Why Amazon Prime Video Cut Costs 90% by Dropping Serverless for a Monolith

Amazon Prime Video’s internal team migrated a monitoring system from a serverless micro‑service architecture, built with AWS Step Functions and Lambda, to a single‑process monolith running on EC2 and ECS, eliminating expensive services and S3 dependencies.

The original design used a highly scalable serverless orchestration to handle thousands of concurrent video streams, but Step Functions became a bottleneck and incurred per‑state‑transition costs, while frequent S3 calls added further expense.

By consolidating components into one process with a lightweight orchestration layer, the team achieved over a 90% reduction in infrastructure costs and improved scaling capabilities, even leveraging EC2 Savings Plans for additional savings.

Industry experts, including Ruby on Rails creator David Heinemeier Hansson and micro‑service author Sam Newman, highlighted that micro‑services and serverless are not universal solutions and should be adopted only after careful cost‑benefit analysis.

Amazon CTO Werner Vogels and former AWS architect Adrian Cockcroft emphasized that software architecture must be evolution‑driven, noting that no single pattern fits all scenarios and that “monoliths are not dinosaurs.”

The case sparked extensive discussion on platforms like Hacker News, Twitter, and Reddit, with many questioning the hype around micro‑services and the hidden costs of over‑engineered cloud solutions.