What Is a GPU?

A GPU, or graphics processing unit, is a processor built to run a very large number of similar calculations at the same time.

Where a CPU has a modest number of powerful cores optimised for sequential logic, a GPU has thousands of simpler cores optimised for parallel arithmetic. That design was created for rendering images, but it maps almost perfectly onto the mathematics of neural networks, which is dominated by large matrix multiplications.

Modern data-center GPUs add three things that matter more than raw core counts: very high bandwidth memory close to the chip, specialised matrix units, and fast interconnects that let many GPUs behave like one larger machine. Training a frontier model is largely a networking and memory problem, not only a compute problem.

This is why GPU supply, memory availability and interconnect capacity now shape the economics of the entire AI industry.