The announcement, stripped down
Anthropic's Claude Platform is now generally available on AWS. This is not another model release. It is an infrastructure shift: AWS customers can now access the full native Claude API — every feature, every beta — through their existing AWS account, billed through AWS, authenticated with IAM.
The important detail: Anthropic operates the service. Your data is processed outside the AWS security boundary. This is the fundamental difference from Amazon Bedrock, where AWS is the data processor and everything stays within AWS infrastructure.
If that distinction sounds like a footnote, it is not. For any enterprise operating under data residency requirements — GDPR, HIPAA, NDPA, or sector-specific regulations — where the data sits and who processes it determines whether a project is feasible at all.
What you actually get
Claude Platform on AWS ships with the complete native API feature set. Not a subset. Not "coming soon." The full thing, with new features landing the same day they go live on Anthropic's direct API.
The headline capabilities:
Claude Managed Agents. Pre-built agent infrastructure that runs in Anthropic's managed containers. You define the agent (model, system prompt, tools, MCP servers), configure an environment (packages, network rules, mounted files), and start sessions. Claude autonomously executes tools, runs code, reads files, and streams results back via server-sent events. Long-running tasks that take minutes or hours — the kind that break synchronous API patterns — run natively.
This eliminates the most painful part of agent deployment: building the loop. The execution sandbox, tool routing, state management, error recovery, and prompt caching are all handled by the harness. For teams that have been building custom agent infrastructure on Lambda and ECS, this is a significant reduction in operational overhead.
MCP connector. Model Context Protocol lets agents connect to external systems — databases, APIs, file stores, internal tools — through a standardized interface. The connector handles authentication via a vault pattern: secrets stay in a pre-registered vault, each session authenticates with its own credentials, and reusable agent definitions never contain secrets. This is how you give an agent access to your CRM, your document store, or your internal ticketing system without writing custom integration code for each.
Code execution. Agents can write and run code in a sandboxed container during a session. Data analysis, file transformation, validation logic — anything that requires computation beyond what the model does natively. The container comes with Python, Node.js, Go, and standard packages pre-installed.
Skills and advisor strategy. Skills are reusable capability modules that extend what an agent can do. The advisor strategy (in beta) provides a pattern for coordinating multiple agents or adding a review layer to agent outputs. Both are in beta, but both signal Anthropic's direction: composable agent architectures that scale beyond single-prompt interactions.
Web search and fetch. Agents can search the web and retrieve page content during execution. For research tasks, competitive analysis, or any workflow that requires current information beyond the model's training data.
Prompt caching, citations, batch processing. The operational features that make production deployments viable. Prompt caching reduces costs on repeated context. Citations trace model outputs back to source documents. Batch processing handles high-volume workloads without per-request overhead.
Claude Console. Anthropic's development environment for prompt engineering, testing, and evaluation. Available through AWS authentication — no separate Anthropic account needed.
Three models, day one
The platform launches with Claude Opus 4.7, Sonnet 4.6, and Haiku 4.5. New model releases sync with the native API — no waiting for AWS availability.
For teams already using our multi-model architecture on Bedrock, this creates an interesting option: use Claude Platform for complex reasoning and agent workloads where you need the latest features, and Bedrock for embedding, classification, and tasks where regional data residency matters.
Platform vs Bedrock: the decision framework
This is the question every AWS customer building with Claude now faces. Both are valid. The right choice depends on your constraints.
Choose Claude Platform on AWS when:
- You need the latest features immediately. Managed agents, MCP connectors, skills, advisor strategy — these ship on Platform first and may never appear on Bedrock in the same form.
- You are building agent-based systems that require long-running execution, tool use, and code execution in managed containers.
- You want Anthropic operating the inference infrastructure. Their team optimizes model serving, handles scaling, and manages the runtime.
- Your compliance requirements allow data processing outside the AWS security boundary (Anthropic processes the data, not AWS).
Choose Amazon Bedrock when:
- You have strict regional data residency requirements. Bedrock processes data within AWS infrastructure, in the region you select.
