Claude Platform on AWS in Practice

Claude Platform on AWS vs. Amazon Bedrock: The Decision Guide

Two ways to run Claude under an AWS account, one fundamental difference: who operates the service and processes your data. Almost every other trade-off follows from that.

Claude 3P 101 · Updated July 2026 · Unofficial guide

Both options let you call Claude with AWS credentials, govern access with IAM, and pay through AWS. The difference is operational ownership. Claude Platform on AWS is Anthropic-operated: models run on Anthropic-managed infrastructure, AWS provides the authentication layer and Marketplace billing, and Anthropic is the data processor for inference inputs and outputs — data may not reside in AWS. Amazon Bedrock is an AWS-operated managed service where AWS is the data processor and retained data stays within your cloud provider's environment. Neither is "better"; they answer different requirements.

Side by side

Claude Platform on AWSAmazon Bedrock
Operator / data processorAnthropic (inference may route to Anthropic's primary cloud)AWS
Feature timingNew models launch simultaneously with the first-party API; features typically same-day, with documented exceptionsAWS's own schedule and feature set
API surfaceClaude API directly (/v1/...), bare model IDs like claude-opus-4-8, anthropic-beta headers pass throughBedrock's surfaces; anthropic.-prefixed model IDs; no anthropic-beta header
Batches, Files API, code execution, web tools, Managed AgentsAvailableNot available (Bedrock offers its own S3-based batch inference instead)
AuthSigV4 (service aws-external-anthropic) or platform API keys; workspace-scoped IAMStandard AWS credentials and IAM
BillingAWS Marketplace, Claude Consumption Units at $0.01/CCU, metered hourly, postpaid arrearsAWS billing
QuotasManaged by Anthropic; orgs start on the Start tier and do not move up automaticallyAWS quota systems
RegionsAll AWS commercial regions; inference_geo for residencyBedrock's own region availability list; regional/multi-region endpoints carry a 10% premium over global (Sonnet 4.5/Haiku 4.5/Opus 4.5 and later)
Compliance fitHIPAA-ready program not available; ZDR available on requestOfficial guidance for FedRAMP High, IL4/IL5, HIPAA-ready needs, or AWS as sole data processor

Two shared notes: cloud marketplace list prices match first-party list prices (committed-use discounts are negotiated with the provider), and the two services draw on separate capacity pools — from each other and from the first-party API — so running both with failover between them is a legitimate pattern, not a hack. Discounts and Marketplace private offers do not transfer automatically between them.

The four questions that decide it

1. Do you need Anthropic's full, current API surface? If your roadmap includes the Message Batches API, Files API, code execution, web search or fetch, Managed Agents, or day-one access to new features via beta headers, Claude Platform on AWS is the only AWS route that has them. If your workload is Messages-plus-tool-use — which all platforms cover — this question doesn't discriminate.

2. Who must process your data? If policy, regulators, or contracts require that data stay in the AWS environment with AWS as the data processor — or you need FedRAMP High, IL4/IL5, or HIPAA-ready operation — the official guidance is Bedrock, full stop. If Anthropic as processor is acceptable (the platform follows the same retention policy as the first-party API, with ZDR available on request), both doors are open.

3. Which operational model fits your team? Bedrock behaves like any AWS service: AWS quotas, AWS-native tooling, cross-region inference profiles, Guardrails. Claude Platform on AWS behaves like the Claude API wearing AWS credentials: quotas and tier changes go through Anthropic, spend limits are not available (use AWS billing controls), and most Admin API endpoints are absent — IAM policies on workspace ARNs replace member management. Ask which support and escalation path your operations team actually wants to live with.

4. How does your code want to be written? Teams standardizing on Anthropic's SDKs get portability from Claude Platform on AWS: AnthropicAWS() is used exactly like Anthropic(), with the same bare model IDs, so first-party code ports with a client swap. Teams deep in AWS SDK conventions may prefer Bedrock's AnthropicBedrockMantle route and AWS-native patterns.

Rule of thumb: feature breadth and first-party velocity → Claude Platform on AWS. Data-boundary and compliance regimes → Bedrock. Genuinely unsure → your compliance answer to question 2 usually settles it before the engineering preferences get a vote.

Where to go next

Go deeper with the Claude Platform on AWS deep dive and the Bedrock deep dive, or compare capabilities in the Bedrock feature availability matrix and the platforms overview. Migrating? See migrating from Bedrock.

Sources