Claude Platform on AWS has been announced, so I've organized the differences between it and Bedrock
This page has been translated by machine translation. View original
On April 20, Anthropic announced an expansion of its partnership with Amazon, which included the introduction of "Claude Platform on AWS".
The following day, a member of the Anthropic team shared that it was currently in preview.
As of the time of writing this article (4/27), the official AWS page has also been published.
"Claude Platform on AWS" is a new service that allows direct access to Anthropic's Claude Platform from AWS accounts. It is currently in preview, with general availability marked as "Coming Soon".
How is it different from Claude via Bedrock? Which one should you use? I've organized the information based on official sources.
Although it's still in the preview stage, understanding the differences now will help you make decisions more quickly when it reaches GA.
Background of the Announcement
First, let's look at the overall picture of this announcement. Claude Platform on AWS is part of a larger expansion of the partnership between Amazon and Anthropic. Here's a summary of key points from Anthropic's official blog:
- Anthropic will invest over $100 billion in AWS technology over the next 10 years, securing up to 5GW of computing capacity
- Amazon is making an additional $5 billion investment (on top of existing $8 billion, potentially adding up to $20 billion more in the future)
- Claude is described as the only frontier AI model available across all three major clouds (AWS / GCP / Azure)
- Over 100,000 customers are using Claude on Bedrock
- Anthropic's run-rate revenue exceeds $30 billion (a rapid increase from approximately $9 billion at the end of 2025)
Claude Platform on AWS was announced as part of this expanded partnership. It provides a new way to use Anthropic's platform directly from AWS, in addition to Bedrock.
What is Claude Platform on AWS
Here are the four benefits described on the AWS official page:
| Benefit | Description |
|---|---|
| AWS Authentication | Access using existing IAM credentials and policies. No separate Anthropic account or API keys required |
| Native Experience | Same API, features, and console experience as accessing Anthropic directly |
| Integrated Auditing | Monitor and audit with CloudTrail, similar to other AWS services |
| Consolidated Billing | Integrated into AWS billing. No separate invoices or vendor contracts needed |
No individual contracts with Anthropic, such as Claude Enterprise, are needed - everything is handled through your AWS account.
*Note: As of the time of writing (April 2026), it's still in preview stage, with general availability marked as "Coming Soon". Pricing has not been announced. If you want access, contacting your account team is recommended.
Differences from Bedrock and How to Choose
Now for the main topic. Based on the four benefits mentioned above, let's compare the positioning with Bedrock.
Core Differences
The AWS authentication, billing, and auditing frameworks are common to both. The differences lie in the inference infrastructure and resulting features.
| Claude Platform on AWS | Claude on Bedrock | |
|---|---|---|
| Authentication, Billing, Auditing | AWS (IAM / Billing / CloudTrail) | AWS (IAM / Billing / CloudTrail) |
| Operating Entity | Anthropic | AWS |
| Data Processing | Anthropic infrastructure (outside AWS boundary) | Within AWS infrastructure |
| Features | Full Anthropic platform experience | AWS managed features (Guardrails, Knowledge Bases, etc.) |
Conceptually, the data flow looks like this:
Claude Platform on AWS:
[User] → [IAM Authentication] → [AWS] → [Inference on Anthropic infrastructure]
↓
[CloudTrail] (audit logs)
[AWS Billing] (consolidated billing)
Claude on Bedrock:
[User] → [IAM Authentication] → [Amazon Bedrock] → [Inference within AWS infrastructure]
↓
[CloudTrail / CloudWatch]
[Guardrails / Knowledge Bases / PrivateLink]
While authentication, billing, and auditing use the same AWS framework, the location where inference occurs differs. With Claude Platform on AWS, data passes through Anthropic's infrastructure, which is the biggest difference from Bedrock. Therefore, Bedrock remains more suitable for organizations with strict data residency requirements.
Bedrock Remains a Key Option
Bedrock continues to evolve as a multi-platform solution and is not being replaced by Claude Platform on AWS.
- Unified use of multiple models (Claude, Nova, Llama, Mistral, DeepSeek, etc.)
- AWS managed features (Guardrails, Knowledge Bases, Agents, PrivateLink, etc.)
- Data residency (data remains within AWS infrastructure)
- Bedrock Mantle adds support for OpenAI-compatible API and Anthropic-compatible API
I've previously analyzed Mantle in these articles:
The Significance of Claude Platform on AWS
For organizations that only need Claude, this adds a new option to access Anthropic's native experience, latest features, and beta access without going through Bedrock.
Organizations currently using the Anthropic API directly will likely see significant benefits. They can maintain existing code and workflows while integrating authentication with IAM and billing with AWS. This eliminates the need for separate contracts with Anthropic or managing separate API keys, simplifying the procurement process.
What Does "Native Experience" Mean Specifically?
The Anthropic blog mentions "more Claude Platform features" for Claude Platform on AWS, suggesting it will offer more functionality than Bedrock.
Specifically, these could include features that are currently not available in Bedrock (marked as unavailable for Bedrock in Anthropic's "Feature availability" documentation):
- Web Search / Web Fetch, Remote MCP, Memory, Files API
- Claude Managed Agents, Message Batches API, Code Execution
*Note: Whether these will actually be available in Claude Platform on AWS is pending official announcement. If provided, they would be clear differentiating factors from Bedrock.
Decision Flowchart
Here's a decision framework for choosing between the options:
Need multi-model usage or AWS managed features?
→ Yes → Bedrock
→ No → Have strict data residency requirements?
→ Yes → Bedrock
→ No → Want Claude's native experience and latest features?
→ Yes → Claude Platform on AWS
→ No → Bedrock is sufficient
If you need multi-model usage, AWS managed features (like Guardrails), or data residency, Bedrock is appropriate. If you only need Claude and prioritize Anthropic's native experience or latest features, Claude Platform on AWS is a candidate.
Points to Verify at GA
Since many specifications haven't been released at this preview stage, here are items to check when GA is announced:
- Network Path: Is communication via the internet or through dedicated connections within AWS? Will private network access like PrivateLink be offered?
- Quotas & Throttling: How do TPM/RPM limits compare to Bedrock's?
- Integration with Existing AWS Services: How well can it be called from Step Functions, EventBridge, etc., and what resource-level control is possible through IAM policies?
- Pricing Structure: How does it compare to using Claude through Bedrock?
- SLA & Support Channel: Is AWS or Anthropic the escalation point for issues?
Network path and support channels in particular are critical points for enterprise adoption decisions. I plan to verify these when GA is announced.
Summary
Claude Platform on AWS offers a new option to use Anthropic's full platform within AWS's authentication, billing, and auditing framework.
This doesn't make Bedrock obsolete. Bedrock remains optimal if you need multi-model usage, data residency, or AWS managed features. Claude Platform on AWS represents a new option for organizations that only need Claude or are currently using the Anthropic API directly.
Including the previously analyzed Bedrock Mantle's Anthropic-compatible endpoints and Codex CLI's Bedrock support, the options for using AI models on AWS are definitely increasing. With more choices available, having a framework for deciding which to choose becomes important. Much information remains unpublished, including pricing structure, network paths, and AWS service integration, which I plan to analyze when GA is announced.