Azure Machine Learning pricing

Azure Machine Learning has no additional charge to use. Costs come from the Azure compute used for training or inference and from other consumed Azure services, such as Blob Storage, Key Vault, Container Registry, and Application Insights. Azure lists prices as estimates; actual pricing can vary by agreement, purchase date, and currency exchange rate.

Select the VM family and size that match the workload: Azure lists general-purpose, compute-optimized, memory-optimized, GPU, and high-performance compute options. The service supports model training and deployment, while the underlying selected compute is the principal price driver.

Pay-as-you-go compute is billed by the second with no long-term commitment or upfront payment. Savings plans can apply through a one- or three-year fixed hourly spend commitment across select compute services, while VM reservations are presented for stable, predictable workloads. Azure bills daily, with each billing day beginning at midnight UTC; bills are generated monthly.

Include the Azure resources used alongside Azure Machine Learning in the budget. The pricing page specifically identifies services including Blob Storage, Key Vault, Container Registry, and Application Insights; its examples also identify services such as virtual networking and Event Hubs as separate consumed-service charges.

Azure Machine Learning is a fit when the team wants an end-to-end ML service while selecting and paying for Azure infrastructure separately. For predictable, sustained compute, evaluate the applicable savings-plan or reservation option against pay-as-you-go pricing; for Managed Spark memory-optimized work, use the published per-vCore-hour rate as the starting input.