how to charge the cost for machine learning workstation if there's several weeks not to open it?

The service named “microsoft.machinelearningservices/workspaces” represents the Azure Machine Learning workspace itself. Even when you stop or delete compute instances, the workspace remains as a control-plane resource that stores metadata, settings, logs, and configuration for experiments, models, environments, registries, and networking. It does not run compute on its own, but it does maintain underlying infrastructure such as storage accounts, key vaults, container registries, and network bindings linked to the workspace. Those linked resources can continue to generate charges even if no compute is running.

To control costs when you are not actively working, you can keep the workspace but remove or pause billable dependencies. Shutting down and deleting compute instances stops their charges entirely. Removing private endpoints, disabling managed virtual networks, and deleting premium-tier linked services such as Azure Container Registry or premium storage tiers can reduce most of the remaining cost. You can also shift the workspace into a minimal state by deleting any attached compute clusters, turning off automated ML features, and cleaning up unused datasets, datastores, or registries. The workspace itself costs very little, but its connected resources may not, so the key is to review every linked resource in its resource group.

If you want to request a cost adjustment or cost investigation from Microsoft support, choose a billing-related support request rather than a technical support category. In the support portal, select “Billing” or “Subscription management” as the issue type and then choose a sub-category such as “Billing discrepancy” or “Unexpected charge”. This route ensures the request reaches the team authorized to review cost disputes, credits, and charge explanations.

If the above response helps answer your question, remember to "Accept Answer" so that others in the community facing similar issues can easily find the solution. Your contribution is highly appreciated.

hth

Marcin