Articles

Amazon Redshift Pricing Explained (2027 Guide)

Every Redshift charge in 2027, from RG, RA3, and DC2 node rates and Serverless RPU-hours to managed storage, snapshots, and discounts, with monthly examples for both deployment options.

Team Eon
Written by
Team Eon
Published: 
Nov 17, 2026
0
 min read

Quick Summary

  • Redshift bills compute two ways: provisioned clusters by the node-hour, or Serverless by the RPU-hour, billed per second with a 60-second minimum.
  • Serverless costs $0.375 per RPU-hour in US East, so the 4-RPU minimum works out to $1.50 per active hour.
  • RG nodes are AWS’s recommended provisioned option, with about 30% lower price per vCPU than RA3 and data lake queries with no per-TB scan fee.
  • Managed storage costs $0.024 per GB-month in US East, about $24.58 a month per terabyte, billed separately from compute.
  • Automated snapshots are free for up to 35 days, but manual snapshots bill until they expire or someone deletes them, even after the cluster is gone.

Most Redshift bills creep up through small things: a Serverless workgroup left at its 128-RPU default, a test cluster that ran all weekend, or a year of manual snapshots nobody deleted after a migration.

Amazon Redshift pricing has more moving parts than most AWS services. You pay for compute, storage, and backups separately, and two deployment options bill on completely different meters.

Every rate below is the on-demand list price in US East (N. Virginia) as of November 2026, unless we say otherwise. Other Regions price differently (Serverless runs $0.494 per RPU-hour in Tokyo and $0.5976 in São Paulo), so check yours before you budget.

Amazon Redshift pricing at a glance

This table covers every Redshift billing component, with US East (N. Virginia) list prices from AWS’s pricing page and price list.

Component How it’s billed US East rate
Provisioned nodes (on-demand) Per node-hour, per second after status changes $0.25 (dc2.large) to $13.04 (ra3.16xlarge) per hour
Reserved Instances 1- or 3-year term, per node About 30% to 63% below on-demand
Serverless compute Per RPU-hour, per second, 60-second minimum $0.375 per RPU-hour ($1.50 an hour at 4 RPUs)
Serverless Reservations Reserved RPUs, billed 24/7 Up to 24% off (1-year), up to 50% off (3-year All Upfront)
Redshift Managed Storage (RMS) Per GB-month, metered hourly $0.024 per GB-month
Redshift Spectrum (RA3, DC2) Per TB scanned, 10 MB minimum per query $5.00 per TB
Concurrency Scaling Per second beyond free credits 1 free hour per day per cluster, then the cluster’s on-demand rate
Automated snapshots Free Up to 35 days of retention
Manual snapshots (RG, Serverless) Unique data blocks across active snapshots $0.023 per GB-month
Manual snapshots (RA3) Standard S3 rates $0.023 per GB-month for the first 50 TB
Cross-region snapshot copy and data sharing Per GB transferred $0.02 per GB
Redshift ML Per million training cells $20, then $15, then $7 per million cells
Zero-ETL integrations No integration fee You pay for the Redshift storage and compute the replication uses

How Redshift bills compute: 2 meters

There’s no upfront fee and no minimum spend on the on-demand rates. The first decision that shapes your bill is the deployment option, because provisioned clusters and Serverless measure compute in different units.

Provisioned clusters: Pay per node-hour

With a provisioned cluster, you pick a node type and a node count, and you pay that hourly rate for as long as the cluster runs. AWS bills partial hours in one-second increments after a billable status change like creating, pausing, or resuming the cluster.

On multi-node clusters, you only pay for compute nodes. The leader node isn’t billed.

Serverless: Pay per RPU-hour

Redshift Serverless measures capacity in Redshift Processing Units (RPUs) and charges only while queries run, so idle time costs nothing for compute. Spectrum and Concurrency Scaling are included in the RPU rate.

If you’ve priced other pay-per-use AWS services, this will feel familiar: AWS Lambda pricing runs on the same per-request, per-second logic at a smaller scale.

Provisioned cluster pricing: RG, RA3, and DC2 nodes

AWS sorts provisioned node types into three generations: RG (new generation), RA3 (current generation), and DC2 (previous generation). AWS now recommends RG for new clusters.

