Using Claude on AWS | How to Use Amazon Bedrock and Pricing

What is the AWS and Anthropic Partnership?

Amazon has made a total strategic investment of $8 billion (approximately 1.2 trillion yen) in Anthropic, and the two companies are pursuing deep collaboration at both the product and infrastructure levels. At the core of this partnership is the integration of Claude into Amazon Bedrock. For Anthropic, Amazon is both one of its largest investors and a key training infrastructure partner, with large-scale training using AWS-proprietary accelerators such as Trainium and Inferentia already announced. This is establishing an environment where "the latest generation of Claude is delivered optimized on AWS."

Amazon Bedrock is a fully managed generative AI platform provided by AWS, allowing you to call multiple foundation models — including Claude — via API. It is valued by enterprises for enabling the use of Claude for custom data fine-tuning and knowledge base construction while maintaining AWS security governance.

What the Investment and Partnership Mean

The $8 billion investment represents more than just a financial relationship — it signals a long-term commitment that "AWS customers get priority access to Anthropic's latest models" and that "both companies' roadmaps will align at the product level." In fact, new Claude models become available on Bedrock quickly, and Claude is solidifying its position as a globally standard enterprise AI foundation. Additional factors in enterprise adoption decisions include the low concern about lock-in to a specific model, and the ease of building compliance frameworks aligned with AWS's shared responsibility model. On Anthropic's side, model availability continues to expand in step with AWS regional rollouts.

What Does It Mean to Use Claude on Amazon Bedrock?

The defining characteristic of Claude on Amazon Bedrock is its design to keep data from leaving the AWS infrastructure. For organizations that need to maintain clarity about where data resides when dealing with internal systems, code, or customer information, Bedrock naturally satisfies those requirements. You can integrate Claude as an extension of the governance controls already in everyday AWS use — including closed-network communication via VPC endpoints (PrivateLink), customer-managed key management with KMS, and full API call audit logging via CloudTrail.

• Handling of inference data: Customer inference data and training data are not stored by AWS unless explicitly configured, and communication occurs over the AWS backbone
• Region availability: Available worldwide, including the Tokyo and Osaka regions. Supports cross-region inference confined to within Japan (a mechanism that distributes inference processing across multiple regions), making it easy to adopt even in regulated industries such as finance and healthcare
• Claude model lineup: Claude Haiku, Sonnet, and Opus models can all be invoked through Bedrock. Models can be mixed and matched for cost optimization
• Integration with AWS services: Can be combined with Bedrock-specific features such as Guardrails (safety filters), Knowledge Bases (RAG: a method of supplementing answers by searching external data), Agents (task automation), and Flows (workflows)

There are two usage patterns: a flexible small-start pattern with pay-as-you-go pricing (via API), and a fixed-price purchase pattern through AWS Marketplace. For details, see the AWS Blog: "Claude Code on AWS Patterns Explained".

Checkpoints When Choosing Bedrock

When adopting Bedrock, first confirm these 3 points: (1) whether the model you plan to use is available in your target region, (2) whether cross-region inference is acceptable, and (3) whether you need to use supplementary features such as Guardrails or Knowledge Bases. Particularly in finance and healthcare, the requirement to "complete processing while keeping data within the country" is strict, and whether or not the Tokyo region is supported directly determines adoptability. Usage can be visualized alongside other services in AWS Cost Explorer, and being able to grasp the entire organization's AI investment efficiency on a single dashboard is another strength of Bedrock.

What is Claude Platform on AWS?

Claude Platform on AWS, announced in 2026, is a new service that provides access to Anthropic's first-party platform using AWS billing, authentication, and audit infrastructure. By subscribing via AWS Marketplace, you can use the Anthropic API linked to your existing AWS account, enabling adoption without significantly changing the operational workflows of procurement, accounting, or security teams.

The biggest difference from Bedrock is that inference runs on Anthropic's infrastructure. Data is processed outside the AWS boundary, but in exchange, you have access to all the latest native features that Anthropic provides. This suits organizations that want to "start using the latest Claude in production as quickly as possible" or "leverage Anthropic-built features like Memory and the Files API while consolidating procurement through AWS."

• Web Search: Claude autonomously searches the web to supplement its responses
• Memory: Retains memory across conversations
• Files API: Upload files and reference them over the long term
• Immediate access to the latest models: Use models released by Anthropic at the earliest possible time

Since billing and auditing can be managed uniformly from an AWS account, it also suits organizations that are "currently using the Anthropic API directly but want to consolidate expenses into AWS." Service details can be found on the AWS official page (Coming Soon).

Differentiation from Bedrock

Claude Platform on AWS is positioned not as a replacement for Bedrock, but as a complementary option. A useful mental model: Bedrock is for enterprises that prioritize infrastructure control, while Claude Platform on AWS is for those that prioritize functionality and immediate access to the latest features. Usage patterns such as "existing users who have been contracting the Anthropic API directly migrate under AWS for improved governance" or "running Bedrock in production while trying the latest features of Claude Platform on AWS in the PoC phase" are expected to increase going forward.

