Amazon Redshift pricing: What to expect and how it compares

Choosing a cloud data warehouse (CDW) is a strategic decision with significant financial implications. Among the leading contenders, Amazon Redshift stands out for its ability to scale efficiently and process petabyte-level datasets. However, maximizing the cost-effectiveness of Redshift deployments requires a thorough understanding of the Redshift pricing model.

Navigating Redshift pricing can be challenging. Compute, managed storage, and the choice between provisioned (node-based) and serverless (RPU-based) options create a complex cost landscape. This post demystifies the billing structure—compute, storage, and add-ons like Spectrum, Concurrency Scaling, ML, and zero-ETL—and shares practical optimization strategies.

Main takeaways

• Redshift offers flexible pricing across provisioned and serverless models; reserved nodes can materially reduce provisioned costs.
• RA3 nodes with Redshift Managed Storage (RMS) decouple compute and storage for better scalability and cost control; RMS in us-east-1 is $0.024/GB-month.
• Redshift Serverless bills per RPU-hour (per second with a 60-second minimum), with a 4-RPU base capacity option that can start around $1.50/hour in us-east-1 examples; Spectrum and Concurrency Scaling are included in RPU billing.
• Additional charges (for provisioned clusters) include Spectrum (priced by data scanned) and Concurrency Scaling beyond free credits; these can add up if not monitored.
• Cost optimization (rightsizing or moving to RA3/Serverless, storage hygiene, and workload tuning) can significantly reduce spend without harming performance.

What is Amazon Redshift?

Amazon Redshift is a fully managed, petabyte-scale data warehouse from AWS designed for large-scale analytics and BI. It integrates with the broader AWS ecosystem (S3, Glue Data Catalog, IAM, SageMaker, and more) and supports both provisioned clusters and serverless.

Core benefits

• Scalability: Scale compute and storage independently (RA3 + RMS) or elastically with Serverless base capacity and autoscaling.
• High performance: Columnar storage and massively parallel processing (MPP) across nodes or serverless capacity.
• Cost flexibility: On-demand, reserved (provisioned), or serverless (RPU) options.
• Fully managed: Automated backups, patching, and deep integrations with AWS services.

Modern architecture (what to know now)

• Leader node and compute nodes (provisioned): The leader coordinates queries; compute nodes execute in parallel (MPP).
• Storage: For RA3, Redshift uses local SSDs for hot data and S3-backed Redshift Managed Storage (RMS) for scale and cost efficiency.
• Serverless: Capacity is expressed in Redshift Processing Units (RPUs); you set a base capacity (commonly 4–512 RPUs) and pay per RPU-hour, per second.
• External data: Redshift Spectrum lets you query open-format files in S3 (Parquet/ORC/CSV) without loading. Use Spectrum for lakehouse-style querying. Load data via COPY (e.g., from S3) or with services like AWS DMS.

Amazon Redshift pricing components

Quick reference table

* Hourly rates change; confirm in the AWS pricing page for your Region, node size, and purchase option.

Compute costs (provisioned clusters)

The current default for new provisioned clusters is RA3: You size compute via node family/size and pay separately for storage via RMS. DC2 is a previous-generation SSD option (appropriate only for small, specific cases); DS2 is retired for new creation. For consistent workloads, reserved nodes can deliver ~41%–76% savings depending on term and payment option. Pause/Resume helps trim on-demand spend when clusters are idle; billing for provisioned compute is per-second once state changes take effect.

Key points to remember

• You’re billed per active node-second (rounded) for provisioned clusters; storage on RA3 is billed separately via RMS.
• Data transfer charges may apply for cross-Region traffic or non-S3 pathways; S3↔Redshift in-Region for COPY/UNLOAD/backup/restore is free.

Redshift Serverless pricing and when to use it

How it bills: Serverless uses RPU-hours with per-second metering (60-second minimum). You set a base capacity (commonly 4–512 RPUs), and scaling capacity is billed at the same RPU rate. In us-east-1 examples, the published RPU rate is about $0.375/RPU-hr; with 4 RPUs that’s roughly $1.50 per hour of active time. Spectrum and Concurrency Scaling are included in the Serverless price.

Good fits: Infrequent, spiky, or unpredictable workloads; ad-hoc analytics; dev/test; external data sharing. You don’t manage clusters, and you don’t pay when idle beyond the minimum billed seconds per execution window.

Redshift storage types and costs

• Redshift Managed Storage (RMS): Billed hourly and aggregated monthly as $0.024/GB-month in us-east-1 (varies by Region). Applies to RA3 and Serverless. Snapshots and backups are separate.
• Backups and snapshots: Automated snapshots are retained per your policy; manual snapshots and long-retention incur S3 charges. (Serverless uses standard backup rates for snapshots.)

Other Redshift charges to watch

Spectrum (provisioned clusters)

You’re charged $5 per TB scanned, rounded up to the next MB (10 MB minimum per query). Use columnar formats (Parquet/ORC), compression, and partitioning to reduce scan size. With Serverless, S3 queries are covered under RPU billing, not a separate Spectrum line item.

