← Back to Overview

Cloud Discount Mechanisms 2026: Savings Plans vs RIs vs CUDs vs SUDs

Every cloud provider has a different discount model with different trade-offs. This is the decision guide for how to buy cloud capacity to save the most.

All Discount Mechanisms

MechanismProviderMax DiscountCommitmentFlexibility
Savings Plans (Compute)AWS66%1-3 year hourly spendFlexible across families, sizes, OS, regions
Savings Plans (EC2)AWS72%1-3 year hourly spendLocked to family and region
Standard RIAWS72%1-3 year, specific configLocked to configuration. Limited modification.
Convertible RIAWS66%1-3 yearExchangeable for different instance types
Reserved InstancesAzure72%1-3 year, specific configSize flexibility within same series
Savings PlansAzure65%1-3 year hourly spendFlexible across families and regions
Committed Use DiscountsGCP70%1-3 yearCommit to vCPU/memory quantities
Sustained Use DiscountsGCP30%AutomaticFully flexible. Zero commitment required.
Spot InstancesAWS90%NoneInterruptible. 2-minute notice.
Spot VMsAzure90%NoneInterruptible. Eviction-based.
Preemptible / SpotGCP80%NonePreemptible: 24-hr max. Spot: no limit.

Decision Framework

Is the workload stable for 1+ years?

Use commitment discounts. AWS: Compute Savings Plans for flexibility, EC2 Instance SPs for maximum discount. Azure: Reserved Instances. GCP: Committed Use Discounts (but check if SUDs already cover enough).

Do you need flexibility across instance types?

AWS/Azure Savings Plans. They commit to hourly spend, not specific configurations. GCP CUDs commit to vCPU/memory quantities rather than machine types.

Want maximum possible discount on a known config?

AWS Standard RI (72%) or Azure RI (72%). Locked to specific configuration but deepest discount. Only if you are confident the workload will not change.

Is the workload fault-tolerant?

Use Spot/Preemptible. AWS Spot (up to 90%), Azure Spot (up to 90%), GCP Preemptible (up to 80%). Diversify across instance pools to reduce interruption risk.

Does usage vary month to month?

GCP SUDs are ideal: they apply automatically with no commitment. On AWS/Azure, avoid over-committing. Consider a conservative 50% coverage with Savings Plans and leave the rest on-demand.

Worked Cost Example: $100,000/month Cloud Bill

Annual cost under different discount strategies.

StrategyAWS AnnualAzure AnnualGCP Annual
No discounts (baseline)$1,200,000$1,200,000$1,200,000
GCP SUDs only (automatic)N/AN/A$960,000 (20% off)
70% RI/SP coverage$756,000 (37% off)$768,000 (36% off)$744,000 (38% off)
70% RI/SP + 20% Spot$612,000 (49% off)$624,000 (48% off)$636,000 (47% off)
Best-case blended$588,000 (51%)$576,000 (52%)$600,000 (50%)

Blended rates assume 70% committed, 20% Spot, 10% on-demand. Azure best-case includes Hybrid Benefit on 30% of Windows workloads.

Common Mistakes

Buying too many commitments

Start at 50-60% coverage. Increase only after validating utilisation for 3 months. Stranded capacity is wasted money.

Not reviewing RI utilisation

Check RI/SP utilisation monthly. Under 80% utilisation means you over-committed. Modify or exchange if possible.

Ignoring Spot for eligible workloads

Any stateless, fault-tolerant workload should run on Spot first. The 60-90% discount is too large to ignore.

Committing before rightsizing

Rightsize instances first. Committing on oversized instances locks in waste. Optimise configuration, then commit on the right-sized fleet.

Enterprise Agreements

ProviderProgramThresholdTypical Discount
AWSEnterprise Discount Program (EDP)$1M+/year5-20% off all spend. Stacks with other discounts.
AzureEnterprise Agreement (EA)$1M+/year5-15% off. Best combined with Hybrid Benefit.
GCPNegotiated Contract$500K+/year5-15% off. Lower threshold than AWS/Azure.