Why Prime Video Dropped Microservices for a Monolith and Cut Costs 90%

Prime Video’s engineering team published a candid case study that surprised the developer community by revealing how they migrated from a micro‑service architecture to a monolithic design to dramatically reduce costs.

The original system, built to monitor “every character stream” viewed by customers, required massive scalability for thousands of concurrent streams and initially relied on AWS Step Functions to orchestrate distributed components.

However, Step Functions became a bottleneck, incurring per‑state‑transition charges and high S3 request costs, leading the team to conclude that the initial architecture was expensive and poorly scalable.

By eliminating the distributed approach, packaging all components into a single process, and running the service on EC2 and ECS with a lightweight orchestration layer, they achieved over a 90 % reduction in infrastructure spend while improving scalability.

The study notes that while micro‑services and serverless can operate at scale, their suitability depends on the specific use case; a monolith proved more cost‑effective for Prime Video’s workload.

Comments from the community, including Hacker News users and experts like David Heinemeier Hansson and Sam Newman, echo the view that micro‑service hype often overlooks operational complexity and cost, and that architects should perform value‑chain analysis before defaulting to distributed designs.

Overall, the paper offers a rare, honest examination of how simplifying architecture can lower cloud expenses and serve as a practical reference for teams using AWS or other cloud providers.