Node type vCPU Memory Managed storage per node On-demand per hour Per month (730 hours)
rg.large 2 16 GiB 8 TB (1 TB single-node) $0.3801 $277.47
rg.xlarge 4 32 GiB 32 TB $0.7602 $554.95
rg.4xlarge 16 128 GiB 128 TB $3.04267 $2,221.15
rg.12xlarge 48 384 GiB 128 TB $9.128 $6,663.44
ra3.large 2 16 GiB 8 TB (1 TB single-node) $0.543 $396.39
ra3.xlplus 4 32 GiB 32 TB (4 TB single-node) $1.086 $792.78
ra3.4xlarge 12 96 GiB 128 TB $3.26 $2,379.80
ra3.16xlarge 48 384 GiB 128 TB $13.04 $9,519.20
dc2.large 2 15 GiB 160 GB local SSD $0.25 $182.50
dc2.8xlarge 32 244 GiB 2.56 TB local SSD $4.80 $3,504.00

On RG and RA3 node types, the hourly rate covers compute only; managed storage bills on top.

RG nodes (new generation)

RG nodes run on AWS Graviton processors. AWS launched them in May 2026 at 30% lower price per vCPU than RA3, and the math checks out: rg.4xlarge works out to about $0.19 per vCPU-hour against $0.27 for ra3.4xlarge.

The bigger pricing change is the built-in data lake query engine. RG clusters query Iceberg and Parquet data in Amazon S3 on their own nodes, so Spectrum isn’t required and those queries carry no per-TB scan charge. Existing external tables and Spectrum syntax keep working.

AWS added the rg.large and rg.12xlarge sizes in July 2026. That makes rg.large ($0.3801 an hour) the cheapest RG or RA3 node. AWS’s pricing page still opens with "Provisioned starts at $0.543 per hour," which is the ra3.large rate.

RA3 nodes (current generation)

RA3 nodes pair compute with Redshift Managed Storage, the same as RG, so you size the cluster for the data you process each day and let storage grow on its own. RA3 doesn’t use Graviton, and it relies on Redshift Spectrum for data lake queries, at $5 per TB scanned.

If you’re on RA3 today, you can move to RG through elastic resize, classic resize, or a snapshot restore. Elastic-resize downtime runs 10 to 15 minutes for compatible configurations.

DC2 nodes (previous generation)

DC2 nodes store data on local SSDs, so storage is included in the node price and doesn’t bill separately. The catch is that storage only grows when you add nodes. DC2 still suits datasets under 1 TB compressed, but RA3 is the better pick once you expect the data to grow.

Redshift Serverless pricing and how RPUs work

Redshift Serverless costs $0.375 per RPU-hour in US East, billed per second with a 60-second minimum. There’s no charge for startup time, and data lake queries bill at the same RPU rate as warehouse queries.

AWS’s own example: a job that runs 13 times a day for 10.5 minutes at 128 RPUs uses 8,190 seconds of compute and costs $109.20 for the day.

What is a Redshift RPU?

An RPU is Redshift Serverless’s unit of compute, and each RPU provides 16 GB of memory. You can set base capacity from 4 to 1,024 RPUs (512 is the ceiling in Regions without the expanded limit).

A few sizing rules affect the bill:

  • The default base is high: Base capacity defaults to 128 RPUs, which is $48 for every hour the workgroup is busy.
  • The 4-RPU floor has limits: It only supports up to 32 TB of managed storage. Above that, the minimum base is 8 RPUs.
  • Scaling past 4 sticks: Once a 4-RPU workgroup scales up, it won’t scale back down to 4.
  • Scale-ups bill at the same rate: Serverless can scale above base capacity under load, and every extra RPU costs the same $0.375.

Base, Max RPU-hours, and MaxRPU settings

Three settings control Serverless spend:

  • Base: The capacity Redshift uses to start serving queries, in RPUs. You can also set a price-performance target and let AWS adjust capacity for you.
  • Max RPU-hours: A usage limit per day, week, or month. When a workgroup hits it, Redshift can log the event, send an SNS alert, or turn off user queries.
  • MaxRPU: The highest RPU count a workgroup can scale to. It caps the hourly rate, while Max RPU-hours caps the total.

Capacity can stay high for a while after query load drops, so a Max RPU-hours limit is worth setting. We’d set both on every workgroup before it goes to production.

