Azure Machine Learning service pricing

Build, train and deploy models from the cloud to the edge

Build and train machine learning models faster, then easily deploy to the cloud or the edge with Azure Machine Learning service.

Use the latest open-source innovations such as TensorFlow, PyTorch and Jupyter. Experiment locally and then quickly scale up or out with large GPU clusters in the cloud. Speed up data science with automated machine learning and hyper-parameter tuning. Track your experiments, manage models and easily deploy with integrated CI/CD tooling.

Pricing details

Below prices are for Deploying Machine Learning models using AKS including the compute costs. Only pay for the underlying compute to train your models using Machine Learning Service. Please refer the Virtual Machines pricing page for compute prices.

General purpose

For websites, small to medium databases and other everyday applications

B-series

Instance vCPU RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
B1S 1 1 GiB $- $- $- $- $-
B2S 2 4 GiB $- $- $- $- $-
B1LS 1 0.5 GiB $- $- $- $- $-
B1MS 1 2 GiB $- $- $- $- $-
B2MS 2 8 GiB $- $- $- $- $-
B4MS 4 16 GiB $- $- $- $- $-
B8MS 8 32 GiB $- $- $- $- $-
B12MS 12 48 GiB $- $- $- $- $-
B16MS 16 64 GiB $- $- $- $- $-
B20MS 20 80 GiB $- $- $- $- $-

Av2 Standard

Instance vCPU RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
A1 v2 1 2 GiB $- $- $- – – Blank – – Blank
A2 v2 2 4 GiB $- $- $- – – Blank – – Blank
A4 v2 4 8 GiB $- $- $- – – Blank – – Blank
A8 v2 8 16 GiB $- $- $- – – Blank – – Blank
A2m v2 2 16 GiB $- $- $- – – Blank – – Blank
A4m v2 4 32 GiB $- $- $- – – Blank – – Blank
A8m v2 8 64 GiB $- $- $- – – Blank – – Blank

Reserved Virtual Machine Instances are currently not available for the Av2-series.

D2-64 v3 latest generation

Instance vCPU RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
D2 v3 2 8 GiB $- $- $- $- $-
D4 v3 4 16 GiB $- $- $- $- $-
D8 v3 8 32 GiB $- $- $- $- $-
D16 v3 16 64 GiB $- $- $- $- $-
D32 v3 32 128 GiB $- $- $- $- $-
D48 v3 48 192 GiB $- $- $- $- $-
D64 v3 64 256 GiB $- $- $- $- $-

D2s-64s v3 latest generation

Instance vCPU RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
D2s v3 2 8 GiB $- $- $- $- $-
D4s v3 4 16 GiB $- $- $- $- $-
D8s v3 8 32 GiB $- $- $- $- $-
D16s v3 16 64 GiB $- $- $- $- $-
D32s v3 32 128 GiB $- $- $- $- $-
D48s v3 48 192 GiB $- $- $- $- $-
D64s v3 64 256 GiB $- $- $- $- $-

D1-5 v2

Instance vCPU RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
D1 v2 1 3.5 GiB $- $- $- $- $-
D2 v2 2 7 GiB $- $- $- $- $-
D3 v2 4 14 GiB $- $- $- $- $-
D4 v2 8 28 GiB $- $- $- $- $-
D5 v2 16 56 GiB $- $- $- $- $-

DS1-S5 v2

Instance vCPU RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
DS1 v2 1 3.5 GiB $- $- $- $- $-
DS2 v2 2 7 GiB $- $- $- $- $-
DS3 v2 4 14 GiB $- $- $- $- $-
DS4 v2 8 28 GiB $- $- $- $- $-
DS5 v2 16 56 GiB $- $- $- $- $-

Compute Optimised

High CPU-to-memory ratio. Good for medium traffic web servers, network appliances, batch processes and application servers.

Fsv2-series

Instance vCPU RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
F2s v2 2 4 GiB $- $- $- $- $-
F4s v2 4 8 GiB $- $- $- $- $-
F8s v2 8 16 GiB $- $- $- $- $-
F16s v2 16 32 GiB $- $- $- $- $-
F32s v2 32 64 GiB $- $- $- $- $-
F48s v2 48 96 GiB $- $- $- $- $-
F64s v2 64 128 GiB $- $- $- $- $-
F72s v2 72 144 GiB $- $- $- $- $-

F series

Instance Core RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
F1 1 2 GiB $- $- $- $- $-
F2 2 4 GiB $- $- $- $- $-
F4 4 8 GiB $- $- $- $- $-
F8 8 16 GiB $- $- $- $- $-
F16 16 32 GiB $- $- $- $- $-

Memory optimised

High memory-to-core ratio. Great for relational database servers, medium to large caches and in-memory analytics.

