Question

A company runs several SageMaker real-time endpoints across multiple product teams in a single AWS account. Finance needs a monthly breakdown of inference infrastructure spend attributed to each team so they can charge costs back internally. The ML platform team wants to enable this reporting with the least ongoing operational effort. Which approach should they use?

Accepted Answer

Apply a consistent cost allocation tag (for example, team) to each endpoint and its resources, activate the tag as a cost allocation tag in the Billing console, and use AWS Cost Explorer or Cost and Usage Reports grouped by that tag.

Answer

Create a separate AWS account for each team using AWS Organizations and move each team's endpoints into its own account so billing is naturally separated.

Answer

Enable CloudWatch billing alarms per endpoint and manually sum the InvocationsPerInstance metric values at the end of each month to estimate each team's share.

Answer

Parse CloudTrail logs of InvokeEndpoint API calls each month with an Athena query and multiply the call counts by the instance hourly rate to allocate cost.