compute

“Compute” is the shorthand the AI world uses for raw computing power: the chips, servers, data centres and electricity needed to train and run models. It has become one of the most important — and most fought-over — resources in technology.

The scale is hard to overstate. Training a frontier model requires vast numbers of specialised chips such as GPUs and TPUs, housed in data centres that can draw as much electricity as a small city. Labs now reserve this capacity years in advance; Anthropic, for example, locked in several gigawatts of computing power through deals with Google and Broadcom.

All this energy has a cost. Google’s own reports show its electricity use rising sharply because of AI — a quiet argument for “right-sizing” your AI and not reaching for a giant model when a smaller, local one would do.