Who processes my data in each option, and where does it reside?

In Claude in Amazon Bedrock, AWS operates the infrastructure: inference data stays inside the AWS security boundary, Anthropic personnel have no access to the inference stack, and data governance is ruled by Amazon Bedrock. In Claude Platform on AWS, Anthropic is the processor of inference data: it may not reside in AWS and inference may route to Anthropic's primary cloud. You can pin the inference geography per request with the inference_geo parameter and request Zero Data Retention (ZDR) from Anthropic.

How is each one billed, and what happens to my AWS spend commitments?

Claude in Amazon Bedrock is billed as a native AWS service. Claude Platform on AWS bills through AWS Marketplace in Claude Consumption Units (CCUs): usage metered hourly and invoiced monthly in arrears on your AWS bill, with no prepaid credits and no committed balance. In both cases the spend arrives through your AWS account, and AWS commitment retirement is preserved when migrating. One detail that surprises more than one team: negotiated discounts and private offers do not transfer automatically between the two offerings — they are agreed separately.

If something breaks, who answers: AWS or Anthropic?

In Claude in Amazon Bedrock, AWS operates the inference stack and manages the service quotas and limits. In Claude Platform on AWS, Anthropic operates the platform, manages rate limits directly and provides direct support; signing up also requires accepting Anthropic's commercial terms (even if you already use Claude through Bedrock). Before choosing, define who your first escalation line will be in production and make sure procurement is clear on who owns each contractual relationship.

Which capabilities actually change between the two?

Claude Platform on AWS exposes Anthropic's full platform: Agent Skills, code execution in a managed sandbox, the Files API, batch processing, beta features via the anthropic-beta header, and new models typically the same day as the Claude API. The Claude in Amazon Bedrock integration centers on the Messages API inside the AWS boundary and receives features per the Bedrock release schedule; as of today, server-side capabilities such as Agent Skills, code execution or the Files API are not available through that path.

Can I use both at once, or switch later without redoing everything?

Yes. Anthropic documents that Claude Platform on AWS uses a capacity pool separate from both the first-party API and Amazon Bedrock, and that you can run workloads on more than one platform and fail over between them. Migrating means changing the base URL, model IDs and some headers; AWS IAM (SigV4) authentication carries over on both. Two practical warnings: Zero Data Retention is opt-in on Claude Platform on AWS (confirm it before moving sensitive workloads) and commercial terms for each path are negotiated separately.

My industry is regulated (banking, insurance, healthcare). Which one applies?

For regulated players in the region — banking and insurance under SBS in Peru or CMF in Chile, healthcare with patient data — the decisive question is about the data boundary: can your data have Anthropic as processor, or must it stay under a single processor inside the AWS boundary? When the latter applies, the path Anthropic documents is Claude in Amazon Bedrock, which runs entirely on AWS-controlled infrastructure (its documentation cites US programs such as FedRAMP or HIPAA-ready as examples; the underlying principle — a single audited processor — is the same one regulators in the region review). In addition, Anthropic's HIPAA-ready program is not available on Claude Platform on AWS. In practice: regulated workloads on Bedrock, with Claude Platform on AWS as a complement for innovation teams.

Claude Platform on AWS vs Amazon Bedrock

Until recently, “using Claude on AWS” meant one thing: enabling Anthropic’s models in Amazon Bedrock. Since May 2026 there is a second door: Claude Platform on AWS, Anthropic’s full platform contracted and billed through your AWS account. The two sound alike, and that similarity confuses: the choice defines who operates the infrastructure, who processes your data, which capabilities you receive, and who you negotiate with.

This guide compares both options against the official Anthropic and AWS documentation, so the decision is made with architecture criteria rather than inertia.

The two offerings in one sentence

Claude in Amazon Bedrock: AWS operates the inference stack. Your data stays inside the AWS security boundary, with AWS-native authentication, billing and monitoring. It is the classic path, today with two integrations: the current one (Messages API, the same shape as the Claude API) and the legacy one (InvokeModel/Converse), which remains available.
Claude Platform on AWS: Anthropic operates the platform. You get the full Claude API experience — Agent Skills, code execution, Files API, batch, betas — with AWS IAM authentication and billing through AWS Marketplace. AWS provides the front door; Anthropic provides the platform.

The underlying difference comes down to one question: do you need AWS to be the sole processor of your data, or do you need Anthropic’s complete platform with your existing AWS commercial relationship?

Claude in Amazon Bedrock: the AWS security boundary

When you consume Claude through Amazon Bedrock, inference runs on AWS-managed infrastructure with zero operator access: per Anthropic’s official documentation, its personnel have no access to the inference infrastructure. Data governance is ruled by Amazon Bedrock, logs flow to CloudWatch and CloudTrail, and access control uses IAM (Bedrock service role, assumed roles, or short-lived tokens).

For data residency, Bedrock offers two endpoint types: global (dynamic routing across regions for maximum availability) and regional (inference resolves in the AWS region you specify, designed for residency requirements, at a surcharge over global).

What you need to be clear about: the Bedrock integration centers on the Messages API. Server-side capabilities of the Anthropic platform — Agent Skills, code execution, the Files API, the anthropic-beta header — are not available through this path, and new features arrive per the Bedrock release schedule.

Claude Platform on AWS: the full platform, billed by AWS

Claude Platform on AWS inverts the model: the models run on Anthropic-managed infrastructure, and AWS provides the authentication layer (SigV4 or API key), IAM-based access control, and billing through AWS Marketplace. In practice, it is the first-party Claude API with your AWS governance and commercial relationship.

