AI · Pillar guide · 2026-06-01

AI's photonics bottleneck: why moving data is now harder than making chips

AI clusters are running out of room on copper. As back-end fabrics climb from 800G to 1.6T and 3.2T, the constraint stops being how many GPUs you can buy and becomes how cheaply you can move data between them. The case for optics as the next chokepoint, and a map of where it bites first.

Dylan Austin Bristot, Founder and publisher, AI Bottlenecks

The cluster is becoming an optical system

AI's photonics bottleneck appears when moving data between accelerators, racks and sites becomes harder than adding more compute. Copper remains useful at short reach, but density, loss, signal integrity and power become punishing as fabrics move from 800G to 1.6T and beyond.

Light moves the constraint

Photonics does not erase scarcity. It relocates it into lasers, silicon photonics wafers, photonic packaging, electro-optical test, fiber attach, connectors and thermal serviceability. A faster transceiver is only useful when the complete optical path can be manufactured and repaired at cluster scale.

Pluggables and co-packaged optics

Co-packaged optics shortens the high-speed electrical path by moving the optical engine closer to the switch or accelerator. It will coexist with pluggable modules because the service model, packaging and installed base move at different speeds. The real stack is hybrid, with architecture deciding where optical value pools.

What to watch

Track qualified laser capacity, 1.6T module yield, photonic packaging and test throughput, fiber-attach automation, connector reliability and named production deployments. Announcements matter less than accepted volume and field uptime.