Servers and storage are the machines that fill a data centre's racks: the servers that run the apps, the storage systems that keep the data, and the AI clusters that train and run AI models. A rack is a tall metal frame, usually about 2 metres high, with 42 slots of 4.45 cm each. Each slot is called a rack unit, or U; the page on server racks explains it.
Consider one ordinary rack. It might hold 20 servers, each 2 units high, with a network switch at the top. A server is built from the same parts as a laptop: a processor, memory, disks and a network connection. But it has many more of each, and it keeps working when one part fails. Inside every server is a small second computer, the baseboard management controller (BMC), which lets engineers reach the server from far away.
Enterprise storage systems are servers packed mostly with disks. Their software keeps extra copies of the data, so that nothing is lost when a disk fails. Many servers read and write their data on one storage system over the network.
AI clusters and rack-scale systems are the newest machines here. An AI cluster joins thousands of GPU servers so that they work on one job, such as training an AI model. For AI, makers now sell whole racks that arrive already wired, cooled and networked.
Much of the engineering in this zone is software. The hardware makers write firmware, the software built into the hardware that starts it and controls it. A security chip, the hardware root of trust, checks that firmware before the server starts. The makers also write management software that lets one team run thousands of servers as if they were one machine; the page on server fleet management explains it. Much of this engineering is done in Bengaluru and Pune.