Canonical and Google Cloud have announced the availability of certified Ubuntu images for Cloud TPU Virtual Machines. The images are offered by default when setting up TPU VMs and are intended to make it easier to run AI workloads on Google’s proprietary AI accelerators.
With this move, both companies aim to make the use of TPU infrastructure more accessible to developers and organizations already working with Ubuntu. Cloud TPU Virtual Machines combine standard computing power with Google’s specialized Tensor Processing Units (TPUs), which are used for AI training and inference.
According to Canonical and Google Cloud, the collaboration aligns with a shift within the AI market. While recent years have been dominated by training increasingly larger models, the focus is now shifting to inference: the actual execution of AI models for applications such as chatbots, AI agents, and automated workflows.
For such applications, general-purpose compute resources and AI accelerators must work closely together. By making Ubuntu available by default on TPU VMs, both parties aim to simplify the implementation of such environments.
Support for the latest TPU generation
Through the new images, developers gain direct access to Google’s latest TPU platform, Ironwood, also known as TPU7x. Additionally, older generations such as TPU v5 and Trillium (TPU v6) remain supported via Ubuntu 22.04 LTS, allowing existing environments to continue functioning without major modifications.
According to The Register, users of TPU v5 and v6 systems have been running on a Google-customized version of Ubuntu 22.04 until now. With the introduction of the certified images, support shifts to Canonical. As a result, these environments will fall under the regular Ubuntu LTS support model, without requiring any changes to existing production environments.
The Register also reports that TPU7x instances come standard with Ubuntu 24.04 LTS. According to Canonical, both Ubuntu 22.04 and Ubuntu 24.04 have been validated for multiple TPU generations, giving organizations greater flexibility in choosing their software environment.
Google Strengthens TPU Ecosystem
The announcement aligns with Google’s broader strategy to promote the use of its own AI hardware. Although Nvidia currently dominates the market for AI accelerators, Google has been investing for years in the development of TPUs as an alternative for both training and inference of AI models.
By officially certifying Ubuntu for Cloud TPU Virtual Machines, Google is lowering the barrier to entry for organizations looking to migrate existing Linux-based AI environments to TPU infrastructure. Administrators can continue to use familiar tools for management, automation, and orchestration, including Kubernetes.
The collaboration also includes optimizations for widely used AI frameworks such as PyTorch, TensorFlow, and JAX. Additionally, support is provided for Ray, a platform used to scale machine learning workloads across multiple systems.
Additional security features on the way
Later this year, Canonical and Google Cloud also plan to add support for Ubuntu Pro on Cloud TPUs. This will give users access to additional security features, such as live kernel patching and comprehensive security updates for open-source software packages.
These capabilities are expected to become available starting in the third quarter of 2026. According to The Register, organizations wishing to gain early access can contact Canonical or their Google Cloud account team. Ubuntu Pro offers, among other things, comprehensive security support for open-source packages and additional security measures for mission-critical AI environments.
Also read: Google presents TPU 8t and TPU 8i chips; splits training and inference