Amazon Redshift

Overview

Amazon Redshift is AWS's petabyte-scale data warehouse, and the first decision is which shape of it to run. Serverless provisions and scales capacity automatically and charges nothing while idle; provisioned clusters are nodes you size and manage yourself, which still suits steady, predictable workloads.

Underneath it is a columnar store built for analytics, with specialised compression encodings and no secondary indexes, so it behaves nothing like the transactional databases it superficially resembles.

Node families matter for cost. RG is the current Graviton generation, which AWS states is up to 2.4 times as fast as the previous RA3 at 30% lower price; RA3 is still sold, and both keep data in Redshift Managed Storage so storage grows independently of compute.

Concurrency is handled by concurrency scaling, free for an hour a day per cluster, which adds transient capacity when queues build, alongside result caching, materialised views and automatic workload management.

Getting data in is increasingly zero-ETL rather than pipelines: managed integrations from Aurora, RDS, DynamoDB and enterprise applications, plus streaming ingestion straight from Kinesis and Kafka.

On the lake side it reads Iceberg, Hudi and Delta Lake. Iceberg tables catalogued in AWS Glue are queried transactionally and writes are supported, though time-travel queries on Iceberg are not.

The SQL dialect is PostgreSQL-derived, with semi-structured data handled by the SUPER type and PartiQL. One date to diary: AWS has published that Python UDFs lose support after 30 June 2026.

AI arrives two ways, Redshift ML training models through SageMaker Autopilot from a CREATE MODEL statement, and Bedrock foundation models called directly from SQL.

Key features and capabilities

The same headings are used for every cloud data warehouses entry, so two tools can be read side by side.

Architecture and storage

• Serverless auto-provisions capacity and costs nothing when idle; provisioned clusters are sized by you
• Columnar storage with compression encodings and no secondary indexes
• Redshift Managed Storage separates storage from compute on RG and RA3
• Multi-AZ deployment with a 99.99% availability commitment

Performance and scaling

• Concurrency scaling absorbs bursts, free for up to an hour a day per cluster
• Result caching for repeated queries, short query acceleration and automatic workload management
• Materialised views with incremental refresh, and layouts that sort themselves by query pattern
• Serverless uses AWS's own scaling and optimisation models rather than manual tuning

SQL and workloads

• PostgreSQL-derived SQL, with transactional features such as secondary indexes deliberately absent
• Semi-structured data through the SUPER type and PartiQL, with dynamic typing and lax semantics
• Streaming ingestion from Kinesis and Kafka, auto-copy from S3, and federated queries
• Python UDFs lose support after 30 June 2026, per AWS's own notice

Open formats and sharing

• Iceberg V1, V2 and V3 tables queried from the AWS Glue Data Catalog, with writes supported
• Iceberg metadata statistics feed the planner and partitions are detected automatically
• Time-travel queries are not supported on Iceberg tables
• Hudi and Delta Lake are also readable, and data sharing works across warehouses and lake tables
• Spectrum scan charges apply on RA3 and DC2 but not on RG or Serverless

Security and governance

• Row-level and column-level access control, integrated with Lake Formation
• Dynamic data masking
• TLS in transit and AES-256 at rest
• Single sign-on through IAM Identity Center
• Named compliance certifications are not listed on the features page

AI and machine learning

• Redshift ML trains models through SageMaker Autopilot from a CREATE MODEL statement
• Bedrock models, including Claude and Titan, called from SQL for text, sentiment and translation
• No extra Redshift charge for Redshift ML, though SageMaker training bills separately
• Amazon Q turns plain English into SQL in query editor v2

Where it runs and what it costs

• AWS only, with published prices across 36 regions including GovCloud
• No self-hosted option
• Billed per RPU-hour on Serverless or per node-hour on provisioned clusters
• Managed storage and Spectrum scans are charged separately

Pricing

Usage-basedServerless at $0.375 an RPU-hour

Serverless costs $0.375 per RPU-hour in US East, billed per second with a 60-second minimum and a four-RPU floor, and includes concurrency scaling. Provisioned nodes are charged per hour, from $0.543 for ra3.large to $13.04 for ra3.16xlarge, with reserved instances for savings; RG hourly rates are not in the published price list. Managed storage is $0.024 a GB a month and Spectrum scans $5.00 a terabyte. Concurrency scaling is free for an hour a day. New accounts get $300 in credits, expiring after 90 days.

Vendor pricing page →

Demos and videos

• Getting started with Amazon Redshift Serverless Tutorial · YouTube
• Amazon Redshift Serverless explained in 90 seconds Product demo · YouTube

About Amazon Web Services

Redshift is a first-party AWS service with no separate ownership or funding of its own. AWS announced it at re:Invent in November 2012 and opened it to all customers in February 2013, pitching launch economics of under $1,000 per terabyte per year, which reset warehouse pricing expectations at the time. It is proprietary, with no open-source or self-hosted edition, billed hourly or per second with optional reserved instances. Openness is delivered by reading and writing open table formats rather than by opening the engine.

aws.amazon.com