Run Ray on TPU, Part 1: The foundations

Ray 2.55 now treats Google Cloud TPUs as a first-class accelerator, with official images and supported APIs.

The post explains how GKE and Ray Core use topology-aware scheduling to place workloads on complete TPU slices.

This brings TPU training and inference into standard Ray workflows, reducing custom setup for users.

Existing GPU-oriented Ray code can more easily run on Google Cloud TPU hardware, including Ray Serve and JAX Trainer use cases.