Markets

AI Data Needs Light, Not Copper: The Silicon Photonics Supercycle

Research Team · June 16, 2026

Copper cable cannot move data fast enough, cool cheaply enough, or scale reliably enough for 100,000-GPU AI clusters. Silicon photonics is the only structural fix. We examine four companies at different points in the optical supply chain, scored against the AIPI, and what their p

The bottleneck in AI infrastructure is no longer compute. It is the wire.

When a data center strings together 100,000 GPUs for next-generation LLM training, the copper cables connecting them become the single largest constraint in the system. Copper carries electrons. Electrons generate heat. Heat demands cooling. Cooling consumes power. Power consumption scales exponentially with the number of connected chips. At the density required for frontier AI training, the physics of copper networking simply stop working.