Alchip Technologies: Expanding Trainium4 TAM Drives 2028 Revenue Upside; 1Q26 Margin Beat Ahead

Core Conclusion

Alchip remains Overweight as the primary AWS Trainium design partner. The Trainium TAM is expanding meaningfully: we now forecast Trainium4 contributing US$4bn in 2028 (500k units at US$8k ASP), supported by a major US AI company's commitment to up to 5GW of new capacity spanning Trainium2 through Trainium4. Near-term, 1Q26 gross margin likely beat consensus at ~45% (vs 39%) due to higher NRE mix, though 2Q revenue guidance may miss. The stock trades at 25x our revised 2027e EPS, below its 5-year average of 29x, and we see 11% upside to our new Street-high PT of NT$5,088.

Market Underappreciation: Trainium4 TAM Scale

Consensus may underestimate Trainium4 revenue potential. The AWS commitment explicitly includes future generations, and Trainium3 is nearly fully subscribed with 30–40% better price/performance. Trainium4 is already reserved, implying mass production in 4Q27. Our US$4bn revenue in 2028 is double previous estimates. Additionally, Alchip’s role in Trainium4 (2nm, electrical I/O) and Trainium5 (1.4nm, optical I/O) solidifies its position versus Marvell/Broadcom.

Evidence Chain

1Q26 Preview: Preliminary revenue NT$4,192mn (-11% Q/Q). Gross margin guided to meaningful improvement; our estimate 45% vs consensus 39%. 2Q revenue likely >50% Q/Q but guidance may trail consensus of NT$9,336mn. 2H26 to account for >80% of full-year revenue, with monthly sales ramping sharply from June through August. The worst of the revenue decline is past.

Trainium TAM Growth: A major US AI company and Amazon expanded collaboration for up to 5GW of new computing capacity. Trainium3 started shipping in 1Q26, is nearly fully subscribed, and offers 30–40% better price performance than Trainium2. The commitment spans Graviton and Trainium2 through Trainium4 chips, with an option to purchase future generations. We model Trainium4 at 500k units × US$8k ASP = US$4bn revenue to Alchip in 2028.

Alchip’s Competitive Position: At the recent TSMC Technology Symposium, Alchip and its CPO partner Ayar Labs demonstrated a chip package with four I/O dies and eight optical engines using TSMC COUPE technology. For Trainium3, Marvell focuses more on 3nm wafer sourcing. Alchip is the major design service partner for Trainium4 (2nm, electrical I/O) and Trainium5 (1.4nm, optical I/O).

Revenue Build-up: 2026e total revenue US$2,978mn (+203% YoY), driven by AWS (US$2bn) and Li Auto (US$350mn). 2028e total revenue US$5,326mn. 2025–28 revenue CAGR of 75%, with AWS alone contributing 75% of turnkey revenue by 2028.

Key Risks

1. Customer Concentration: AWS accounts for ~64% of 2026e revenue. Any loss of share or push-out would severely impact estimates.
2. Competition: Broadcom, Marvell, GUC, and MediaTek all target AI ASIC. Sustaining design win pipeline is critical.
3. Near-term Execution: 2Q26 revenue guidance may miss consensus by ~28%. Delays in Trainium3 ramp could pressure sentiment.
4. Export Controls: US restrictions on Chinese customers may slow some projects, though offset by non-China demand.

Valuation / Trade Implication

We raise price target to NT$5,088 (from NT$4,388), based on a residual income model (10.4% cost of equity, 16% intermediate growth, 5% terminal growth). This implies 28x our 2027e EPS of NT$180.46. The stock currently trades at 25x 2027e EPS, well below the +1SD five-year average of 39x and below the 29x average. Given the 75% revenue CAGR, multiple compression appears unwarranted. Near-term catalyst: May 8 earnings call, where management likely discusses TAM upside and potential new wins (2nm Meta MTIA v5, US auto chip). Risk-reward is favorable: bull case NT$5,950 (33x 2027e), bear case NT$2,130 (12x).

Appendix: Key Estimates