Rates last reviewed: July 2026.

How Redshift Pricing Works

Redshift has two completely different pricing stories: provisioned clusters (pay-per-node-hour) and Redshift Serverless (pay-per-RPU-hour), plus managed storage and Spectrum query costs. What you pay depends on the node family you pick, whether you commit to a reservation, and whether you need Serverless flexibility or provisioned predictability.

The two Redshifts: provisioned vs Serverless

Redshift started as a provisioned data warehouse. You picked some nodes, paid by the hour, managed the cluster. Then Amazon launched Redshift Serverless, which works more like Snowflake — compute suspends when idle, you pay for what you use. The two models share the same SQL engine and storage layer but have completely different pricing and operational profiles.

Provisioned cluster pricing

With provisioned clusters, you are renting virtual machines that run Redshift. The pricing is straightforward: pick a node type, pick a count, pay per node-hour. The minimum charge is one hour — spin up a cluster for 10 minutes and you pay for 60.

The dc2 family uses local SSD storage. When you resize or lose a node, the data is gone unless you have it backed up somewhere. The RA3 family separates compute from storage — data lives in S3 and caches locally on SSDs — so you can resize without moving data. RA3 also charges for "managed storage" at $0.024/GB-month for the data stored in S3.

Reserved instances

You can buy provisioned Redshift nodes on 1-year or 3-year reservations. A 1-year partial upfront typically saves 30-35% off on-demand. Three-year partial upfront can save 50-60%. This is the strongest Redshift cost lever for steady workloads, and it is the main reason provisioned Redshift can beat Snowflake on price at scale.

Redshift Serverless pricing

Redshift Serverless charges per RPU-hour. One RPU is roughly equivalent to 1 GB of RAM plus proportional CPU. The base configuration is 8 RPUs, and Serverless auto-scales up to handle concurrency (you set a max RPU limit to control costs).

• Base RPU compute: 8 RPUs minimum, ~$0.25/RPU-hour in us-east-1 = ~$2.00/hour at base
• Auto-suspend: Configurable from 0 to 24 hours (zero compute cost when suspended)
• Per-second billing: After the first minute, you pay per second, same as Snowflake
• Managed storage: Same $0.024/GB-month as RA3 provisioned

Serverless is the direct competitor to Snowflake's consumption-based model. At 8 RPUs running 8 hours/day, 22 days/month: 8 × 176 hours × $0.25 = ~$350/month in compute. A Snowflake Small warehouse (2 credits/hr) running the same schedule would be 2 × 176 × $2.50 = ~$880/month. Serverless is cheaper at this scale — but the Snowflake benchmark assumes 2 credits whereas Serverless assumes 8 RPU, and the actual performance per unit differs, so you need your own benchmarks to compare apples-to-apples.

Managed storage (RA3 and Serverless)

Both RA3 provisioned nodes and Redshift Serverless use managed storage. You pay $0.024/GB-month for the compressed data stored in S3. This includes automatic backups and replication within the region. A 10 TB warehouse adds about $245/month in storage cost. Snowflake charges $23/TB-month, so Redshift managed storage is roughly 5% more expensive at list price.

Redshift Spectrum pricing

Redshift Spectrum lets you query data directly in S3 without loading it into Redshift tables. This costs $5/TB scanned on top of your cluster or Serverless compute. It is useful for data lake queries, but the per-TB cost adds up fast if you scan large volumes frequently.

Concurrency scaling

Provisioned clusters add concurrency scaling clusters when the workload exceeds the base cluster's capacity. These additional clusters are billed per-hour at the node rate and show up as a separate line on your bill. The feature is useful for handling peak load, but the costs can be unpredictable if you do not set a limit.

Hidden costs to watch

• Data transfer out of AWS. If you connect BI tools or users that are outside AWS, you pay standard EC2 data transfer rates ($0.05-0.09/GB depending on region).
• VACUUM and ANALYZE operations. On provisioned clusters, these consume node-hours and can slow down query performance while they run.
• Snapshot storage. Automated snapshots stored beyond the retention period incur S3 storage charges.
• Cross-region data copy. Disaster recovery or cross-region replication adds data transfer costs.

Getting the best Redshift price

• Use RA3 nodes so you can resize compute without migrating data.
• Buy reserved instances for baseline capacity (saves 30-60%).
• Use Serverless for variable or bursty workloads.
• Set a concurrency scaling budget limit to avoid surprises.
• Minimize Spectrum scans by loading frequently-accessed data into managed storage.

Compare Redshift against other platforms

Use the calculator to compare Redshift Serverless cost against Snowflake, Databricks, and BigQuery with your actual workload numbers.

Compare Redshift cost with your workload