Anthropic has launched the Claude Platform on AWS as generally available — a first-party offering that puts Anthropic’s complete native API directly on AWS infrastructure, under your existing AWS account, billing, and IAM credentials.
This is not Amazon Bedrock. That distinction matters, and it is probably the first thing worth understanding.
Two ways to run Claude on AWS
AWS developers now have two distinct pathways to Claude, and the choice between them is not arbitrary.
Claude on Amazon Bedrock is the familiar path: AWS acts as the data processor, your data stays within AWS’s infrastructure boundary, and you interact with Claude through Bedrock’s unified model APIs alongside other providers like Meta, Cohere, and Stability AI. This is the right choice for workloads where data residency, compliance frameworks, or existing Bedrock integrations drive the architecture.
Claude Platform on AWS is something different. Here, Anthropic operates the service. Data is processed outside AWS’s infrastructure boundary (just as it would be if you called Anthropic’s API directly). In exchange, you get complete feature parity with Anthropic’s native Claude API — including capabilities that do not exist in Bedrock today — while the billing, authentication, and audit trail stay inside your AWS account.
Neither is universally better. The right one depends on whether your primary requirement is data residency (Bedrock) or full access to Anthropic’s latest platform capabilities (Claude Platform on AWS).
What “full feature parity” actually means
This is where the announcement gets interesting for builders. The Claude Platform on AWS ships with features that represent Anthropic’s current frontier of what you can do with Claude in production.
Agentic capabilities:
- Claude Managed Agents (beta) — Deploy and orchestrate agents at scale without managing infrastructure for agent execution.
- Advisor Strategy (beta) — An architectural pattern where a second, advisory model consults during an agent’s reasoning process to improve decision quality.
- MCP Connector (beta) — Connect Claude to remote Model Context Protocol servers without writing and maintaining custom connector code.
- Skills (beta) — Package best-practice instructions so Claude can be taught consistent, repeatable behaviors for your domain.
Data and tool access:
- Web Search and Web Fetch — Give Claude access to current information from the internet as part of your API calls.
- Code Execution — Run Python and generate visualizations directly within an API call, without provisioning a separate execution environment.
- Files API (beta) — Upload documents once and reference them across multiple conversations, rather than re-sending context on every request.
Cost and quality:
- Prompt Caching — Reduce cost and latency by caching frequently repeated context. If you are sending large system prompts or static reference documents repeatedly, this pays for itself quickly.
- Citations — Ground Claude’s responses in specific source passages, with verifiable references back to the source material.
- Batch Processing — Submit high-volume asynchronous workloads for processing at reduced cost.
Customers also get access to the Claude Console, including prompt generation, prompt improvement, and evaluation tooling — the same development environment available on Anthropic’s direct platform.
For teams building complex agent workflows, this is more than a model-access change. It shifts more of the agent stack into Anthropic’s native platform. The tradeoff is that workloads now depend on Anthropic-operated infrastructure rather than Bedrock’s AWS-operated boundary — that distinction matters for procurement, compliance reviews, and teams with existing Bedrock-native integrations.
The AWS integration story
For enterprise AWS customers, the integration mechanics may be the most immediately compelling part.
Authentication runs entirely through AWS IAM. Your existing IAM roles, policies, and permission boundaries apply. No separate API keys to rotate, no new identity provider to federate. For teams already running least-privilege IAM policies across their AWS accounts, Claude just becomes another service in the policy.
Billing is unified through AWS invoicing, which means spend against the Claude Platform draws against your existing AWS Enterprise Discount Program commitments or Private Pricing Agreements. For organizations managing AWS spend centrally, this eliminates a separate vendor relationship and the overhead that comes with it.
Audit logging flows through AWS CloudTrail, so Claude API calls appear in the same audit trail as every other AWS API call you make. For security and compliance teams, this is significant: you do not need a new logging pipeline or a new integration with your SIEM to capture Claude API activity.
Regional availability covers most AWS commercial regions, with both global and U.S. inference geographies available at launch.
Model availability
The platform launches with Anthropic’s current model family:
- Claude Opus 4.7 — highest capability, suited for complex reasoning and analysis
- Claude Sonnet 4.6 — the productivity sweet spot for most production workloads
- Claude Haiku 4.5 — fastest and most cost-efficient for high-volume, latency-sensitive tasks
New models will launch simultaneously across the Claude Platform on AWS and Anthropic’s direct API — so you will not need to wait for a separate rollout cycle.
What this does not replace
Claude Platform on AWS does not replace Bedrock for every use case. Bedrock still makes sense when teams need multi-model routing, Bedrock-native features like Knowledge Bases and Bedrock Agents, AWS as the data processor, or strict regional data-residency controls. Claude Platform on AWS is better suited when access to Anthropic’s full native API surface and day-one feature availability matter more than keeping data within AWS’s infrastructure boundary.
Getting started
The Claude Platform on AWS is available now. Access is through the Claude Platform on AWS page and Anthropic’s developer documentation at platform.claude.com/docs; AWS customers should follow the documented onboarding flow for their account.
Note on private offers: If you already have a Bedrock private offer, check with your AWS or Anthropic account team before starting usage. Anthropic states that discounts cannot be applied retroactively to usage before a Claude Platform private offer is accepted.
If you are building a new AI workload today and data residency is not the dominant constraint, the Claude Platform on AWS is worth evaluating — particularly if you want access to the agentic capabilities and are already running workloads against AWS commitments.
The most interesting development is not the launch itself — it is what the full feature set signals about where production AI engineering is heading: away from raw API calls towards orchestrated, tool-using agents with structured memory, verifiable citations, and audit trails that your security team can actually work with.
Claude Platform on AWS was announced on May 11, 2026. More details: claude.com/blog/claude-platform-on-aws