Neural networks
A GPU (graphics processing unit) is a chip designed to do thousands of simple calculations at once, which makes it ideal for training and running AI models.
GPUs were built for video games, where millions of pixels need updating every frame. That requires doing the same simple arithmetic on huge amounts of data in parallel. Neural networks need exactly that: enormous grids of numbers multiplied and added. A GPU can do this many times faster than a general-purpose processor.
Around 2012, researchers showed that training deep networks on gaming GPUs made previously impossible experiments practical. That discovery launched the modern AI boom. Today, training a frontier model can use tens of thousands of specialized GPUs running for months, and access to them is a major cost.
The dominant maker is Nvidia, though Google, AMD, and others build competing AI chips. When people mention 'compute' as a limit on AI, they mostly mean GPU supply. Running a model on your own laptop is possible if your GPU has enough memory to hold the model's weights.
The same graphics card that renders a video game at 60 frames per second can train a small image classifier in an afternoon.