Light, Optics, and What a Cluster Can Actually Do
Every fast link in the stack, from subsea cable to in-rack fabric, ultimately runs on light, and the small devices that convert electricity to light and back are a genuine bottleneck in building AI infrastructure. Optical transceivers are consumed by the millions in a large cluster, and their supply, cost, and power draw increasingly constrain how fast fabrics can grow. Faraday's read: the humble transceiver is where the network's ambitions meet physical and manufacturing limits, and it deserves far more attention than it gets.
Knowledge check
Why are optical transceivers a meaningful constraint on the growth of AI cluster fabrics, rather than a mere commodity component?
Show the answer
Because a large cluster requires an enormous number of transceivers to connect its accelerators and switches, and each one converts between electrical and optical signals while consuming power and generating heat. Their manufacturing capacity, cost, and power draw scale with the fabric, so limits on transceiver supply, efficiency, or reliability directly cap how large and how fast a fabric can grow.