Anthropic's Claude Platform Now Available on AWS (2026)
TL;DR
- AWS is now the first cloud provider offering native Claude Platform access — developers can use Anthropic’s APIs and tools directly through AWS credentials
- Data stays outside AWS’s security boundary — Claude Platform requests are processed by Anthropic, making this unsuitable for organizations with strict data residency requirements
- Different from Amazon Bedrock — Claude on Bedrock keeps all data within AWS boundaries, while Claude Platform on AWS does not
- Same pricing, expanded monitoring — costs match Anthropic’s direct pricing, but you gain AWS CloudTrail integration for auditing
What Happened
AWS launched general availability of Anthropic’s Claude Platform on Monday, following the companies’ expanded partnership announcement in April. Developers can now access Claude’s full suite of developer tools and APIs — including the Messages API, Claude Managed Agents (beta), web search, and code execution — using their existing AWS credentials.
The integration makes AWS the first cloud provider to offer the native Claude Platform experience. Supported features include advisor tools (beta), MCP connector (beta), Agent Skills (beta), the files API (beta), and web fetch capabilities.
But there’s a critical architectural detail: Anthropic still operates the underlying Claude Platform infrastructure. Your requests and data get processed outside AWS’s security boundary, which fundamentally changes who can use this service.
Why It Matters
This launch creates a confusing but important choice for AWS customers: use Claude Platform on AWS or stick with Claude on Amazon Bedrock. The difference matters because of where your data lives.
Claude on Amazon Bedrock keeps all processing within AWS’s security perimeter. If your organization has regulatory requirements about data residency — healthcare, financial services, government work — Bedrock is your only compliant option.
Claude Platform on AWS routes requests to Anthropic’s infrastructure outside AWS boundaries. AWS explicitly states this “makes it well-suited for teams without specific regional data residency requirements.” Translation: if you’re in a regulated industry, this probably isn’t for you.
The integration does streamline authentication and billing through AWS’s existing systems. For teams already using AWS CloudTrail for compliance monitoring, you now get built-in auditing for Claude API usage. That’s valuable for tracking AI usage patterns and costs across your organization.
Key Details
Supported Features (as of GA):
- Messages API (full)
- Claude Managed Agents (beta)
- Advisor tool (beta)
- Web search and web fetch
- MCP connector (beta)
- Agent Skills (beta)
- Code execution
- Files API (beta)
Pricing: Identical to direct Anthropic Claude Platform pricing
Data Processing: Outside AWS security boundary (Anthropic infrastructure)
Authentication: Standard AWS credentials and IAM
Monitoring: AWS CloudTrail integration for auditing
Availability: General availability now
Comparison to Bedrock:
| Feature | Claude Platform on AWS | Claude on Amazon Bedrock |
|---|---|---|
| Data location | Outside AWS boundary | Within AWS boundary |
| Feature completeness | Full Claude Platform APIs | Model access only |
| Suitable for regulated industries | No | Yes |
| CloudTrail integration | Yes | Yes |
| Authentication | AWS credentials | AWS credentials |
Implications
This move signals AWS’s strategy to offer multiple integration paths for leading AI models, letting customers choose between feature completeness and compliance boundaries. It’s a pragmatic acknowledgment that not every workload requires fortress-level data residency.
The partnership’s scale matters more than the technical integration. Anthropic committed to over $100 billion in AWS compute capacity over the next decade, with access to up to 5GW of capacity and AWS’s Trainium chips. That’s a direct response to Anthropic’s well-documented capacity constraints that have frustrated developers during peak usage periods.
For the broader AI infrastructure landscape, this establishes a pattern: foundation model companies maintaining control of their platforms while offering cloud-native access through hyperscalers. It’s different from the fully integrated approach of Azure OpenAI Service, where Microsoft operates the infrastructure.
The deal also includes less publicized benefits. Amazon’s internal developers recently gained access to Claude Code alongside AWS’s own Kiro coding tool — a sign that AWS sees value in offering developers choice rather than forcing a single internal solution.
Our Take
This launch exposes a fundamental tension in enterprise AI: the gap between what developers want (full API access, latest features) and what compliance teams need (data residency guarantees).
AWS is betting both sides win — developers without regulatory constraints get the full Claude Platform experience, while regulated workloads stick with Bedrock. That’s probably correct, but it creates complexity. IT teams now need to maintain clear governance about which Claude integration to use for which workloads.
The real story is the $100 billion compute commitment. Anthropic’s capacity issues have been the company’s biggest operational weakness, directly impacting developer experience and enterprise adoption. This partnership doesn’t just give Anthropic infrastructure — it gives them 10 years of predictable scaling capacity.
Watch for how this affects Claude’s competitive position against GPT-4 and Gemini. Reliable capacity matters as much as model quality for enterprise adoption. If Anthropic can maintain consistent availability while continuing to improve Claude’s capabilities, AWS’s infrastructure advantage becomes Anthropic’s competitive moat.
The architecture choice — keeping Claude Platform outside AWS boundaries — is controversial but honest. Anthropic maintains control over its platform while AWS gets to offer it as a service. That’s better than forcing everything through Bedrock’s more constrained model-access-only approach, even if it creates compliance headaches for some customers.