Charging on SageMaker AI models with autoscaling

When using SageMaker endpoints with autoscaling for marketplace machine learning models, the pricing structure depends on how the seller has configured their pricing model.

For machine learning products with hourly pricing in AWS Marketplace, you are typically charged per instance of the software running in SageMaker. This means that if your endpoint scales up to multiple instances due to autoscaling, you would be charged for each running instance.

The pricing works as follows:

• Sellers can charge a different hourly price for each instance type that their software runs on
• Usage is tracked by AWS Marketplace and billed accordingly
• Usage is prorated to the minute

For model package products specifically, there are two ways buyers can run the software:

1. Hosting an endpoint continuously for real-time inference
2. Running a batch transform job on a dataset

Sellers can set different pricing for both of these usage patterns.

In addition to the software costs from the marketplace, you'll also be responsible for the SageMaker infrastructure costs for each running instance. These infrastructure costs are set by AWS, not the marketplace seller, and scale with the number of instances you're running.

So in summary, for marketplace models on SageMaker endpoints with autoscaling, both the model subscription charges and the infrastructure charges typically scale with the number of running instances, not as a fixed cost independent of instance count.
Sources
Machine learning product pricing for AWS Marketplace - AWS Marketplace
Free trials of products listed on AWS Marketplace - AWS Marketplace

Hello,

I would like to inform you that, for Amazon SageMaker Marketplace models deployed on endpoints with autoscaling, the model subscription charges are generally billed per running instance, not as a fixed cost.

When a model package is purchased from AWS Marketplace, customers pay two components:

1. Infrastructure cost — the hourly cost for the underlying SageMaker instances.
2. Software (model) cost — the price set by the model provider, which is usually defined per hour per instance or per inference, depending on the listing.

In the case of hourly software pricing, each active instance of the model contributes to the overall charge. Therefore, if the endpoint scales out to multiple instances due to increased traffic, both the SageMaker instance cost and the model subscription fee scale proportionally.

If the model listing uses inference-based pricing instead, charges are incurred per invocation rather than per instance, and autoscaling would not directly affect the software cost (only the infrastructure cost).

For confirmation, it’s always best to review the “Pricing” section of the specific model package in AWS Marketplace, as pricing behaviour may vary by provider.

For better assistance, I kindly request you to please reach out to Premium Support so that we can check your use case and current environment better.

Additionally, I kindly request you to please stay updated through these blogs:

[+] What's New with AWS - https://aws.amazon.com/new/

[+] AWS News Blog - https://aws.amazon.com/blogs/aws/

If you have any questions or would like more details about this query, please reach out to AWS Support, and we would be happy to assist you further.

—References—

[1] Creating support cases and case management: https://docs.aws.amazon.com/awssupport/latest/user/case-management.html#creating-a-support-casehttps://docs.aws.amazon.com/awssupport/latest/user/case-management.html#creating-a-support-case