E2-64 v3 latest generation

Instance vCPU RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
E2 v3 2 16 GiB $- $- $- $- $-
E4 v3 4 32 GiB $- $- $- $- $-
E8 v3 8 64 GiB $- $- $- $- $-
E16 v3 16 128 GiB $- $- $- $- $-
E20 v3 20 160 GiB $- $- $- $- $-
E32 v3 32 256 GiB $- $- $- $- $-
E48 v3 48 384 GiB $- $- $- $- $-
E64i v3 1 64 432 GiB $- $- $- $- $-
E64 v3 64 432 GiB $- $- $- $- $-
1 Azure Compute offers virtual machine sizes that are isolated to a specific hardware type and dedicated to a single customer. These virtual machine sizes are best suited for workloads that require a high degree of isolation from other customers for workloads involving elements such as compliance and regulatory requirements.

E2s-64s v3 latest generation

Instance vCPU RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
E2s v3 2 16 GiB $- $- $- $- $-
E4s v3 4 32 GiB $- $- $- $- $-
E8s v3 8 64 GiB $- $- $- $- $-
E16s v3 16 128 GiB $- $- $- $- $-
E20s v3 20 160 GiB $- $- $- $- $-
E32s v3 32 256 GiB $- $- $- $- $-
E48s v3 48 384 GiB $- $- $- $- $-
E64is v3 1 64 432 GiB $- $- $- $- $-
E64s v3 64 432 GiB $- $- $- $- $-
1 Azure Compute offers virtual machine sizes that are isolated to a specific hardware type and dedicated to a single customer. These virtual machine sizes are best suited for workloads that require a high degree of isolation from other customers for workloads involving elements such as compliance and regulatory requirements.

D11-15 v2

Instance vCPU RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
D11 v2 2 14 GiB $- $- $- $- $-
D12 v2 4 28 GiB $- $- $- $- $-
D13 v2 8 56 GiB $- $- $- $- $-
D14 v2 16 112 GiB $- $- $- $- $-
D15 v2 20 140 GiB $- $- $- $- $-

DS11-S15 v2

Instance vCPU RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
DS11 v2 2 14 GiB $- $- $- $- $-
DS12 v2 4 28 GiB $- $- $- $- $-
DS13 v2 8 56 GiB $- $- $- $- $-
DS14 v2 16 112 GiB $- $- $- $- $-
DS15 v2 20 140 GiB $- $- $- $- $-

G series

Instance Core RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
G1 2 28 GiB $- $- $- – – Blank – – Blank
G2 4 56 GiB $- $- $- – – Blank – – Blank
G3 8 112 GiB $- $- $- – – Blank – – Blank
G4 16 224 GiB $- $- $- – – Blank – – Blank
G5 32 448 GiB $- $- $- – – Blank – – Blank

Reserved Virtual Machine Instances are currently not available for the G-series.

M-series

Instance vCPU RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
M64 64 1,024 GiB $- $- $- $- $-
M32ls 32 256 GiB $- $- $- $- $-
M64ls 64 512 GiB $- $- $- $- $-
M64m 64 1,792 GiB $- $- $- $- $-
M128m 128 3,892 GiB $- $- $- $- $-
M8ms 8 219 GiB $- $- $- $- $-
M16ms 16 438 GiB $- $- $- $- $-
M32ms 32 875 GiB $- $- $- $- $-
M64ms 64 1,792 GiB $- $- $- $- $-
M128ms 128 3,892 GiB $- $- $- $- $-
M64s 64 1,024 GiB $- $- $- $- $-
M128 128 2,048 GiB $- $- $- $- $-
M128s 128 2,048 GiB $- $- $- $- $-
M32ts 32 192 GiB $- $- $- $- $-

