Lumentum is undershipping customer orders by 30% as AI infrastructure demand outpaces optical component production capacity. The company's electroabsorption modulated laser (EML) capacity is fully committed through contracts extending to the end of 2027.
The supply gap is widening despite aggressive capacity expansion. Lumentum increased indium phosphide manufacturing capacity by over 20% in Q4 2025 alone, with additional expansions planned. Demand is growing faster than new production lines can come online.
EMLs convert electrical signals to optical light for data transmission in AI clusters. Every major AI networking deployment relies on these components to connect GPUs and transfer training data between servers. Meta, Microsoft, Google, and Amazon are building facilities requiring millions of optical transceivers.
Optical circuit switch (OCS) orders reflect the capacity crunch. Lumentum's OCS backlog surged past $400 million, with most units slated for shipment in H2 2026. These switches route optical signals between servers without electrical conversion, reducing latency in AI training workloads.
The bottleneck affects data center construction timelines. Hyperscalers cannot fully deploy new AI clusters without optical transceivers and switches. Lead times for optical components have stretched from weeks to quarters as suppliers prioritize existing commitments.
Long-term supply agreements lock in customers but limit spot market availability. Lumentum has allocated its entire EML production capacity through multi-year contracts, leaving no room for new customers or expanded orders from existing ones. Companies without secured supply face extended waits.
The shortage extends beyond Lumentum. Competitors face similar constraints as indium phosphide wafer supply, clean room capacity, and specialized manufacturing equipment limit industry-wide production. Adding new fabrication capacity requires 18-24 months and significant capital investment.
Industry analysts expect the supply-demand imbalance to persist through 2027. Each wave of AI infrastructure buildout requires exponentially more optical components as cluster sizes grow from thousands to millions of GPUs. Current expansion plans will close only part of the 30% supply gap.