Redshift Managed Storage pricing

Redshift Managed Storage (RMS) holds the data for RG clusters, RA3 clusters, and Serverless workgroups. It costs $0.024 per GB-month in US East, metered hourly on the total data stored. You pay the same rate whether the data sits on a node’s local SSD or has been offloaded to S3.

At that rate, 1 TB costs about $24.58 a month and 40 TB costs $983.04. AWS’s own sizing example of four rg.4xlarge nodes with 40 TB of RMS comes to $9,867.64 a month, and storage is about 10% of that bill.

RMS doesn’t include manual snapshots on RA3; those bill separately as backup storage.

Spectrum, Concurrency Scaling, ML, and zero-ETL charges

These four charges sit outside the main compute and storage lines. Whether you pay them depends on your node type and the features you turn on.

Redshift Spectrum

Spectrum lets RA3 and DC2 clusters query data in S3 for $5 per TB scanned, rounded up to the next MB, with a 10 MB minimum per query. Failed queries and DDL statements are free. Serverless and RG clusters don’t pay it.

File format has a big effect on this bill. In AWS’s example, one query against a 4 TB text file costs $20. Compress it with GZIP and it’s $5. Convert it to Parquet and read one column of 100, and it’s $0.05.

Concurrency Scaling

Provisioned clusters earn one free hour of Concurrency Scaling credit for every 24 hours they run, and they can bank up to 30 hours, enough to cover 97% of customers. 

Beyond the credits, you pay the cluster’s on-demand rate per second, with a one-minute minimum each time a scaling cluster starts. A usage limit on Concurrency Scaling keeps a week of afternoon dashboard spikes from doubling that line.

For a 10-node rg.4xlarge cluster ($30.43 an hour), two scaling clusters running five minutes past the free credits add about $5.07.

Redshift ML

Redshift ML trains models through Amazon SageMaker. First-time SageMaker users get two free CREATE MODEL requests a month for two months, up to 100,000 cells each. 

After that, training costs $20 per million cells for the first 10 million, $15 per million for the next 90 million, and $7 per million above that.

The default MAX_CELLS setting of 1 million keeps most training runs under $20.

Zero-ETL integrations

Zero-ETL integrations replicate data from Aurora, RDS, DynamoDB, Oracle Database@AWS, and apps like Salesforce into Redshift. 

AWS charges no integration fee, but you pay for the storage the replicated data uses, the compute or RPUs that process changes, and cross-AZ data transfer. On Serverless, that replication keeps using RPUs even in hours when nobody is querying, so check overnight usage before you assume an idle workgroup costs nothing.

Some sources add their own fees. Oracle Database@AWS charges $0.04 per GB exported and $2.00 per GB of change data capture in US East.

Provisioned vs. Serverless: Which costs less?

Provisioned costs less when a warehouse stays busy most of the day, and Serverless costs less when it sits idle for long stretches. The two examples below use the same 10 TB warehouse. The usage numbers are illustrative assumptions, so swap in your own.

Provisioned example: Two rg.4xlarge nodes running all month

Line item Calculation Monthly cost
Compute 2 nodes × $3.04267 × 730 hours $4,442.30
Managed storage 10 TB × 1,024 GB × $0.024 $245.76
Total (on-demand) $4,688.06
Total with 1-year No Upfront Reserved Instances 2 × $2.12987 × 730 + $245.76 $3,355.37

The cluster bills every hour it runs, busy or not. Reserved Instances cut compute by about 30% here. You pay for every hour of the term, though, so a 1-year No Upfront reservation only comes out ahead if the cluster would otherwise run about 70% of the year.

Serverless example: A 32-RPU workgroup busy eight hours a weekday

Line item Calculation Monthly cost
Compute 32 RPUs × 8 hours × 22 weekdays × $0.375 $2,112.00
Managed storage 10 TB × 1,024 GB × $0.024 $245.76
Total $2,357.76

This assumes the workgroup stays at its 32-RPU base. Any scale-up during busy hours adds to the compute line at the same $0.375 rate.

Where the break-even point sits

At 32 RPUs, Serverless costs $12 for every active hour. It passes the cluster’s $4,442.30 monthly compute bill at about 370 active hours a month, roughly 12 hours a day, every day.

AWS doesn’t publish a conversion between RPUs and nodes, so treat 32 RPUs vs. two rg.4xlarge nodes as an assumption. Before you commit, run a week of your real queries on both and compare usage in Cost Explorer.

