Skip to main content
All terms
Hardware & Systems

Tensor Processing Unit

Google's custom AI chip built specifically for the matrix math of neural networks.

Definition

A Tensor Processing Unit (TPU) is a custom chip designed by Google to accelerate the matrix and tensor math at the core of deep learning. Built around a systolic-array architecture and high-bandwidth memory, it executes large matrix multiplications efficiently and integrates tightly with JAX and TensorFlow. Unlike general-purpose GPUs, TPUs are purpose-built for neural network arithmetic, and they are available through Google Cloud for both training and serving.