That carries concrete consequences:

Same-day capabilities. New models and features typically arrive the same day as the direct Claude API: Agent Skills, code execution in a managed sandbox, the Files API, batch processing, full prompt caching and beta features.
Data processed by Anthropic. Anthropic is the processor of inference data; data may not reside in AWS and inference may route to Anthropic’s primary cloud. The inference_geo parameter pins the inference geography per request (for example, keeping it in US data centers), and Zero Data Retention is available on request.
Billing through AWS Marketplace. Usage is denominated in Claude Consumption Units (CCUs), metered hourly and invoiced monthly in arrears on your AWS bill. They are not prepaid credits: there is no CCU balance or commitment.
End-to-end AWS governance. Claude Platform workspaces are the primary IAM resource (with their own ARN for policies), CloudTrail can capture every request, and AWS PrivateLink is supported to connect your VPC to the endpoint.

Head-to-head comparison

Aspect	Claude Platform on AWS	Claude in Amazon Bedrock
Who operates the stack	Anthropic	AWS
Inference data processor	Anthropic	AWS
Data residency	May leave AWS; pinned with `inference_geo`	Inside the AWS boundary; global or regional endpoint
API surface	Full Claude API	Messages API (plus native Bedrock APIs on the legacy integration)
Agent Skills, code execution, Files API, batch	Available	Not available on this path
Beta features	Yes, via `anthropic-beta` header	Per Bedrock release schedule
New model availability	Same day as the Claude API	Per Bedrock release schedule
Authentication	AWS IAM (SigV4) or API key	AWS IAM (SigV4), roles or short-lived tokens
Billing	AWS Marketplace (CCUs, monthly in arrears)	Native AWS service
Rate limits and quotas	Managed by Anthropic	Managed by AWS
First-line support	Anthropic	AWS
Regulated industries (AWS as sole data processor)	Not available today	The path Anthropic recommends

When to choose each one

Choose Claude in Amazon Bedrock when the data boundary rules. If your regulator, your contracts or your internal policy require inference data to stay under a single processor inside the AWS boundary — the typical case for banking, insurance and healthcare in the region — or if your security team already audits everything inside AWS, Bedrock is the documented answer. It is also the natural option if you already run your agents there with Bedrock AgentCore.

Choose Claude Platform on AWS when platform velocity rules. If your use case depends on Agent Skills, code execution, the Files API, or trying betas as soon as they ship — typical for teams building business-facing agentic AI — and your data framework accepts Anthropic as processor (with inference_geo and ZDR as controls), this path gives you the complete platform without opening a new procurement relationship: everything arrives on your AWS bill.

And it is not an irreversible decision. Both use separate capacity pools, you can operate the two in parallel and fail over between them. Many organizations land on a mixed model: regulated workloads on Bedrock, product teams on Claude Platform on AWS.

What if my team already works with Cowork or Claude Code?

The platform decision does not lock you out of Anthropic’s agentic tooling either — Claude’s official documentation covers both routes:

Claude Cowork (the agent mode of the Claude desktop app) runs in its standard mode on Anthropic’s platform with Claude plans. And for organizations that need inference to stay in their own AWS account, the official documentation describes the third-party mode with Amazon Bedrock: the IT administrator sets inferenceProvider=bedrock, distributes the profile via MDM (Jamf, Intune or group policies), enables corporate sign-in via AWS IAM Identity Center, and gets per-user identity in CloudTrail.
Claude Code documents official configuration for both paths: Amazon Bedrock (the CLAUDE_CODE_USE_BEDROCK variable) and Claude Platform on AWS (the CLAUDE_CODE_USE_ANTHROPIC_AWS variable plus the workspace ID), so the same development team can point its tooling at whichever platform the organization chose.

And Cowork directly on Claude Platform on AWS? As of today that combination is not documented: the desktop app’s third-party mode covers Amazon Bedrock (plus Vertex AI and compatible gateways), and organizations choosing Claude Platform on AWS tap it from Claude Code and the Agent SDK, while Cowork operates on Anthropic’s platform with Claude plans.

In other words: first decide the platform on data, feature and contract criteria; the day-to-day tools align afterwards through configuration, not repurchasing.

How Caleidos helps

At Caleidos, as an AWS Advanced Tier Services Partner, we help define this architecture before it becomes debt: we assess your regulatory and data framework, design the consumption path (Bedrock, Claude Platform on AWS, or mixed) and land it in your account with IAM, monitoring and cost governance from day one. Our Agentic AI on AWS practice builds the agents on that foundation, and our Amazon Bedrock page goes deeper into the AWS-native path.

Frequently asked questions

Who processes my data in each option? On Bedrock, AWS operates the infrastructure and data stays inside its security boundary. On Claude Platform on AWS, Anthropic processes inference; geography is pinned with inference_geo and ZDR is available on request.

How is each billed? Bedrock as a native AWS service; Claude Platform on AWS through AWS Marketplace in CCUs, monthly in arrears. In both cases the spend arrives on your AWS bill; discounts on one do not transfer automatically to the other.

Which one fits a regulated industry? If the data must stay under a single processor inside the AWS boundary — the usual case for banking, insurance and healthcare in the region — the path Anthropic documents is Claude in Amazon Bedrock.

Evaluating which path to adopt Claude through in your organization?

Let’s talk about your generative AI architecture: we help you choose the right platform for your regulatory framework and take it to production on AWS.

Claude Platform on AWS vs. Claude in Amazon Bedrock: what changes and which one fits your company