- You need AWS as the data processor for regulatory or contractual reasons (GDPR data processing agreements, sector-specific regulations that mandate AWS as processor).
- You are using multiple foundation models through a single API and want Bedrock's unified interface across providers.
- You need AWS PrivateLink, VPC endpoints, or other AWS networking features for model inference.
Use both when:
- You route complex agent workloads through Claude Platform and simpler inference tasks through Bedrock.
- You use Bedrock for data-residency-sensitive workloads and Platform for everything else.
- You want Bedrock for embedding and classification (Titan, Cohere) alongside Claude Platform for reasoning and agent tasks.
This is not an either-or decision. Most enterprises we work with will end up using both, with a routing layer that sends each task to the right service based on data sensitivity, feature requirements, and cost.
What this means for enterprise teams
Three things matter about this announcement, regardless of where your team operates.
Billing consolidation. Claude Platform billing flows through AWS Marketplace. No separate Anthropic contract, no separate invoicing, no separate spend approval process. For procurement teams in regulated industries — banking, insurance, healthcare, government — this removes a real barrier. One vendor relationship, one billing stream, one set of commitment discounts.
Regional availability. Claude Platform launches in 18 regions across North America, Europe, Asia Pacific, and South America. For teams with latency-sensitive workloads, you can select the region closest to your users. For teams in regions without coverage yet — including Africa and parts of the Middle East — European endpoints (Frankfurt, Stockholm, London) provide sub-200ms latency for most workloads, with real-time applications being the exception to plan around.
Data processing and compliance. This is where the Platform vs Bedrock distinction gets concrete. Claude Platform has Anthropic as the data processor. Bedrock has AWS. Your compliance team needs to evaluate which processing arrangement fits your regulatory obligations — whether that is GDPR in Europe, HIPAA in the US, NDPA in Nigeria, POPIA in South Africa, or sector-specific frameworks like PCI-DSS, SOC 2, or ISO 27001. The architecture decision follows the compliance decision, not the other way around.
How we help
We have been building production AI systems on AWS — including multi-model Bedrock architectures, RAG pipelines, and managed agent deployments — across financial services, insurance, legal, and e-commerce clients. This announcement adds a new deployment surface, but the engineering patterns remain the same.
Architecture and deployment planning. We assess your workloads and design the right split between Claude Platform and Bedrock. Which tasks need managed agents? Which need data residency? Where does the routing layer sit? We build the decision framework specific to your regulatory environment and technical requirements.
Managed agent implementation. Claude Managed Agents eliminates the agent loop infrastructure, but you still need to design the agent: system prompts, tool definitions, MCP server configurations, environment setup, and session management. We build and deploy production agents that connect to your internal systems through MCP, execute domain-specific workflows, and operate within your security and compliance boundaries.
MCP server development. The MCP connector is only as useful as the servers it connects to. We build custom MCP servers that expose your internal systems — databases, APIs, document stores, CRMs, ticketing systems — to Claude agents through a standardized, secure interface. This is the integration layer that turns a general-purpose agent into one that actually operates within your business.
Migration and optimization. If you are already running Claude workloads on Bedrock, we help you evaluate which workloads benefit from moving to Claude Platform and handle the migration. If you are running custom agent infrastructure on Lambda and ECS, we help you consolidate onto Managed Agents where it makes sense — reducing operational overhead without losing control over critical workflows.
Compliance assessment. For regulated industries, we map your data flows against your regulatory obligations and determine the right processing arrangement. Anthropic as processor vs AWS as processor is not a technical decision — it is a compliance decision that depends on your specific regulatory context.
The bottom line
Claude Platform on AWS is the most significant change to enterprise AI deployment on AWS since Bedrock launched. It gives AWS customers full access to Anthropic's platform — including features that Bedrock does not and may never offer — while keeping everything within the AWS billing and identity ecosystem.
The decision is not "should we use this." It is "which workloads go where." That is an architecture question, a compliance question, and a cost optimization question. It is the kind of question we answer for clients every week.
If you are building AI systems on AWS and want to understand how Claude Platform fits your architecture, start a conversation with us.