A specialised processor used for high-speed parallel computing. In AI data centres, GPUs are often deployed in large numbers and create high rack power densities and concentrated heat loads.