A Google TPU (Tensor Processing Unit) is a chip Google designed specifically to speed up machine learning, available to others through Google Cloud.
In plain words
A TPU is built around matrix-multiply units, hardware that does the large grids of multiplications neural networks need. It has fast high-bandwidth memory on the chip, and many TPUs can be joined into slices for bigger jobs. Code for TPUs is compiled by XLA, which turns a model from frameworks such as JAX or PyTorch into TPU machine code.
A quick example
Imagine a researcher training a model written in JAX. They rent a TPU slice on Google Cloud, and XLA compiles the training step so the matrix units on every chip work on it together.
Where you'll meet it
- Training and serving large models on Google Cloud.
- Machine learning work written in JAX or PyTorch.
Related
Sources
- DocsIntroduction to Cloud TPU, Google Cloud · read 29 Sept 2026
- DocsTPU architecture, Google Cloud · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.