Infrastructure

Inside an AI rack

One liquid-cooled AI rack, and the five systems that keep it running: power, cooling, compute, network, and storage. Click a system to zoom in, or send a request to watch it flow. The numbers are illustrative.

Back to the AI industry stack

Five systems keep one AI rack running. Pick one, or send a request through the rack.

One liquid-cooled AI rack, drawn as a cutaway with the front door off. Power comes from the building through the UPS and an overhead whip into the rack's power shelf. Coolant leaves the CDU at 30 °C, climbs the supply riser, takes heat off cold plates on every GPU tray and comes back warmer on the return. The switch links the four GPU trays (32 accelerators), the storage shelf and the patch panel that leads out to the rest of the cluster. Numbers are illustrative, computed from a small model: rack power from tray load, feed current at 415 V three-phase, and return temperature from 60 L/min of coolant.