On July 9, 2026, Canonical made its fully managed Kubeflow service available through the Azure Marketplace. Unlike a typical SaaS offering, this deployment model runs the entire platform inside the customer’s own Azure tenancy.

The move gives organizations a way to get a production-grade, open-source MLOps platform without the ongoing operational overhead—while still keeping data, code, and compute resources under their direct control and compliance boundary.

What actually changed

Canonical Managed Kubeflow is now listed as a transactable offer in the Azure Marketplace. When you subscribe, the service provisions Kubeflow into your Azure subscription, typically on top of an Azure Kubernetes Service (AKS) cluster you provide or that is provisioned as part of the deployment.

Here’s the concrete breakdown:

  • Deployment model: The entire Kubeflow stack—including notebooks, pipelines, KFServing, Katib, and the dashboard—runs inside your Azure tenant. Canonical manages the Kubeflow components, but the underlying virtual machines, networking, and storage remain your resources.
  • Billing: You pay Canonical through the Azure Marketplace for the management service; Azure infrastructure costs are billed separately and directly to your Azure account.
  • Identity integration: The service integrates with Microsoft Entra ID (formerly Azure Active Directory) for authentication and role-based access control. So your teams can use existing corporate credentials and conditional access policies.
  • Day-2 operations: Canonical handles upgrades, patching, security updates, and monitoring of the Kubeflow control plane and workloads, presumably through a secured operator that runs in your cluster.

This is different from the managed Kubeflow pipelines available in Vertex AI or Azure Machine Learning. Those are proprietary abstractions over Kubeflow. Canonical’s offering is native Kubeflow—the standard open-source distribution—without vendor lock-in.

What it means for you

For data scientists and ML engineers

You get a Kubernetes-native MLOps environment where you can spin up Jupyter notebooks, define and run training pipelines, tune hyperparameters, and serve models using the tools you’re already familiar with. No more filing IT tickets to set up Kubeflow or waiting for days to get a secure, compliant environment.The platform is fully armed with the latest Kubeflow release and is automatically patched, so you can focus on model development.

For IT administrators and platform owners

This solves a real governance headache. Because the stack runs in your tenant, you can apply Azure Policy to the managed resource group, enforce network security groups, integrate with your existing naming conventions, and monitor everything with Azure Monitor. Billing flows through your existing Azure commitment, and you get a single pane of glass for compliance and cost management.

Crucially, the Entra ID integration means you can enforce multi-factor authentication, session durations, and conditional access rules for Kubeflow—something that’s complex to set up manually.

For the business

Operationally, you’re paying a premium over plain open-source Kubeflow, but you’re offloading the constant hassle of keeping a complex, fast-moving Kubernetes application secure and up-to-date. The “customer tenancy” model also means you don’t have to worry about data leaving your control, which is a hard requirement in regulated industries like finance, healthcare, or government.

How we got here

Kubeflow started at Google as a way to run TensorFlow Extended (TFX) on Kubernetes, but it quickly evolved into a general-purpose MLOps platform. It became a Cloud Native Computing Foundation (CNCF) project in 2022 and is now one of the most popular open-source MLOps tools.

Canonical—the company behind Ubuntu—has been steadily building a managed services portfolio on top of its expertise in Kubernetes and open-source software. It launched a managed Kubeflow service in 2023, initially on bare metal and then on AWS, targeting organizations that wanted an upstream-compliant Kubeflow experience without the DIY burden.

The Azure expansion makes sense. Microsoft’s cloud has a large enterprise base, and many of those users already run Ubuntu Server, use AKS, and have MLOps initiatives. Azure Machine Learning is powerful, but many organizations prefer standards-based, portable MLOps tools. Canonical is filling that gap.

Previous attempts to offer managed Kubeflow on Azure were piecemeal—either third-party guides for self-managed deployments or experimental offers that didn’t stick. This launch is the first time a major Linux company has put its weight behind a fully managed Kubeflow service on the Azure Marketplace.

What to do now

If you’re an existing Kubeflow user on Azure, evaluate whether the managed service can reduce your operational load. Run a proof of concept in a staging environment.

For new ML projects, consider Canonical Managed Kubeflow alongside other managed options like Azure Machine Learning or DIY Kubeflow on AKS. Key steps:

  1. Review prerequisites: You’ll need an active Azure subscription with Contributor or Owner rights, an AKS cluster (or let the deployment create one), and a domain that supports TLS termination.
  2. Deploy from the Azure Marketplace: Search for “Canonical Managed Kubeflow” in the portal. The offer will guide you through the necessary resource group, region, and sizing parameters.
  3. Configure Entra ID: Set up app registrations and grant admin consent for the Kubeflow application so that users can sign in. Canonical’s documentation should provide a step-by-step script.
  4. Integrate your tooling: Connect your Azure Container Registry, Azure Blob Storage (backed by ADLS Gen2), and your preferred CI/CD pipelines. Kubeflow Pipelines can trigger from these services.
  5. Define a cleanup strategy: Since infrastructure costs are yours to bear, create a policy to shut down idle development namespaces or to scale clusters to zero when not in use.

Talk to your Microsoft account team: Canonical’s offer likely counts toward your Azure consumption commitment, so the service might effectively reduce your Microsoft budget burn.

Outlook

Canonical’s move signals a maturing of the MLOps market on Azure. Expect more open-source platform providers to follow with similar “managed-in-your-tenant” offerings—especially as data sovereignty regulations tighten.

For Microsoft, this is a welcome addition: it shows the Azure Marketplace can host first-class, enterprise-grade services from major open-source players without forcing them into a SaaS mold. Longer term, watch for deeper integration with Azure Arc for hybrid deployments and a possible joint go-to-market between Canonical and Microsoft for regulated industries.