34  Introducing the Claude Platform on AWS

34.1 Overview

On May 11, 2026, Anthropic announced the general availability of the Claude Platform on AWS — a fully featured deployment of the Claude Platform that integrated natively with AWS authentication, billing, and commitment programs. For organizations already standardized on AWS, this meant they could adopt the complete Claude Platform feature set without leaving the AWS ecosystem for credentials, invoicing, or governance.

The announcement addressed a recurring tension in enterprise AI adoption. Companies wanted the full power of the Claude Platform — Managed Agents, the Advisor strategy tool, web search, code execution, Skills, and more — but their security and procurement teams wanted everything to flow through existing AWS controls. The Claude Platform on AWS resolved this tension by making AWS the billing and identity layer while Anthropic remained the service operator.

This chapter covers what the Claude Platform on AWS includes, how it differs from Amazon Bedrock, and the architectural and governance implications for enterprises choosing between the two.

34.2 What the Claude Platform on AWS Includes

The Claude Platform on AWS delivered the complete feature set that Anthropic offered through its direct API, but accessed through AWS-native authentication and billed through a single AWS invoice. Every major Platform capability was available:

Feature Description Enterprise Value
Managed Agents Cloud-hosted agents that run autonomously Walk-away workloads, scheduled tasks
Advisor Strategy and guidance tool Decision support, planning
Web Search & Fetch Real-time information retrieval Grounded, up-to-date responses
Code Execution Sandboxed code runtime Data analysis, computation
Files API Upload, store, and reference files Document processing
Skills Reusable procedural knowledge Consistent task execution
MCP Connector Connect to external tools/data Integration with enterprise systems
Prompt Caching Cache context for faster, cheaper calls Cost reduction on repeated context
Citations Source attribution for responses Trust and verifiability
Batch Processing Asynchronous bulk processing Cost-effective large-scale work

34.2.1 Available Models

At launch, the Claude Platform on AWS offered three models:

  • Claude Opus 4.7 — The most capable model, for complex reasoning, agentic tasks, and high-stakes work.
  • Claude Sonnet 4.6 — The balanced model, offering strong performance at lower cost for most production workloads.
  • Claude Haiku 4.5 — The fast, efficient model for high-volume, latency-sensitive tasks.
# Example: Calling Claude on AWS using IAM authentication
aws claude invoke-model \
  --model-id claude-opus-4-7 \
  --content '{"messages":[{"role":"user","content":"Summarize this quarterly report."}]}' \
  --region us-east-1

34.3 AWS-Native Authentication and Governance

The defining characteristic of the Claude Platform on AWS was its deep integration with AWS identity and governance services. Rather than managing separate API keys, organizations used AWS IAM for authentication.

34.3.1 IAM Integration

Authentication flowed through standard AWS IAM mechanisms:

# Authentication via AWS IAM (no static API keys needed)
import boto3

client = boto3.client(
    'claude',
    region_name='us-east-1',
    # Credentials come from IAM role, SSO, or environment
)

response = client.invoke_model(
    modelId='claude-opus-4-7',
    body={
        'messages': [{'role': 'user', 'content': 'Analyze this dataset.'}]
    }
)

This meant organizations could leverage their existing IAM patterns:

  • IAM roles for EC2 instances, ECS tasks, and Lambda functions.
  • IAM Identity Center (formerly AWS SSO) for human users.
  • STS temporary credentials for cross-account access.
  • IAM policies for fine-grained access control.

34.3.2 CloudTrail Logging

Every API call to the Claude Platform on AWS was automatically logged in AWS CloudTrail, providing a complete audit trail of who called what, when, and from where. This was a significant advantage for organizations subject to compliance regimes that required immutable access logs.

CloudTrail integration meant that Claude API calls appeared alongside other AWS service calls in security information and event management (SIEM) systems, enabling unified threat detection and compliance reporting.

34.3.3 Single AWS Invoice

All Claude usage was billed through the standard AWS billing system, appearing as a single line item on the organization’s AWS invoice. This had several practical benefits:

  • Simplified procurement: No separate vendor relationship or contract with Anthropic.
  • Commitment utilization: Organizations could apply AWS commitment discounts (Savings Plans, EDP commitments) to Claude usage.
  • Consolidated billing: Finance teams received one invoice covering infrastructure and AI services.
Note

For organizations with significant AWS Enterprise Discount Program (EDP) commitments, routing Claude spend through AWS meant that commitment dollars could be applied — effectively reducing the marginal cost of Claude usage against pre-negotiated discount rates.

34.4 Claude Platform on AWS vs. Amazon Bedrock

The announcement inevitably raised a question: How is this different from Amazon Bedrock, which already offers Claude models? Anthropic provided a clear framework for understanding the distinction.

Dimension Claude Platform on AWS Amazon Bedrock
Operator Anthropic operates the service AWS operates the service
Data processor Anthropic AWS
Feature set Full Claude Platform (all features) Claude models (subset of features)
Authentication AWS IAM AWS IAM
Billing AWS invoice AWS invoice
Data residency Anthropic’s infrastructure AWS region (customer chooses)
Governance CloudTrail + Anthropic Console CloudTrail + AWS console

The key distinction was about who operates the service and who is the data processor:

  • Claude Platform on AWS: Anthropic operates the platform and is the data processor. The full feature set is available, including Managed Agents, Advisor, Skills, and all Platform capabilities. Authentication and billing flow through AWS.