Concurrency Scaling (provisioned clusters)

Each cluster earns 1 free hour per 24 hours (bank up to 30 hours). Excess usage is billed per-second at your cluster’s on-demand rate (1-minute minimum per activation). The free credits cover the majority of customers’ usage.

Redshift ML

Train and use ML models via SQL. Training is billed by “cells” processed with published tiers ($20/M, $15/M, $7/M) plus small S3 artifact charges. Some usage may qualify for the SageMaker free tier initially.

Zero-ETL (Aurora ↔ Redshift)

No additional fee: you pay for Aurora and Redshift resources (storage, I/O, transfer). This enables near-real-time analytics without building ETL pipelines.

Data transfer

$0 between S3 and Redshift in the same Region for COPY/UNLOAD/backup/restore; other transfers (e.g., cross-Region, JDBC/ODBC over public internet or VPC egress) follow standard AWS data transfer pricing.

How to reduce Redshift costs without sacrificing performance

1. Rightsize (or migrate) to RA3 Use RA3 for modern workloads so you can scale compute and storage independently via RMS. If you’re still on DC2, plan a migration path. Consider reserved nodes for steady usage.
2. Leverage Serverless for variable workloads For spiky or ad-hoc loads, Serverless eliminates idle cluster cost. Tune SQL, apply workload management to control concurrency and priorities, and set budget/usage alerts.
3. Optimize storage Purge stale data, set snapshot lifecycles, and archive cold data to S3 (query via Spectrum) to reduce RMS and snapshot costs. Partition large tables to shrink scans.

Power Redshift with clean, cost-effective pipelines

Understanding Redshift pricing is just the first step. RudderStack delivers clean, schema-enforced data directly to Redshift—helping your queries run leaner and cheaper. RudderStack is customer data infrastructure that’s data-cloud native and integrates tightly with Snowflake and Redshift. Explore the data foundation solution.

Cut Redshift costs without slowing down your pipelines Struggling with rising Redshift bills? The problem often isn’t Redshift, but the quality of the data flowing into it. RudderStack delivers clean, schema-consistent data that reduces query overhead, eliminates inefficiencies, and keeps your warehouse lean. Request a demo.

Discover how RudderStack makes Redshift more efficient Amazon Redshift offers powerful analytics capabilities with flexible pricing, but navigating its costs requires a thoughtful strategy. From compute and storage tiers to serverless options, each element of Redshift pricing impacts total cost of ownership. To capitalize fully, your pipelines should be clean, consistent, and cost-aware—where RudderStack helps with real-time event streaming, schema enforcement, and Reverse ETL into Redshift.

Request a demo to learn more.

FAQs

How is Amazon Redshift priced overall?

You pay for compute and storage, with optional charges for features like Spectrum and Concurrency Scaling on provisioned clusters. You can choose provisioned nodes or Redshift Serverless. Rates vary by Region and can change.

What is the difference between provisioned and serverless pricing?

Provisioned bills by node-hours for your cluster. Serverless bills by RPU-hours with per-second metering after a 60-second minimum. Serverless includes Spectrum and Concurrency Scaling in the RPU price.

How do RA3 nodes and Redshift Managed Storage reduce costs?

RA3 decouples compute and storage. You size compute with RA3 nodes and pay storage separately via RMS. This helps you scale storage without over-provisioning compute.

What does Redshift Serverless cost?

Pricing is based on RPU-hours with per-second billing. Documentation for us-east-1 cites about $0.375 per RPU-hour. A 4-RPU base can be roughly $1.50 per active hour. Always confirm for your Region.

How much does Redshift Managed Storage cost?

RMS is billed per GB-month. An example reference is $0.024 per GB-month in us-east-1. Charges vary by Region and snapshot retention.

When do Spectrum charges apply?

On provisioned clusters, Spectrum bills by data scanned, with a reference of $5 per TB and a 10 MB minimum per query. Under Serverless, Spectrum usage is included in RPU billing.

How does Concurrency Scaling pricing work?

Provisioned clusters earn one free hour per 24 hours and can bank up to 30 hours. Excess usage is billed per second at your on-demand cluster rate with a one-minute minimum per activation.

Are there extra charges for Redshift ML and zero-ETL?

Redshift ML training is priced by cells processed with tiered rates and small S3 artifact costs. Zero-ETL between Aurora and Redshift has no separate fee. You pay for the underlying services.

What data transfer costs should I expect?

S3 to Redshift transfers in the same Region are $0 for COPY, UNLOAD, backup, and restore. Cross-Region and other network paths follow standard AWS data transfer pricing.

How do reserved nodes impact cost?

For steady workloads on provisioned clusters, one- or three-year reserved nodes can materially reduce cost compared to on-demand. Typical savings references range from about 41% to 76% depending on term and payment option.

When should I use Serverless instead of provisioned?

Choose Serverless for spiky or unpredictable workloads, ad-hoc analytics, and dev or test. You avoid paying for idle clusters and still get autoscaling within your base and limits.

What are the best ways to cut Redshift costs without hurting performance?

Rightsize or move to RA3 or Serverless. Purge or tier cold data and tune SQL to reduce scans. Partition and compress data for Spectrum. Use workload management and budget alerts. Keep schemas clean to avoid wasted compute.