Jamf Establishes Real-Time System to Limit Amazon Bedrock Spending

Jamf, which provides Apple device management and security, has implemented a new real-time spending control mechanism to keep artificial intelligence costs under control for engineers using Amazon Bedrock.

Artificial Intelligence Cost Problem

Generative artificial intelligence spending exhibits a different structure from traditional IT costs. While traditional costs scale based on capacity, artificial intelligence expenditures vary directly depending on user behavior.

Intensive work carried out by engineers on premium models can quickly generate high costs. This situation makes it difficult for expenditures to be visible before the billing period.

Developed Infrastructure Solution

Jamf has established a new system in the production environment to achieve periodic cost visibility while providing its engineering team with access to Amazon Bedrock.

This system brings together AWS IAM Customer Managed Policies, Amazon Athena cost views, and a serverless AWS Lambda auditing loop.

Gradual Model Restrictions

The developed solution tracks each engineer's daily Amazon Bedrock spending and applies gradual model restrictions when approaching budget limits.

When eighty percent of the daily budget is reached, access to certain models is restricted while lower-cost models are left open, ensuring that engineers can continue their work without interruption.

Serverless Architecture Components

The system's architecture utilizes Amazon S3 for call logging, Amazon Athena to calculate daily expenditures, and Amazon EventBridge components to trigger the audit handler.

Thanks to IAM policy versioning features, live access evaluations are performed instantly without interrupting active sessions.