As a rule of thumb, we’d lean toward provisioned for warehouses that load data and serve dashboards around the clock. Serverless fits ad hoc analysis, dev and test, and spiky workloads that go quiet overnight.

How to lower Redshift costs with Reserved Instances, reservations, and pause/resume

Redshift has three discount levers. Which one fits depends on your deployment option and how steady your usage is.

Reserved Instances for provisioned clusters

Reserved Instances (AWS also calls them reserved nodes) trade a 1- or 3-year commitment for a lower rate on a specific node type. You can pay No Upfront, Partial Upfront, or All Upfront.

Based on AWS’s US East price list, the discount lands at about 30% to 34% on a 1-year term and about 57% to 63% on a 3-year term, depending on the payment option. A 3-year All Upfront rg.4xlarge works out to about $1.14 an hour, against $3.04 on demand.

The trade-off is that you pay for every hour of the term whether the cluster runs or not, even while it’s paused. Run on demand for a few months first, so you know the node type and count you’ll actually keep.

Serverless Reservations

Serverless Reservations commit to a set number of RPUs at a discount. A 1-year reservation saves up to 20% with No Upfront or up to 24% All Upfront. AWS’s pricing page lists up to 45% for a 3-year term, and the 3-year All Upfront option added in July 2026 goes up to 50%.

Reservations are managed at the payer account level and shared across linked accounts. Usage above the reserved RPUs bills at on-demand rates.

Reserved RPUs bill around the clock, so only reserve the RPUs you use 24/7. In our Serverless example above, reserving 32 RPUs for a year at $0.30 per RPU-hour would cost about $7,008 a month. On-demand costs $2,112 for the same eight hours a weekday.

Pause and resume for provisioned clusters

You can pause an on-demand cluster manually or on a schedule, and compute billing stops while it’s paused. Storage keeps billing, and Redshift takes a snapshot each time you pause.

Pausing pays off for dev and test clusters that only run during working hours. Pausing the two-node example cluster outside a 12-hour weekday would cut its compute from $4,442.30 to about $1,606.53 a month.

AWS cost optimization tools track Redshift spend alongside the rest of your AWS bill, and the account-wide habits behind these savings are well-worn AWS cost reduction strategies.

Hidden Redshift costs: Snapshots, data transfer, and defaults

The charges below don’t show up on the headline rate, and they’re usually where a Redshift bill outgrows its estimate.

Manual snapshots that outlive their cluster

Automated snapshots are free and can be kept for up to 35 days. The default retention is one day, and on RG and RA3 clusters you can’t turn automated snapshots off.

Manual snapshots are a different story. By default, Redshift keeps them indefinitely, even after you delete the cluster, and they bill until they expire or someone deletes them. That includes paused clusters.

How they bill depends on the node type:

  • RG and Serverless: Since June 2026, you pay $0.023 per GB-month only for the unique data blocks across all active manual snapshots in the account. Blocks shared between snapshots count once.
  • RA3: Manual snapshots bill as backup storage at standard S3 rates. In AWS’s example, a 10 TB RA3 cluster with 30 TB of manual snapshots pays for all 30 TB of backup storage. At $0.023 per GB-month, that’s about $706.56 a month in US East.
  • DC2: Storage comes with the nodes, but backups beyond the provisioned storage size bill at S3 rates.

The fix is a retention period on every manual snapshot and a quarterly sweep for snapshots of clusters that no longer exist. Sound backup retention policies set how long to keep each copy, and a snapshot on its own isn’t a full backup strategy.

AWS Backup can also manage Redshift snapshots for provisioned clusters and Serverless, alongside your other AWS resources, though AWS Backup pricing adds its own line to the bill.

Cross-region snapshot copies and data sharing

Copying snapshots to another Region for disaster recovery costs $0.02 per GB from US East. AWS charges this on automated snapshots, manual snapshots, and Serverless recovery points alike. The first copy moves the whole snapshot (about $204.80 for 10 TB), and later copies move only the changed blocks.

For a 10 TB warehouse that changes 0.5 TB a day, those daily copies add about $307.20 a month in transfer fees. Cross-region data sharing bills per GB in the consumer Region ($0.02 when that’s US East).

Other traffic in and out of Redshift (such as JDBC/ODBC traffic to a cluster endpoint in a VPC) bills at standard AWS data transfer rates. Loads, unloads, backups, and restores between Redshift and S3 in the same Region are free.