  • Amazon Bedrock: AWS is the data processor. Claude models are available as part of the Bedrock model marketplace. Data residency is determined by the AWS region the customer selects, and AWS governs the data handling.

Tip

Choosing between them: If your priority is maximum features (Managed Agents, Skills, MCP connector, Advisor), choose the Claude Platform on AWS. If your priority is data residency in a specific AWS region or you’re already standardized on Bedrock for multi-model access, choose Amazon Bedrock. Both use AWS authentication and billing.

34.4.1 Decision Framework

For most organizations, the decision came down to three questions:

  1. Do you need the full Platform feature set? If you need Managed Agents, Skills, or the MCP connector, the Claude Platform on AWS is the only option that provides them through AWS billing.
  2. Do you have strict data residency requirements tied to a specific AWS region? If so, Bedrock gives you region-level control that the Platform on AWS does not.
  3. Are you already using Bedrock for other models? If you’re already managing access to multiple models through Bedrock, adding Claude there may be simpler operationally — even if you sacrifice some Platform features.

34.5 Architecture and Data Flow

Understanding how the Claude Platform on AWS handled data was important for security and compliance teams.

┌─────────────────────────────────────────────────┐
│              Your AWS Environment                 │
│                                                   │
│  ┌──────────┐    IAM Auth    ┌────────────────┐  │
│  │ Your App │───────────────►│ Claude Platform│  │
│  │ /Service │                │   (on AWS)     │  │
│  └──────────┘                └───────┬────────┘  │
│                                      │            │
│  ┌──────────┐                        │            │
│  │CloudTrail│◄───── Logs ────────────┘            │
│  └──────────┘                                    │
└──────────────────────────────────────────────────┘
                        │
                        │ Inference
                        ▼
              ┌──────────────────┐
              │  Anthropic Infra │
              │  (Data Processor)│
              └──────────────────┘

The flow was:

  1. Your application or service authenticates using AWS IAM.
  2. Requests are routed to the Claude Platform endpoint.
  3. All calls are logged in CloudTrail.
  4. Inference is performed on Anthropic’s infrastructure, where Anthropic acts as the data processor.
  5. Billing appears on your AWS invoice.

34.6 Common Use Cases

34.6.1 Enterprise Application Integration

Organizations building applications on AWS could now call Claude using the same IAM roles their applications already used — no API key management, no secrets rotation.

# Example: ECS task role policy for Claude access
PolicyDocument:
  Version: '2012-10-17'
  Statement:
    - Effect: Allow
      Action:
        - claude:InvokeModel
        - claude:CreateAgent
        - claude:ManageAgent
      Resource: '*'

34.6.2 Managed Agents on AWS

Managed Agents — cloud-hosted agents that run autonomously — were fully supported. An organization could create an agent that monitored CloudWatch alerts, diagnosed issues, and proposed remediations, all running on Anthropic’s infrastructure but authenticated and billed through AWS.

34.6.3 Batch Processing for Cost Efficiency

For large-scale workloads like document processing or data enrichment, the Batch API offered significant cost savings (typically 50% versus synchronous calls). On AWS, these batch jobs could be triggered by S3 events, processed through the Claude Platform, and results written back to S3 — all within the AWS ecosystem.

34.7 Migration and Adoption Patterns

For organizations already using Claude through the direct Anthropic API, migrating to the Platform on AWS was straightforward. The API surface was identical; only the authentication mechanism changed. Anthropic provided migration tooling that translated existing API key configurations to IAM-based authentication.

For organizations using Bedrock, the decision to add the Claude Platform on AWS was typically additive rather than a replacement — they might use Bedrock for models where data residency in a specific region was critical, and the Claude Platform on AWS for workloads that needed Managed Agents or Skills.

34.8 Key Takeaways

  • Full Platform features, AWS-native access. The Claude Platform on AWS delivers the complete feature set — Managed Agents, Advisor, Skills, MCP connector, and more — accessed through AWS IAM authentication and billed on a single AWS invoice.
  • Three models at launch. Opus 4.7, Sonnet 4.6, and Haiku 4.5 are all available, covering the full spectrum of capability and cost.
  • IAM, CloudTrail, and AWS billing are the governance backbone. No separate API keys to manage; every call is auditable in CloudTrail; spend applies to AWS commitments.
  • The distinction from Bedrock is about the operator and data processor. Claude Platform on AWS = Anthropic operates (full features); Bedrock = AWS operates (region-level data residency). Both use AWS auth and billing.
  • Choose based on your priorities. Maximum features → Platform on AWS. Strict region-level data residency → Bedrock. Multi-model consolidation → Bedrock.
  • Commitment dollars apply. AWS EDP and Savings Plan commitments can offset Claude Platform costs, reducing effective spend for committed customers.
  • Migration from direct API is low-friction. The API surface is identical; only authentication changes from API keys to IAM.
  • The Claude Platform on AWS is the right default for AWS-centric organizations that want the full Platform experience without leaving the AWS ecosystem.