Explain like I'm five · Cloud

What is an AI Data Center?

AI data centers, told as a story about a building annex constructed just for power-hungry GPU equipment.

GPU rackthis rack eats an enormous amount of power!an AI tenant has brought in equipment that draws enormous power

An AI tenant has brought in equipment that draws enormous power.A GPU rack draws far more power than ordinary office equipment.

the breaker keeps tripping, and the AC can't keep upordinary office power and AC can't handle GPU equipment

Ordinary office power and AC can't handle GPU equipment.The breaker trips, and the AC can't keep up with the racks' heat.

ordinary buildingAI-only annex: high power, heavy coolingthe annex gets its own power lines and cooling pipes, just for the GPU racks

An AI data center is an annex built just for the GPUs.The annex gets its own power lines and cooling pipes, just for the GPU racks.

Much higher power density than an ordinary buildingfar more power packed into each rack
Cools far harder than an ordinary buildingAC air isn't enough, so liquid cooling joins in
Runs an ultra-fast dedicated networkGPUs need to talk to each other fast
Built only where power is abundantfew places can supply that much power
ordinary building vs AI-only annexpower density: normal vs much highercooling: AC air vs even liquid coolingnetwork: normal wiring vs ultra-fast dedicated wiringwhere it can be built: almost anywhere vs only where power is abundant

On power, cooling, network, and siting, the AI-only annex demands far more than an ordinary building.That's why the annex can't be built just anywhere.

the annex costs a fortune, and local power falls shortan AI-only annex is costly to build, and the power it needs is hard to secure

An AI-only annex is costly to build, and the power it needs is hard to secure.Getting enough power can itself become the bottleneck that limits expansion.

In one breath

An AI data center is a dedicated annex with power, cooling, and networking all designed separately, just for power-hungry equipment like GPUs.

GPUs and accelerators used for AI training and inference draw far more power and generate far more heat than ordinary servers, requiring data centers purpose-built for high-density power delivery, liquid cooling, and low-latency high-speed interconnects between GPUs. Construction cost and securing enough power have become the new bottlenecks.

When grown-ups say it

AI data center
An annex built just for the GPUs. Power, cooling, and networking are all designed separately.
GPU / accelerator
The specialized equipment that draws enormous power. Runs AI training and inference faster than ordinary CPUs.
Power density
How much power is packed into one rack. Much higher in an AI-only annex than in an ordinary building.
Liquid cooling
A cooling method stronger than just AC air. Needed because GPU heat is too much for AC air alone.
Interconnect
The dedicated wiring GPUs use to talk to each other fast. Slow wiring here slows AI training down.
VM / Container
The two ways to divide space in an ordinary office. GPU equipment is sometimes divided up with VMs or containers too. → the private office and the partitioned desk
Scale-up / Scale-out
Making one office bigger, or renting more of them. AI annexes grow both ways: beefier machines (scale-up) and more of them (scale-out). → bigger office or more offices
Cloud service model
Bare shell / interior done / move-in ready. GPU infrastructure can be rented as IaaS too. → pick one to rent
Next story
Bricks, a Folder Tree, and Tagged Boxes →
Want another term as a picture book? Request →