why does AI need so many GPUs?
AI needs many GPUs because one chip solves problems one at a time, but many chips split the same work and finish it together. That's what a GPU is for and why they cost so much. Watch the difference yourself.
A CPU is built to do one hard thing at a time, in order. AI training and inference are millions of small, identical multiplications that can all happen at once — exactly the shape a GPU’s thousands of simple cores are built for. That parallel shape is also why GPUs, not CPUs, became the scarce, expensive resource behind the AI boom.