What's the difference between TPU vs. GPU?

Google’s Tensor Processing Unit (TPU) is a proprietary AI accelerator built for cloud-based machine-learning workloads in its data centers, using a systolic-array architecture that passes results directly between multiplier units, eliminating memory bottlenecks that GPUs encounter. In Pixel 11 smartphones the “TPU” functions as an on-device NPU, handling camera, image and local AI tasks while delivering up to 3.5× faster processing and 3.5× lower energy use compared with traditional NPUs.