The short answer
Amazon Redshift is available in two distinct deployment models: traditional Provisioned Clusters and Redshift Serverless. Understanding the trade-offs between predictable flat-rate billing and elastic demand-driven scaling is essential for data platform design.
Redshift Serverless
In the serverless model, you do not configure compute nodes, manage cluster resizing, or patch maintenance windows. Redshift automatically provisions and scales compute capacity based on incoming query complexity.
Compute is measured in Redshift Processing Units (RPUs). You configure a base capacity (such as 32 RPUs to 512 RPUs) and an optional maximum cap to prevent runaway bills. Billing is strictly per-second while queries are actively running. When zero queries run, your compute cost drops to exactly zero.
Provisioned RA3 clusters
In a provisioned cluster, you choose an exact node family (such as ra3.4xlarge or ra3.16xlarge) and count. The cluster runs 24 hours a day, 7 days a week, incurring a fixed hourly fee regardless of whether queries are executing.
To reduce costs, AWS offers 1-year and 3-year Reserved Instance (RI) commitments, providing discounts between 30% and 60% compared to on-demand pricing.
Workload Profile Recommended Architecture
---------------------------------------------------------
Unpredictable ad-hoc queries Redshift Serverless
Intermittent daytime reporting Redshift Serverless
24/7 high-frequency production Provisioned RA3 (Reserved)
Predictable batch ETL runs Provisioned RA3 (Reserved)Decision criteria
Choose Serverless for development environments, departmental analytics with variable schedules, and sporadic workloads where clusters would otherwise sit idle overnight. Choose provisioned RA3 with Reserved Instances for mature enterprise platforms with continuous 24/7 query traffic, where predictable fixed monthly costs beat variable per-second billing.