AI is advancing exponentially. The models driving breakthroughs in science, medicine, and human knowledge require unprecedented computational scale—100,000+ GPUs training in coherent clusters.
But these systems are bandwidth-starved. GPU utilization sits at 40-60% because chips spend more time waiting for data than computing.
The interconnect is the bottleneck.
Lightmatter builds photonic interconnects that eliminate this constraint. Our 3D integrated photonic circuits deliver 114 Tbps bandwidth—the industry’s highest—enabling frontier AI models to train 8x faster.
We’re not incrementally improving existing technology. We’re creating a fundamentally new architecture—edgeless I/O—that will scale AI infrastructure for the next decade.
