Compute costs are often the largest portion of a managed Postgres bill, yet they receive less attention than storage. This post breaks down how Amazon RDS bills for compute (instance class, on-demand vs reserved, single vs multi-AZ) and walks through strategies to reduce costs: rightsizing instances, manually downsizing or pausing, query optimization, fixing ORM N+1 problems, and using caching. It then contrasts these manual, error-prone approaches with Neon's autoscaling and scale-to-zero features, which dynamically match compute to actual load without downtime or manual intervention. Neon's approach also simplifies one-database-per-tenant architectures and makes dev/test instances cheap by scaling them to zero when idle.
