Ray gains official TPU support on Google Cloud

Ray 2.55 now treats Google Cloud TPUs as a firstclass accelerator, with official APIs, prebuilt images, and support across the core libraries. The update lets teams that already use Ray on GPUs run similar Python workloads on TPU through Google Kubernetes Engine (GKE). The article explains that TPU chips are organized into slices, which must stay intact so workers can communicate over the interconnect. GKE provisions and labels those slices, while Ray Core uses those labels to reserve an entire slice at once, helping ensure that distributed jobs are scheduled correctly. Google also describes how the Ray Operator addon installs KubeRay and a TPU webhook on GKE. Users can request TPU resources in RayCluster manifests by specifying the TPU generation, topology, and host count, and Ray AI libraries such as Data, Train, and Serve can use the TPU support without requiring custom placement code.