Extreme Memory Optimised

Instance Active vCPU/
Underlying vCPU
RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
DS11-1 v2 1 / 2 14 GiB $- $- $- $- $-
DS12-1 v2 1 / 4 28 GiB $- $- $- $- $-
DS12-2 v2 2 / 4 28 GiB $- $- $- $- $-
DS13-2 v2 2 / 8 56 GiB $- $- $- $- $-
DS13-4 v2 4 / 8 56 GiB $- $- $- $- $-
DS14-4 v2 4 / 16 112 GiB $- $- $- $- $-
DS14-8 v2 8 / 16 112 GiB $- $- $- $- $-
E4-2s v3 2 / 4 32 GiB $- $- $- $- $-
E8-2s v3 2 / 8 64 GiB $- $- $- $- $-
E8-4s v3 4 / 8 64 GiB $- $- $- $- $-
E16-4s v3 4 / 16 128 GiB $- $- $- $- $-
E16-8s v3 8 / 16 128 GiB $- $- $- $- $-
E32-8s v3 8 / 32 256 GiB $- $- $- $- $-
E32-16s v3 16 / 32 256 GiB $- $- $- $- $-
E64-16s v3 16 / 64 432 GiB $- $- $- $- $-
E64-32s v3 32 / 64 432 GiB $- $- $- $- $-
Gs4-4 4 / 16 224 GiB $- $- $- Not available Not available
Gs4-8 8 / 16 224 GiB $- $- $- Not available Not available
Gs5-8 8 / 32 448 GiB $- $- $- Not available Not available
Gs5-16 16 / 32 448 GiB $- $- $- Not available Not available
M64-16ms 16 / 64 1,750 GiB $- $- $- $- $-
M64-32ms 32 / 64 1,750 GiB $- $- $- $- $-
M128-32ms 32 / 128 3,800 GiB $- $- $- $- $-
M128-64ms 64 / 128 3,800 GiB $- $- $- $- $-

Storage optimised

High disk throughput and IO. Ideal for Big Data SQL, and NoSQL databases.

L-series

Instance Core RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
L4 4 32 GiB $- $- $- $- $-
L8 8 64 GiB $- $- $- $- $-
L16 16 128 GiB $- $- $- $- $-
L32 32 256 GiB $- $- $- $- $-

GPU

Specialised virtual machines targeted for heavy graphic rendering and video editing available with single or multiple GPUs.

NC-series

Instance Core RAM GPU
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
NC6 6 56 GiB 1X K80 $- $- $- $- $-
NC12 12 112 GiB 2X K80 $- $- $- $- $-
NC24 24 224 GiB 4X K80 $- $- $- $- $-
NC24r 24 224 GiB 4X K80 $- $- $- $- $-

NCv2 series

Instance Core RAM GPU
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
NC6s v2 6 112 GiB 1X P100 $- $- $- $- $-
NC12s v2 12 224 GiB 2X P100 $- $- $- $- $-
NC24s v2 24 448 GiB 4X P100 $- $- $- $- $-
NC24rs v2 24 448 GiB 4X P100 $- $- $- $- $-

NCv3 series

Instance Core RAM GPU
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
NC6s v3 6 112 GiB 1X V100 $- $- $- $- $-
NC12s v3 12 224 GiB 2X V100 $- $- $- $- $-
NC24s v3 24 448 GiB 4X V100 $- $- $- $- $-
NC24rs v3 24 448 GiB 4X V100 $- $- $- $- $-

NV-series

Instance Core RAM GPU
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
NV6 6 56 GiB 1X M60 $- $- $- $- $-
NV12 12 112 GiB 2X M60 $- $- $- $- $-
NV24 24 224 GiB 4X M60 $- $- $- $- $-

ND-series

Instance Core RAM GPU
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
ND6 6 112 GiB 1X P40 $- $- $- $- $-
ND12 12 224 GiB 2X P40 $- $- $- $- $-
ND24 24 448 GiB 4X P40 $- $- $- $- $-
ND24r 24 448 GiB 4X P40 $- $- $- $- $-

High performance compute

Our fastest and most powerful CPU virtual machines with optional high-throughput network interfaces (RDMA).

H series

Instance Core RAM
Linux VM price
Machine Learning
Service surcharge
Pay as you go
Total Price
One year reserved
(% Savings)
Total Price
Three years reserved
(% Savings)
Total Price
H8 8 56 GiB $- $- $- $- $-
H16 16 112 GiB $- $- $- $- $-
H8m 8 112 GiB $- $- $- $- $-
H16m 16 224 GiB $- $- $- $- $-
H16mr 16 224 GiB $- $- $- $- $-
H16r 16 112 GiB $- $- $- $- $-

FPGA

Accelerated hardware microservices mainly for inferencing workload

Using ML service on FPGA-based VMs is currently free. Pricing for using ML service on FPGA VMs will be announced at a later date.

Note: FPGA-based VMs are currently restricted for Machine Learning service inferencing workloads only.

Additional information

  • In addition to the above costs, three additional resources will be deployed that will incur additional charges

    1. Azure Container Registry Basic account
    2. Azure Block Blob Storage (general purpose v1)
    3. Key Vault
  • We provide technical support for all Azure services released to general availability through Azure Support starting at $29/month. Billing and subscription management support is provided at no cost.
  • Monthly uptime calculation and Service Levels for Machine Learning Services Realtime Scoring*

    “Total Transaction Attempts” is the total number of API requests by Customer during a billing month for a given Microsoft Azure subscription.

