Amazon Redshift

Use it when

• The organization is committed to AWS; VPC networking, IAM, Glue, and existing S3 data make Redshift the shortest path.
• Zero-ETL from Aurora and RDS, or streaming ingest from Kinesis and MSK straight into a materialized view, removes pipelines you would otherwise build.
• Usage is spiky and you want Serverless RPUs billed per second, or steady and you want provisioned RA3 nodes at a predictable rate.
• The team already speaks PostgreSQL-flavored SQL.

Think twice when

• Multi-cloud or an exit path matters; Redshift is AWS only and its storage format is proprietary, though it reads S3 and Iceberg in place.
• Concurrency spikes are constant; concurrency scaling adds transient clusters, but engines built for high QPS serving (ClickHouse, Pinot) fit that shape better.
• You want open table formats as the primary store; Databricks SQL and lakehouse stacks keep the data engine-neutral.

How it runs

Fully managed on AWS in two compute models over the same managed storage: provisioned RA3 clusters (hot data on local SSD, the rest tiered to S3) or Serverless capacity in RPUs. Spectrum queries S3 in place; WLM queues and concurrency scaling manage load. Clients connect over JDBC, ODBC, and the Data API.