Canonical has introduced a fully managed upstream Kubeflow service on Microsoft Azure that keeps organisations on the standard open source platform while removing operational overhead, preserving workload portability and keeping data within customer cloud environments.
Canonical has launched Managed Kubeflow on Microsoft Azure, bringing a fully managed deployment of upstream open source Kubeflow to enterprises without introducing a proprietary fork of the machine learning platform.
The managed service keeps customers on standard upstream Kubeflow while Canonical handles upgrades, security patches, version migrations, backups and operator maintenance. According to the company, the approach reduces the day-two operational burden of running Kubeflow without changing the underlying open source platform, allowing workloads to remain portable across environments.
The platform runs entirely inside a customer’s Azure tenancy, meaning data, models and training workloads never leave the organisation’s cloud environment. Canonical says this design supports compliance and data sovereignty requirements while giving organisations full control of their infrastructure.
The company positions the service as a solution to common Kubeflow operational challenges, including Istio configuration, Kubernetes API changes, storage provisioning, GPU scheduling and the complexity of integrating multiple open source components such as Katib, Pipelines, Notebooks and Central Dashboard.
Canonical added that the Azure deployment shares the same architecture as its on-premises OpenStack implementation, with managed services for additional public clouds planned. The platform supports generative AI and traditional machine learning workloads, includes an integrated MLflow server for experiment tracking and compliance logging, and offers Microsoft Entra ID integration with role-based access controls. Canonical says production-ready clusters can be deployed in under 30 minutes through Azure Marketplace.
