Canonical has made its fully managed Kubeflow MLOps platform generally available on the Microsoft Azure Marketplace. The pitch is easy to summarise: a production-ready machine learning environment running inside your own Azure tenant, without you having to assemble and maintain the plumbing yourself.
What was announced
The product is managed Kubeflow, running natively on Azure Kubernetes Service (AKS). Canonical says it goes from Marketplace listing to operational in under an hour, which for a platform of this size is a lot less than the usual hand-built setup takes.
There is no closed magic underneath. It is built on the proven upstream Kubeflow, MLFlow, and KServe projects. Both the application and the automation code are open source, so you are not locked into a proprietary fork that only Canonical knows how to move. If you ever want to take it elsewhere, the pieces are the same ones everyone else uses.
Why it matters for the people running it
The detail that carries the most weight is the in-tenancy deployment: everything runs inside your own Azure Virtual Network (VNet). Proprietary models and training data stay within your security perimeter. For teams handling sensitive data, that boundary is exactly what tends to stop them from trying third-party managed AI platforms in the first place.
Authentication hooks into what you already run. It integrates natively with Microsoft Entra ID, Okta, or any OpenID Connect (OIDC)-compliant identity provider, so there is no separate user system to stand up.
For production, the platform ships with high availability built in. Compute is split across independent worker pools with auto-scaling: cost-effective CPUs for development, powerful GPUs when it is time to train models in earnest. You pay for the expensive hardware only when you need it, instead of keeping GPUs spinning all day.
Maintenance sits with Canonical. Its engineers handle 24/7 management, including version upgrades. For a small team this changes the maths: you do not have to hire MLOps specialists before you even know whether your product has found its market.
The billing detail
It is a transactable listing on the Marketplace, and there is one concrete point worth knowing: subscriptions draw down your Microsoft Azure Consumption Commitment (MACC) on a 1-for-1 basis. If you already have a spend commitment with Azure, what you pay for managed Kubeflow counts against it in full, with no penalty.
Canonical is aiming this at both enterprise and startup AI teams. The underlying argument is the familiar one: get started with MLOps without first building a dedicated team to run it.
To see how this fits with the rest of Canonical’s lineup, our Ubuntu page collects the versions and support details of the distribution that underpins much of its server and cloud stack.
Source
Original announcement from Canonical on its blog: Canonical announces fully Managed Kubeflow AI operations platform on the Microsoft Azure Marketplace. All information comes from Canonical.