The AI boom is often framed as a story about chips and models, but its most enduring impact may lie elsewhere: in the infrastructure that moves data at scale. As hyperscalers race to build ever larger AI systems, the limiting factor is no longer just compute, but connectivity – between data centres, across facilities, and increasingly inside them.

Training clusters are being fragmented out of necessity, driven less by latency or sovereignty than by power availability. At the same time, AI architectures are decomposing internally, with accelerators no longer confined to a single board or chassis. Both trends expose the physical limits of copper interconnects, even over very short distances, and are pulling fibre deeper into the data centre and closer to compute.

This marks a quiet but profound shift. For decades, networking followed the maxim “copper if you can, fibre if you must”. AI is inverting that logic. Optical links, once reserved for long haul and metro networks, are now being considered at rack level and beyond, not because they are cheaper or simpler, but because they are the only viable way to scale bandwidth, power efficiency, and reliability.

The market response is telling. From optical line systems to coherent pluggables, and from co-packaged optics to fibre management, value is migrating into the interconnect layer. Companies like Ciena sit at the centre of this transition – not as AI vendors, but as enablers of the networks that AI increasingly depends on.