Cost allocation tags are AWS's standard mechanism for slicing a bill: key–value labels (like team=search or cost-center=4102) that, once activated in the Billing console, become filterable dimensions in Cost Explorer and billing exports. Teams coming from Amazon Bedrock often assume the same tag-everything workflow applies to Claude Platform on AWS. It partly does — but because the platform is Anthropic-operated and billed as a consolidated AWS Marketplace charge, the sharpest attribution tool is actually the workspace, with AWS tags playing a supporting role.
Workspaces are the native attribution unit
On Claude Platform on AWS, usage, quotas, cost, files, batches, and Skills all roll up per workspace. A workspace is bound to a single AWS region, has a tagged ID like wrkspc_01AbCdEf23GhIj, and is the platform's primary IAM resource — its ARN looks like arn:aws:aws-external-anthropic:{region}:{account-id}:workspace/{workspace-id}. That makes "one workspace per team, product, or cost center" the foundational move for cost attribution:
Each application or team gets its own workspace; IAM policies scope each team's principals to their workspace ARN, so usage cannot leak across attribution boundaries; and the Claude Console's Usage and Cost pages then show consumption broken down along exactly the lines finance cares about. This is attribution enforced by access control, which is stronger than attribution by tagging convention — a mis-tagged Lambda can charge the wrong cost center, but a workload without IAM access to another team's workspace cannot consume against it. See workspace isolation and multi-workspace strategy for the design patterns.
Where AWS cost allocation tags fit
The AWS-side charge for Claude Platform arrives as a consolidated Marketplace line item, metered in Claude Consumption Units and invoiced monthly in arrears (see how billing works). Standard AWS tagging practice still applies to everything around the platform: the EC2 instances, Lambda functions, and ECS services that call Claude carry your normal cost allocation tags, and separating callers by AWS account gives you a hard billing boundary as well.
For tags on the Claude Platform resources themselves, the official documentation we rely on describes the workspace as the platform's single IAM resource type but does not detail a resource-tagging workflow for it — so check the current official documentation for whether and how tags can be attached to workspaces in your account before building a chargeback pipeline on that assumption. Whatever you find, the durable pattern is the same one AWS recommends everywhere: activate your organization's standard cost allocation tag keys in the Billing console, apply them consistently to the calling infrastructure, and use account or workspace boundaries where you need attribution you can enforce.
A practical chargeback recipe
Putting the pieces together, a setup that works without building a separate billing system:
1. One workspace per chargeback unit. Create workspaces from the AWS Console Workspaces page (the Claude Console's Workspaces view is read-only), named for the team or product they serve.
2. IAM policies per workspace ARN. Grant each team's roles inference actions only on their own workspace, so the attribution boundary is enforced, not advisory.
3. Monthly reconciliation. Pull the per-workspace usage and cost views from the Claude Console, and match the total against the Marketplace line item on the AWS invoice. Note that the programmatic Usage and Cost API is not available on Claude Platform on AWS, so this step uses the Console views rather than an automated Anthropic-side export.
4. Tag the callers. Keep normal cost allocation tags on the compute that invokes Claude, so the surrounding infrastructure cost lands in the same cost-center buckets as the model spend.
The result: finance gets team-level Claude numbers from workspaces, the invoice reconciles at the Marketplace line, and no one maintains a parallel metering system.
Where to go next
Read viewing Claude Platform costs in AWS Cost Explorer for the AWS-side reporting setup, or the workspace ID explainer if workspaces are new to you.