Comparison of Amazon Bedrock and Claude Platform on AWS

Source: DevelopersIO "Differences between Claude Platform on AWS and Bedrock"

How to Read This Table

The two most impactful columns for decision-making are "where inference runs" and "scope of access to Anthropic's latest features." Industries with strict constraints on data location tend to default to Bedrock, while teams for whom features like Web Search or Memory are business-critical will find Claude Platform on AWS the natural candidate. Since both services allow unified billing and auditing through AWS, a major advantage is that "from an accounting perspective, either choice looks the same." In technical selection discussions, locking in which takes priority — functional requirements or compliance requirements — at the outset allows the conversation to converge quickly.

Which One Should You Choose?

Cases Where You Should Choose Amazon Bedrock

• Regulated industries such as finance, healthcare, or public sector where data cannot leave AWS infrastructure
• You want to run multi-model operations with models other than Claude (Nova, Llama, Mistral, etc.)
• You want to combine AWS-managed AI features such as Guardrails, Knowledge Bases, and Agents
• Your AWS environment is already well-established and you prioritize integration with IAM, VPC, and CloudTrail

Cases Where You Should Choose Claude Platform on AWS

• You want to use all of Anthropic's native features such as Web Search and Memory
• You want same-day access to Anthropic's latest models
• You are currently using the Anthropic API directly but want to consolidate into AWS billing
• You will only use Claude and want to maximize the Anthropic experience

Priority Order When You're Unsure

Applying decision criteria in the order "data location → feature requirements → cost" makes it easier to reach a sound conclusion without overcomplicating the decision.

1. Data location: Confirm whether it is operationally and regulatorily acceptable to process data outside AWS infrastructure. If not acceptable, Bedrock is effectively the only option.
2. Feature requirements: Evaluate whether Anthropic-native features (Web Search, Memory, Files API) are essential to your operations. If they are essential, Claude Platform on AWS becomes a candidate.
3. Cost: Estimate monthly token consumption and future growth, then compare whether Marketplace fixed-price contracts or pay-as-you-go pricing is a better fit for your organization's budget management.

Using Claude Code on AWS for Organizations

There is growing adoption of Claude Code (Anthropic's coding assistance tool) deployed across organizations via Amazon Bedrock. Through Bedrock:

1. Security: Designs can keep code and business information from leaving AWS
2. Cost management: Centrally manage usage in AWS Cost Explorer
3. Access control: Fine-grained per-developer access control using IAM roles
4. Scalability: Gradually scale from startups to large enterprises with pay-as-you-go pricing

Detailed benefits for organizational adoption, with real-world examples, are explained in the AWS Startup Blog.

Operational Points When Adopting Claude Code

When deploying Claude Code across an organization via Bedrock, keeping the following 4 points in mind will allow for safe operations.

• Separate permissions with IAM roles: The standard approach is to design separate IAM roles per developer to isolate usage and access permissions.
• Visualize with tags in Cost Explorer: By tagging at the project level and visualizing per-team token consumption in Cost Explorer or Cost & Usage Reports, you can identify opportunities to improve usage efficiency.
• Mask sensitive information with Guardrails: Combined with Bedrock Guardrails, you can apply uniform policies such as "automatically mask confidential organizational information if it appears in a prompt."
• Small-scale validation → phased rollout: A safe approach is to solidify operational guidelines with a small validation team early in adoption, then roll out to departments in stages.

Steps to Get Started with Claude on AWS

Getting Started with Amazon Bedrock

1. Sign in to the AWS Management Console
2. Navigate to the Amazon Bedrock service page
3. Request access to Anthropic's Claude models under "Model access"
4. After approval, invoke Claude from the Bedrock Playground or via API

With just an AWS account, you can start experimenting in minutes. Pricing is pay-as-you-go based on input and output token counts, with Claude Haiku being the most affordable (followed by Sonnet and Opus in order of capability and price).

Getting Started with Claude Platform on AWS

As of 2026, the service is in a pre-launch announcement phase, so it is recommended to register for the waitlist at aws.amazon.com/claude-platform/. To receive GA notification when the service launches, organizing the AWS accounts your organization actually plans to use (production and testing) in advance will allow you to proceed immediately to Marketplace contracts after release.

Notes for Existing Anthropic API Users Migrating

Organizations currently using the Anthropic API directly will need to plan the handling of API keys and the timing of billing account switchover. Switching to Claude Platform on AWS is best understood as: "the API endpoint itself doesn't change, but the billing and authentication routes move under AWS." Setting a parallel operation period to rebuild logging, monitoring, and budget alerts on the AWS side before migrating production traffic will reduce the risk of disruptions during the switchover.

Summary

"Using Claude on AWS" offers two paths: Amazon Bedrock (data residency focus, multi-model, AWS feature integration) and Claude Platform on AWS (full access to Anthropic's latest features, AWS billing integration). Bedrock is well-suited for regulated industries and large-scale enterprise deployments, while Claude Platform on AWS is a strong option for organizations prioritizing Anthropic's latest features. Both share AWS authentication, billing, and auditing, making them easy to combine with existing AWS governance frameworks.

The choice between the two paths becomes clearer when you anchor your decision on two questions: "where will data be processed?" and "which features are required?" Many organizations will likely start by advancing internal adoption with Bedrock, then settle into a combined model that supplements with Claude Platform on AWS for the latest features as needed. The AWS-Anthropic partnership is still evolving, and the feature differences and pricing structures of both platforms are expected to continue expanding and adjusting. For the latest information, it is advisable to regularly check the AWS official Claude Platform page and the AWS Blog.