Is Underestimating

The market fixates on SEMCO's ~US$1bn contract with Winbond as an existential threat to AP Memory's Si-Cap franchise. This ignores that (1) non-TSM silicon capacitor supply shortage is projected at 57% in 2026, 16% in 2027, and 11% in 2028 — meaning the pie is expanding faster than any single competitor can fill; (2) VHM (Vertical Hybrid Memory) represents a largely consensus-unmodeled revenue stream with a clearer near-term path than Si-Cap; and (3) SEMCO's 3D stacking capacity via Winbond is only 4kwpm in 2027-2028, versus AP Memory/PSMC's 4-8kwpm ramp starting 2026, with qualification requirements spanning substrate vendors, EMIB integration, and end customers.

Evidence Chain

1. Non-TSM Si-Cap supply shortage is structural and multi-year Demand drivers include EMIB (TPU v10/v11, potential auto chips), China AI GPUs, and smartphone SoC. Total wafer demand reaches 160.9k units by 2028, against a global non-TSM supply of only 144k units. The deficit is 57% in 2026, falling to 16% in 2027 and 11% in 2028, but even the latter implies pricing power remains intact. The supply base remains concentrated: PSMC/AP Memory's 3D stacking capacity scales from 4kwpm in 2026 to 8kwpm in 2028 (48k/72k/96k wafers per year), while Winbond/SEMCO adds only 4kwpm in 2027 and maintains 4kwpm in 2028 (48k wafers per year each year). The incremental supply from SEMCO is insufficient to close the gap.

2. SEMCO's competitive threat is back-end loaded and overestimated SEMCO's Si-Cap technology also uses stacking and targets EMIB, similar to AP Memory. The advantage of Winbond's foundry process is real, but the timeline for equipment procurement, qualification with substrate vendors, EMIB integration, and end-customer validation is measured in years. The effective capacity ramp is 4kwpm by 2027, half of AP Memory/PSMC's 8kwpm base by 2028. Even if SEMCO captures some share, the addressable market is growing so rapidly that absolute volumes for AP Memory can still expand. The market is pricing in a winner-takes-most outcome that the supply/demand math does not support.

3. VHM is a near-term revenue catalyst not in consensus estimates Beyond Si-Cap, AP Memory's VHM technology addresses a different memory architecture need and has clearer near-term commercialization. No current consensus model includes VHM revenue. The upside from this product alone could drive earnings revisions without requiring Si-Cap share gains. The combination of Si-Cap shortage pricing and VHM ramp creates a compound catalyst that the market has not priced.

Key Risks

• Weaker-than-expected consumer demand slowing Si-Cap adoption across EMIB, China AI GPU, and smartphone SoC end markets.
• Slower-than-expected IPD or VHM ramp and WoW packaging development, delaying revenue inflection.
• Intensifying competition from foundry partners, memory houses, GPU vendors, or other design houses like SEMCO with faster certification or broader capacity.

Valuation & Trade Implication

Maintain Overweight. The NT$1,555 price target is derived from a residual income model (9.2% cost of equity, 3% terminal growth, 16% medium-term growth). Current NT$876 implies 78% upside. Accumulate on weakness. Near-term catalysts include VHM revenue updates and continued evidence of Si-Cap supply tightness. The stock trades at 49.6x 2026e P/E and 25.5x 2027e P/E, which becomes 14.4x 2028e P/E — pricing in zero terminal value if the competitive thesis were fully correct. The structural shortage and VHM opportunity invalidate that assumption.

Appendix: Silicon Capacitor Supply/Demand Summary (Non-TSM, wafers in thousands)