TPU is an application specific chip (ASIC) designed by Google to make inferencing and training ML models faster and more efficient.