Lumentum CEO Michael Hurlston warned at the RAISE Summit that indium phosphide—the compound semiconductor critical for every high-speed laser in AI data center transceivers—faces a supply crunch worse than the DRAM shortage. Despite operating five fabs, Lumentum ships more than 30% below customer demand, with supply-demand imbalance widening. Hurlston said telecom customers once bought lasers in hundreds; now hyperscalers demand hundreds of millions.
The bottleneck stems from silicon's inability to emit light; every silicon photonic chip, co-packaged optic, and transceiver still requires indium phosphide lasers. AI clusters demand 1.6 Tbps optical interconnects, each 800G+ link needing multiple InP-based electro-absorption modulated (EML) lasers per rack. Nvidia's March decision to write $2 billion checks to both Lumentum and Coherent reflected the crisis. Indium itself is geopolitically constrained: China supplies ~69% of global refined indium, and placed export controls on the metal in February 2025, halting shipments until June 2025.
For architects, this is not a temporary chip delay. Ramping from thousands to millions of InP wafers in a material with multi-year fab timelines means capacity crunches extend through 2027 at minimum. Hyperscaler designs that depend on co-packaged optics or high-speed EML transceivers face allocation risk. Alternative paths (silicon photonics alone, copper scale-up) don't eliminate the InP dependency; they embed it deeper. Lumentum guidance of 35–36% margins and $960M–1.01B Q4 revenue shows pricing power, but customers queuing for allocation should plan for sustained scarcity.