From chips to datacenters: Scaling the AI fabric with light

AI is redefining the physical architecture of computing. As agentic and physical AI deploy at scale, traffic shifts from human-to-machine to machine-to-machine, with agents orchestrating agents and device fleets learning through the datacenter. The limiting factor is no longer compute alone, but moving data efficiently across chips, racks, and datacenters. This talk explores how photonics becomes the backbone of scalable AI infrastructure, with co-packaged and 3D-integrated optics reaching deeper into the package while pluggable optics extend the fabric across scale-out and scale-across networks. Together they are reshaping how accelerators compose into systems, how compute and memory disaggregate, and how AI infrastructure grows sustainably.