GPU (Graphics Processing Unit) is a type of computer chip, originally designed for rendering graphics, that turned out to be exceptionally good at the calculations behind AI. GPUs are the workhorse hardware of modern machine learning.
What it means in plain English
Training and running AI models involves doing enormous numbers of simple mathematical operations in parallel. GPUs were built to do exactly that for graphics, and it happens the same kind of parallel maths powers neural networks. This coincidence is a big reason deep learning took off: GPUs made it practical to train large models in reasonable time. Today, AI progress is closely tied to the availability of powerful GPUs.
Specialised chips like TPUs have since been designed specifically for AI, but GPUs remain the most widely used.
A simple example
Training a large language model can require thousands of GPUs running for weeks. Even running a model to answer your questions typically happens on GPUs in a data centre, because they process the maths far faster than an ordinary computer chip.
Why it matters
GPUs are the physical foundation of the AI boom. The demand for them, and the companies that make them, have become central to the whole industry — which is why AI hardware is now a topic of global economic importance.
Related terms
- TPU — a chip designed specifically for AI.
- Deep Learning — the field GPUs made practical.
- Inference — often run on GPUs.
Frequently asked questions
Why are GPUs used for AI?
GPUs can perform many calculations in parallel, which suits the matrix maths at the heart of neural networks, making them far faster than CPUs for training and running AI models.
Do you always need a GPU for AI?
For training large models and heavy workloads, yes — but many smaller models and inference tasks can run on CPUs or specialised chips, and cloud GPUs let you rent power as needed.
