Lumentum says AI laser capacity is sold out into 2029 as hyperscalers help fund expansion

InP laser chips have shifted from commoditized component to the binding constraint on AI interconnects. Hyperscalers co-funding fab expansion is not procurement—it is quasi-equity lock-in with no precedent in the transceiver value chain. Technical cascade: InP epiwafers are the atomic bottleneck for 800G/1.6T optical engines. Capacity booked through 2029 extends MOCVD order visibility (Aixtron, Veeco) past 2027, while module makers absorb a 30-50% chip premium. Silicon photonics deepens CW-DFB InP dependency at the 1.6T node; no viable substitute exists before 2026. Risk architecture: Customer-funded expansion blurs the buyer-investor boundary. If two or three accounts control 70%+ of output, pricing power migrates structurally to the buy side, compressing Lumentum's margin profile. InP substrate concentration (Sumitomo, JX) layered with export-control variables makes resilience a more urgent variable than raw capacity. Competitive response: Coherent will mirror the co-funding model within two quarters. Broadcom accelerates internal laser vertical integration. Chinese module makers face a pincer—upstream chip inflation against downstream price pressure—forcing upstream extension into InP epitaxy. 12-24 month outlook: InP lasers will replicate the HBM allocation logic of 2023. Hyperscalers will begin specifying chip-level parameters, creating a private-label tier. MOCVD capacity becomes the new allocation battleground, and the InP M&A window reopens in 2026.