Serverless settings that keep RPUs running

A few Serverless behaviors bill more than teams expect:

  • Open transactions that never end, which keep using RPUs until the session times out.
  • Canceled queries and system table queries, which still bill for the time they ran.

When Redshift stores data you already back up

Some Redshift spend exists only to hold copies of production databases, loaded through ETL pipelines so analysts and auditors can query history. Those copies bill for storage, the pipelines bill for compute, and the same data usually sits in a backup somewhere already.

With Eon, that backup can serve the query. Eon stores database backups as deduplicated tables in Amazon S3, and you choose which ones to share with Redshift. Redshift then queries them in place, read-only, with no restore and no ETL pipeline to maintain.

Eon’s Cloud Backup Posture Management (CBPM), the category we created, discovers databases and other cloud resources, classifies them by data type, and applies backup policy automatically.

Eon also cuts backup storage costs by 30% to 50% through global deduplication; NETGEAR cut its backup storage 35% after moving off a legacy provider. Most cloud teams can’t say what they spend on backup, and Eon’s Cost Explorer breaks that spend down by resource, resource type, or source account.

Pros and cons of Redshift’s pricing model

Redshift’s pricing rewards teams that match the deployment option to their workload, and it punishes defaults left untouched.

Pros

  • Steady warehouses get predictable node-hours, and spiky ones pay nothing for idle time on Serverless.
  • 3-year Reserved Instances run about 57% to 63% below on-demand, and 3-year Serverless Reservations go up to 50%.
  • RG clusters query S3 data with no per-TB charge, which removes the Spectrum line entirely.
  • Automated snapshots are free for up to 35 days.

Cons

  • A 128-RPU Serverless base and manual snapshots with no expiry both bill more than most teams need.
  • Compute, storage, snapshots, data transfer, Spectrum, and Concurrency Scaling each bill separately.
  • Reserved Instances and Serverless Reservations charge for every hour of the term, used or not.

The bottom line on Amazon Redshift costs

Redshift’s rates are fair at list price and expensive on autopilot. The biggest savings in our examples came from matching the deployment option to how often the warehouse is busy, then committing only to capacity that runs around the clock.

The next place to look is what the warehouse holds. Any table that exists only so someone can query production history is one your backups could already serve.

Want Redshift to query your database backups where they sit? Book a demo to see Eon backups queried from Redshift with no restore.

Frequently asked questions

Is AWS Redshift free?

No. Redshift bills for compute and storage, but first-time Serverless users get a $300 credit that expires 90 days after sign-up. That usage shows in the Redshift console or SYS_SERVERLESS_USAGE view (not the billing console) until the trial ends. Regions without Serverless get a two-month DC2 large trial instead.

How much does 1 TB of storage cost on AWS?

In Redshift, 1 TB of managed storage costs about $24.58 a month in US East ($0.024 per GB-month × 1,024 GB). A terabyte of RA3 manual snapshots bills at S3 Standard rates, about $23.55 a month.

How much does Amazon Redshift cost per hour?

Redshift Serverless starts at $1.50 an hour at the 4-RPU minimum. Provisioned nodes start at $0.25 an hour for dc2.large and $0.3801 for rg.large, while AWS’s pricing page advertises $0.543 an hour, the ra3.large rate.

What is an RPU in Redshift?

An RPU (Redshift Processing Unit) is the compute unit for Redshift Serverless, and each one provides 16 GB of memory. You can set base capacity from 4 to 1,024 RPUs, and each RPU-hour costs $0.375 in US East.

Is Redshift Serverless cheaper than provisioned?

Yes, for workloads that sit idle for long stretches, because Serverless bills only while queries run. For warehouses busy most of the day, a provisioned cluster, especially with Reserved Instances, usually costs less.

Do Redshift snapshots cost money?

Yes, manual snapshots do. Automated snapshots are free for up to 35 days, but manual snapshots bill at $0.023 per GB-month in US East until they expire or are deleted, even after the cluster is deleted.

FAQ

No items found.
Team Eon
Team Eon

Turn your backups into usable data

Eon turns your backups into instantly searchable, usable data so you can recover exactly what you need without delays.

  • Instantly search backup data with Global Search
  • Find any backup data with Global Search
  • Restore only what you need
See eon in action