AWS Introduces Automation for Managing Private AI Model Pricing on Amazon Bedrock

Amazon Web Services (AWS) has unveiled new tools designed to streamline the management of private pricing offers for artificial intelligence models accessed through Amazon Bedrock. The move aims to reduce operational complexities for organizations managing AI model subscriptions across multiple AWS accounts, particularly when dealing with custom pricing agreements facilitated by AWS Marketplace Private Pricing Offers (MPPOs).

Simplifying Multi-Account Access to Custom AI Pricing

Previously, enterprises with numerous AWS accounts faced the cumbersome task of individually subscribing to and accepting MPPOs for each workload account. This process created significant operational friction. To address this, AWS is enhancing Amazon Bedrock with managed entitlements, a feature that allows organizations to subscribe to third-party AI models once from a central account and then distribute access across their entire organization.

The Chenab Times has learned that AWS is providing two primary methods for activating these MPPO entitlements within AWS License Manager. The first is a lightweight, one-time script for immediate activation of trusted licenses. The second is a more robust, event-driven automation solution built using the AWS Cloud Development Kit (AWS CDK), designed for continuous verification and grant management.

Automating Grant Activation for Negotiated Rates

AWS Marketplace Private Pricing Offers enable customers to negotiate custom rates for various third-party models available on Amazon Bedrock. Once a private offer is accepted, AWS Marketplace generates a license within AWS License Manager. For organizations utilizing AWS Organizations with all features enabled, these organization-level grants are automatically accepted but remain inactive until manually activated. This manual activation process, often done through the AWS License Manager console, can be time-consuming and prone to oversight, meaning negotiated pricing does not take effect until the grant is actively enabled.

The need for automation becomes more pronounced as organizations adopt new models or renew existing agreements, as each instance requires similar manual intervention. This can lead to delays in realizing cost benefits from negotiated pricing.

Two Paths to Automated Entitlement Management

The provided sample repository offers two distinct approaches to automate the activation of these entitlements.

Path 1: The Lightweight Script

This standalone script is intended for users who require a quick, infrastructure-light solution for activating licenses they already trust. It lists all received licenses and generates a dry-run plan, allowing for review before implementation. To apply the changes and activate organization-wide grants, users re-run the script with specific confirmation flags. This method is recommended only when all received licenses in the management account are trusted, as it lacks a seller allow-list.

Implementation involves cloning the repository, running the script in dry-run mode to review licenses and their issuers, and then executing the script with confirmation flags to apply the changes. Post-activation, users are advised to validate the billing path to ensure the negotiated discount is correctly applied, potentially using a provided utility script.

Path 2: Event-Driven CDK Automation

For organizations seeking ongoing, automated management of private pricing agreements, the AWS CDK approach offers a more comprehensive solution. This event-driven automation deploys within the AWS management account and continuously monitors for new AWS Marketplace agreement events. Key components include an Amazon EventBridge rule that triggers an AWS Lambda function upon the creation of a purchase agreement.

The Lambda function then verifies the seller against an allow-list, locates the relevant license, creates a distributed grant, and attempts to activate it. For auditability, all actions and outcomes are logged to Amazon CloudWatch Logs. The grants can be targeted to the entire organization, specific organizational units (OUs), or individual accounts. Negotiated pricing becomes effective only after the grant reaches an active state.

Prerequisites for this path include using the AWS Organizations management account in the us-east-1 region, having Python 3.12+ and AWS CDK v2 installed, and ensuring AWS Organizations is configured with all features enabled. The solution also specifies that it is applicable to License Manager-backed Bedrock models, excluding Software as a Service (SaaS) products.

The implementation steps for the CDK path involve cloning the repository, bootstrapping the necessary AWS organization prerequisites, configuring deployment inputs including a seller allow-list, deploying the CDK stack, and seeding the Amazon DynamoDB table with verified sellers. The system is then ready to process eligible agreement events, with grant activation contingent on seller verification and License Manager’s processing workflow.

Benefits and Customer Endorsement

This automation is expected to accelerate the application of negotiated pricing, ensure consistent handling of grants across different accounts, and provide a robust audit trail through CloudWatch Logs. Marcus Lau, Head of Platform Engineering at Heidi Health, highlighted the value of private pricing and automated grant management, stating, “Private pricing on AWS Marketplace gives us a cost-effective and efficient way to manage that growth: we can granularly attribute usage and pricing across AWS accounts and organizational units, and automate grant management as we adopt new models. That lets our platform team scale cost governance without scaling manual work.”

AWS provides a GitHub repository with the sample code for managed entitlements, along with links to relevant AWS documentation for License Manager, Amazon EventBridge, and AWS CDK, to help users get started.

---

---

Follow us on

---

(We don't allow anyone to copy content. For Copyright or Use of Content related questions, visit here.)