Commentary: AOI locks in China's InP substrates for AI optics, but export controls threaten supply

InP substrates are quietly becoming the strategic chokepoint of AI optical interconnects—functionally equivalent to silicon wafers in the logic chip world. AOI's rush to lock in Chinese InP supply is fundamentally a time-buying exercise against 18-month certification cycles and VGF crystal-growth ramp-up. On the technical stack, the 1.6T bottleneck sits not in packaging but in EML laser chips. InP's bandgap makes it the only viable III-V substrate at 100G-per-lane rates; silicon photonics can replace modulators but cannot eliminate the InP light source. An upstream cutoff means hollow shells without cores. The compliance risk is a materials-level embargo—more insidious than chip-level controls. Drawing on the 2023 gallium-germanium precedent, Western inventory buffers last 6–9 months while a new InP production line requires 18–24 months, creating an unbridgeable supply vacuum. Competitive response: Lumentum and Coherent will deepen prepayment commitments to Chinese suppliers while accelerating SiPh + CW-DFB architectures to dilute EML dependence. Over the next 12–24 months, expect InP price centers to shift up 30–50%, substrate BOM share to climb from under 5% to 10–15%, and a rare-earth-style de-single-sourcing scramble. But InP's lattice-matching and epitaxy moat dwarfs rare earths; full substitution is a 3–5 year minimum.