    “Failed Transactions” is the set of all requests within Total Transaction Attempts that fail to return either a Success Code, HTTP 4xx, HTTP 502 or HTTP 503 response.

    “Monthly Uptime Percentage” is calculated as Total Transaction Attempts less Failed Transactions divided by Total Transaction Attempts in a billing month for a given Microsoft Azure subscription. Monthly uptime percentage is represented by the following formula:

    Monthly Uptime % = (Total Transaction Attempts - Failed Transactions) / Total Transaction Attempts

    The following Service Levels and Service Credits are applicable to Customer’s use of the Machine Learning Services Realtime Scoring.

    Monthly uptime percentage Service credit**
    < 99.95% 10%
    < 99% 25%

    *We guarantee to call the compute target (AKS) that user has chosen for service deployment and commit to SLA as long as there are no user errors (e.g. Corrupt model or error in Web service code) and there are no issues with the environment (e.g. not enough AKS agents to handle the request). Any service failures due to user errors like the ones explained above will not count against the SLA

    **Service credit would apply to the surcharge only

    To learn more about the SLA, please visit the SLA page.

FAQ

  • Please review the migration process here: How to migrate.

  • You can continue to use your experimentation and model management accounts as well as the Workbench app for slightly longer after September 2018. Support for the following resources will be removed progressively in the 3-4 months after that release. You can still find the documentation for the old features in the Resources section at the bottom of the table of contents.

    Phase Support details for earlier feature
    1 The ability to create an Azure Machine Learning Experimentation account and a Model Management account in the Azure portal and from the CLI. The ability to create ML Compute Environments from the CLI will also end. If you have an existing account, the CLI and the desktop Workbench will continue to work in this phase.
    2 The underlying APIs for creating old workspaces and projects in the desktop Workbench and with the CLI will end. You can still open existing projects, add additional scripts to them, run scripts in existing projects and deploy web services to existing ML Compute Environments in this phase.
    3 Support for everything else, including the remaining APIs and the desktop Workbench, end in this phase.

    Start migrating today. All of the latest capabilities are available using the new SDK, CLI and portal.

    Please review the migration details.

  • Azure Machine Learning service and the Azure Machine Learning SDK for Python replaces the Azure Machine Learning Experimentation service and the Azure Machine Learning Model Management service, currently in preview. Customers of Azure Machine Learning Experimentation and Model Management services will need to migrate all experiments, models and deployments to the new service. We sincerely apologise for any inconvenience this may cause. Migration is required because we made architectural changes in response to customer feedback received during preview. Azure Machine Learning will now enable faster, more efficient, end-to-end data science from a single Azure Service.

    We have also simplified the pricing for the service and now you only pay for what you use.

    Customers of Azure Machine Learning Experimentation can now use Machine Learning service to train their models with no additional surcharge for the service. You only pay for the underlying compute, BatchAI or DSVM and other Azure services consumed.

    Customers of Azure Machine Learning Model Management service now only pay for what they want to deploy via AKS. You pay for the underlying compute, a small surcharge for the Machine Learning service and any other Azure services consumed.

  • No. Training models using Azure Machine Learning service is free. You have to pay for using underlying compute and VMs. Please refer here for the VM prices.
  • No. You do not pay additional Azure Machine Learning surcharge for deploying models on your own VMs. You will however have to pay for provisioning the VMs. Please refer here for the VM prices.
  • You will be billed daily. For billing purposes, a day commences at midnight UTC. Bills are generated monthly. As a specific example, let’s say that you deploy your models using 10 DS14 v2 VMs. For a billing month of 30 days, you can calculate the prices

    Azure Machine Learning service charges: (10 * 30 * 24 ) * hourly PAYG rate

    Please note that the compute charges are included with Azure Machine Learning service rates above. In addition, you will incur separate charges for any Azure services consumed in conjunction with Azure Machine Learning, including but not limited to HDInsight, Azure Container Registry, Azure Blob Storage, Application Insights, Azure Key Vault, Virtual Network, Azure Event Hub and Azure Stream Analytics.

Resources

Estimate your monthly costs for Azure services

Review Azure pricing frequently asked questions

Learn more about Azure Machine Learning service

Review technical tutorials, videos and more resources

Added to estimate. Press 'v' to view on calculator

Learn and build with $200 in credit, and keep going for free