Data Center Rack

Overview

The rack is the basic building block of a data center's physical layout. It is a metal frame of standard dimensions, almost always 19 inches wide internally and typically 42U tall, into which equipment is bolted in horizontal slots. One rack unit, 1U, is 44.45mm of height, so a "2U server" fills two of those slots. Everything in a facility, compute, storage, switching, and the cabling between them, is organized around this grid, which is what lets hardware from different vendors share the same infrastructure.

How it works

A rack is more than a shelf. It delivers power to each device through rack-mounted distribution units, carries the structured cabling for networking, and channels airflow so cool air reaches the front of every machine and hot exhaust is pulled away from the back. In a modern facility, racks are arranged in hot-aisle and cold-aisle rows so the two air streams never mix. How much power and cooling a single rack can support, its density, is what separates a general-purpose data hall from one built for AI.

Why it matters

Rack-level power and cooling limits set the ceiling on how much compute a facility can hold. A conventional rack might support a handful of kilowatts; an AI rack packed with GPUs can demand ten times that, and every extra kilowatt is heat that has to leave the same physical space. This is why AI data centers are effectively designed rack outward: fix how much power and cooling each rack can take, and the rest of the building follows.