Canonical and Google Cloud have agreed to make Ubuntu the default image on Google Cloud’s TPU virtual machines. If you run training or inference on Google’s accelerators, you no longer have to assemble the system base yourself. When you create the VM in Compute Engine, you pick the TPU machine type and Ubuntu provisions on its own.
The change covers the TPU generations Google currently runs in production. There are certified images for the TPU 7x, the seventh generation known as Ironwood; for v6 (Trillium); and for v5p and v5e. The Ubuntu version mapping goes by generation: v5 and Trillium instances use Ubuntu 22.04 LTS, while the TPU 7x machines run newer releases.
Why it matters if you train or serve models
These images are built for real accelerator work rather than being a generic install you bolt extras onto later. They include optimized access to the usual frameworks (JAX, PyTorch and TensorFlow) and to tools like Ray for scaling machine learning workloads. You boot the VM and the stack is ready to go, with no fighting over drivers or mismatched versions.
Canonical keeps the image lightweight, with minimal system overhead. That pays off when you want your memory and cycles going to the model, not to the operating system itself. The images work with Kubernetes, Snap packages and standard MLOps tooling, so they fit into pipelines you already have.
Maintenance and the arrival of Ubuntu Pro
On support, these images give you up to five years of standard security maintenance through Ubuntu’s LTS cycle. That already covers most of the working life of a serious project.
The part that matters most for production is still on the way: Ubuntu Pro will be available on these VMs in the third quarter of 2026. Once it lands, it adds live kernel patching (livepatch) so you can apply fixes without rebooting, plus Expanded Security Maintenance (ESM), which extends coverage to more than 30,000 open-source packages for up to 15 years with the Legacy add-on. For anyone keeping models served around the clock, patching without taking the machine down is exactly what you need.
This collaboration follows what Canonical has been doing across other cloud and AI acceleration platforms, where Ubuntu aims to be the default base. If you want to see how this connects to the AI tuning Canonical did in the latest LTS, read our piece on the AI optimizations in Ubuntu 26.04 LTS.
Getting started takes no special recipe: create a VM in Compute Engine, select the TPU machine type, and Ubuntu is provisioned automatically.
Source
Based on Canonical’s official announcement: Canonical announces optimized Ubuntu images for TPU virtual machines by Google Cloud (ubuntu.com/blog). Credit to Canonical.