Term explained
What is a GPU and why does AI need them?
A GPU is a chip originally designed for graphics that performs huge numbers of simple calculations at once, making it the workhorse for training and running AI models.
In plain English
GPU stands for graphics processing unit. It was invented to draw video game images, a job that means doing the same simple arithmetic on millions of pixels simultaneously. It turns out neural networks need exactly that kind of work — vast numbers of multiplications and additions that can all happen in parallel.
The contrast is with a CPU, the general-purpose chip at the heart of a computer. A CPU has a few very capable cores that handle varied tasks one after another; a GPU has thousands of simpler cores working side by side. For AI, the GPU’s approach is dramatically faster. Specialised AI chips with other names exist too, but the principle is the same.
Why it matters
GPUs are the physical bottleneck of the AI industry. Training a large model can occupy thousands of them for weeks, and serving millions of users requires enormous fleets more. This has made advanced chips extraordinarily valuable, reshaped the fortunes of chipmakers, driven data centre construction and electricity demand, and turned chip exports into a matter of international policy. When you read about AI companies raising billions, much of it goes on this hardware and the power to run it.
An example
Asking one simple question of a chatbot is like having one person answer it. Training a model is like having a stadium full of people each do a tiny piece of a calculation at the same moment — that stadium is what a GPU cluster provides.
What to watch out for
GPUs are not magic accelerators for all software; they only help with work that can be split into many identical parallel pieces. Chip performance figures quoted by vendors are measured under ideal conditions and rarely match real workloads. And raw chip count tells you little without the memory, networking and power to use it well.