Global AI Transceiver Demand: Supply Constraints Masking Structural Up-cycle; Upward Revisions Insufficient

Core Conclusion

Global AI transceiver demand is accelerating faster than consensus expects, driven by hyperscaler capex growth and 1.6T adoption gaining traction earlier than modeled. Current supply-side bottlenecks obscure the true demand trajectory, making upward revisions to 2026-2028 shipments—now projecting >4x TAM expansion to $102B by 2028—only partially priced into equities. The key opportunity lies in companies with outsized exposure to 1.6T ramp and proven share gains, while those tied to legacy telecom or supply-constrained components face structural headwinds.

What the Market Likely Underestimates

The market underestimates the durability and velocity of the AI transceiver up-cycle. Three factors are not fully discounted: 1) 1.6T transceiver adoption is accelerating far ahead of prior cycles, with Morgan Stanley raising its 2027 shipment forecast from 71M to 141M units – a 98% upward revision; 2) component shortages, especially EML lasers and pump lasers, are constraining output, masking actual end-demand strength; 3) hyperscaler capex is set to expand 60% YoY in 2026, with Amazon ($200B), Google ($180B), and Meta ($125B) all accelerating spend. The implied 2026-2028 CAGR of 43% (base case) is not captured in current forward P/E multiples.

Evidence Chain

1. Demand signals are unequivocal and recent. Lumentum's CEO stated in April 2026 that hyperscaler demand is accelerating and the company "could be sold out for two quarters through 2028." It expects InP optical channel capacity demand to compound at ~85% CAGR through 2030. Coherent plans to double InP capacity over two years; Sumitomo is expanding 100% in 2024-2026 and another 40% in 2026-2028. Broadcom is undertaking "significant" capacity expansion. These are not isolated signals—they form a consistent pattern.

2. Shipment forecasts are being materially revised upward. Morgan Stanley's new base case for 800G+1.6T shipments: 73M units in 2026 (vs. 53M previously), 141M in 2027 (vs. 71M), and 150M in 2028 (vs. 80M). The 2026 shipment growth is projected at >200% YoY, versus LightCounting's 65%. This gap suggests consensus remains too conservative. The TAM is forecast to grow from $18B in 2025 to $102B in 2028, a >4x expansion.

3. Company-level evidence confirms demand pull. AOI received its first mass-production order for 1.6T transceivers in March 2026, valued at >$200M, followed by an $124M order for 800G from the same customer. Eoptolink's 1Q26 results beat Morgan Stanley estimates by 17% on revenue and 23% on net profit. Its 12-month forward P/E has expanded from 8x to 22x since early 2025, reflecting accelerating earnings power.

4. Supply-side constraints are the binding choke point, not demand. Fabrinet indicated that without supply limitations, data center revenue would be at record levels. Shortages span lasers, memory, and certain ASICs. Lumentum's pricing power—evidenced by recent margin performance—confirms the supply-constrained environment. Morgan Stanley views laser/pump laser supply as tighter than transceiver assembly itself.

Key Divergences and Risks

1. CPO disruption is the single largest downside risk for incumbent transceiver suppliers. If co-packaged optics achieves manufacturing breakthroughs and scales by 2027, the current pluggable transceiver architecture could face structural obsolescence. Morgan Stanley's probability-weighted valuation assigns 50% base case, 30% bull case (CPO delayed), and 20% bear case (CPO early), implying CPO is not fully discounted in base-case earnings.

2. Component shortages could constrain 2026-2028 volume ramp. The rapid capacity expansion by Lumentum, Coherent, and Sumitomo may not keep pace with demand, particularly for EML lasers and pump lasers. Any delay in capacity additions would cap transceiver shipments and revenue for suppliers.

3. Competitive dynamics vary sharply by company. Eoptolink (96% overseas revenue) is positioned as a share gainer in 800G/1.6T, with aggressive upward revisions to 2027-2028 net profit (+74% and +72% respectively). Accelink, despite earnings upgrades, remains Underweight due to 70% domestic exposure, weak telecom capex, and lack of US hyperscaler customers. TFC's 1Q26 miss—revenue 30% below estimates—offsets the positive demand backdrop, leaving 2026 forecasts unchanged.

Valuation or Trading Implications

The probability-weighted residual income valuation yields intrinsic values of RMB 710/share for Eoptolink (target up 54% from RMB 460), RMB 371 for TFC (unchanged), and RMB 166 for Accelink (up 177% from RMB 60). The wide dispersion reflects divergent earnings trajectories and risk profiles.

Eoptolink is the clear structural winner: its earnings revisions are the most aggressive, its overseas exposure insulates it from domestic competition, and its current forward P/E of 22x remains below its historical mean of 15.8x when adjusting for the accelerating earnings growth trajectory. The CPO risk is well-flagged and likely already priced into debate.

TFC faces near-term execution headwinds from new low-margin production lines, and its stock trades above +1 standard deviation of its historical mean. Accelink's lower-quality earnings base and domestic market risks justify its Underweight rating, despite the dramatic target price revision.

Investment conclusion: Overweight Eoptolink as the most direct pure play on the 1.6T cycle with strong earnings momentum and manageable CPO risk. Avoid TFC and Accelink until execution improves or CPO clarity emerges.

Appendix Data Summary

Shipment Forecast Comparison (Morgan Stanley vs. Consensus)