Redshift Pricing Negotiation: Reserved Nodes, Serverless, and EDP Leverage
Redshift is the AWS analytics service where idle capacity and over-provisioning compound fastest. RA3 reserved nodes, Serverless RPU sizing, pause/resume policy, and EDP-level private pricing together routinely cut warehouse spend in half.
Redshift cost optimisation has changed substantially with the RA3 node family (compute and storage decoupled), Redshift Serverless (pay-per-RPU elasticity), and the broader integration with S3 via Spectrum and Lake Formation. The structural moves we see deliver the most leverage: pause/resume on non-production clusters, RA3 reserved coverage on steady-state, Serverless adoption on bursty workloads, and EDP-level negotiation on the residual.
The Redshift pricing models
| Configuration | How it bills | Best for |
|---|---|---|
| Provisioned RA3 nodes | Per node-hour + managed storage per GB-month | Steady-state warehouses with predictable concurrency |
| Redshift Serverless | Per RPU-hour (compute) + managed storage | Ad-hoc, bursty, or development workloads |
| Redshift Spectrum | Per TB scanned in S3 | Querying S3 data without ingest |
| Concurrency Scaling | Per second of additional cluster capacity | Burst absorption on provisioned clusters |
RA3 reserved nodes
RA3 node reservations are the highest-leverage commitment for steady-state Redshift clusters. Rates:
| Term | Payment | Approximate discount |
|---|---|---|
| 1 year | No upfront | 20 to 25 percent |
| 1 year | All upfront | 30 to 35 percent |
| 3 year | No upfront | 45 to 50 percent |
| 3 year | All upfront | 60 to 65 percent |
The coverage rule: reserve only clusters that genuinely run 24/7 at stable size. RA3 reservations are node-specific (size and family); resizing a cluster mid-term creates partial-coverage scenarios that look like waste on the bill.
Redshift Serverless economics
Serverless bills RPUs (Redshift Processing Units) per second. Configuration:
- Base capacity: the minimum RPU level (8, 16, 32, etc.) that the workgroup runs at.
- Maximum RPU-hours: a monthly cap to prevent runaway spend.
- Auto-pause: idle workgroups suspend automatically (configurable).
The decision against provisioned:
- Serverless wins on bursty workloads. Ad-hoc analytics, monthly reporting cycles, dev/QA environments.
- Provisioned wins on steady-state workloads. Continuous BI workloads, ELT pipelines that run hourly, anything with predictable concurrency.
- Hybrid: use provisioned RA3 for the steady core, Serverless for ad-hoc and dev/QA.
The most common Serverless mistake is leaving the base capacity at the default (32 RPU minimum). For development workloads, 8 RPU is appropriate and cuts idle cost 75 percent.
Pause and resume on non-production clusters
Provisioned RA3 clusters can be paused on a schedule. Paused clusters bill only for storage (no compute). For non-production warehouses running standard business-hours BI:
- Pause nights and weekends. Recovery time on resume: 5 to 10 minutes.
- Typical non-prod savings: 65 to 75 percent of compute spend.
- EventBridge schedules or third-party automation manage the pause/resume cycle.
This single lever is the biggest immediate Redshift saving for most enterprise estates.
Concurrency Scaling cost control
Concurrency Scaling adds transient cluster capacity when queue depth exceeds a threshold. Pricing:
- 1 free hour per day per cluster.
- Beyond free tier: charged per second at on-demand RA3 node rate.
The trap: a misconfigured workload management policy that triggers Concurrency Scaling on every cycle can rack up six-figure overruns. Configure Concurrency Scaling thresholds explicitly; monitor with the ConcurrencyScalingActiveClusters CloudWatch metric.
Spectrum and Lake Formation cost integration
Spectrum lets Redshift query S3 data directly, billed at $5 per TB scanned. The optimisation surface:
- Parquet conversion. Columnar Parquet versus row-format CSV reduces scan volume by 50 to 90 percent.
- Partition pruning. Properly partitioned S3 datasets allow Spectrum to scan only relevant partitions.
- Result caching. Repeated queries against the same data should hit the result cache, not re-scan.
See our Athena query cost reduction guide for the broader S3-scan cost discipline that applies equally to Spectrum.
Managed storage and snapshot strategy
RA3 managed storage bills $0.024 per GB-month. Automated snapshots above the free tier and cross-region copies add cost. The audit:
- Snapshot retention windows: 7 to 14 days for most workloads, 30 days only where compliance requires.
- Cross-region snapshot copies: justify each, especially at multi-PB scale.
- Manual snapshots: lifecycle to S3 Glacier where long-term retention is required.
The EDP private pricing layer
Redshift is one of the most flexible AWS services on custom pricing. Negotiable items:
- Custom RA3 node-hour rates. For multi-rack RA3 deployments, AWS will discount node-hour rates 15 to 30 percent on EDP commits.
- Custom Serverless RPU rates. For sustained high-volume Serverless workloads.
- Managed storage discounts. Petabyte-scale storage attracts discount below the published rate.
- Spectrum scan-rate concessions. Negotiable at multi-PB monthly scan volumes.
- Reserved node early-exit credits. If you must exit early, AWS will sometimes credit the unused portion against new commitments.
Case study: $1.8M Redshift estate
A retail customer with $1.8M annualised Redshift spend across one production cluster, two staging clusters, and a Spectrum-heavy analytics environment.
Findings:
- Production cluster running RA3.4xlarge x 12 nodes, on-demand, no reservation.
- Two staging clusters running 24/7 with average 8 percent utilisation.
- Spectrum scanning 240 TB/month against unpartitioned CSV.
- 180-day automated snapshot retention.
Interventions:
- 3-year reserved coverage on production RA3 nodes. $420K annualised saving.
- Pause/resume schedule on staging clusters. $310K annualised saving.
- Parquet conversion and partitioning on Spectrum source data. Scan volume dropped 75 percent. $108K annualised saving.
- Snapshot retention reduced to 14 days with Glacier export for compliance copies. $42K annualised saving.
- Custom node-hour discount of 22 percent negotiated at EDP renewal. $108K annualised saving on residual production spend.
Combined annual run-rate dropped from $1.8M to $812K, a 55 percent reduction.
Action checklist
- Inventory every Redshift cluster and Serverless workgroup.
- Identify non-production clusters running 24/7. Implement pause/resume.
- Pull CloudWatch utilisation on production clusters. Right-size oversized nodes.
- Purchase 3-year reserved nodes on stable production clusters.
- Audit Spectrum scan volume and source data format. Convert to Parquet, partition aggressively.
- Audit Concurrency Scaling triggers; cap on misbehaving workloads.
- Reduce snapshot retention; lifecycle long-term snapshots to S3 Glacier.
- Scope custom Redshift pricing into EDP renewal.
- Contact our advisory team for a Redshift cost audit benchmarked against $2.4B+ of reviewed AWS spend.
Redshift cost is one of the most negotiable lines on the AWS bill at scale. See our AWS database cost strategy guide and Redshift Serverless pricing guide for the surrounding context.