Cloud Cost Optimization — Commitment Instruments

Reserved Instances vs
Savings Plans: 2026 Guide

Reserved Instances offer the deepest discounts. Savings Plans offer greater flexibility. GCP Committed Use Discounts offer a third model entirely. This complete comparison explains which instrument to use — and when the flexibility of Savings Plans is worth the discount trade-off.

72%
Max RI Discount (AWS 3yr)
66%
Max SP Discount (AWS 3yr)
55%
GCP 3-Year CUD Discount
65%
Azure 3-Year RI Discount

This guide is part of the Cloud Cost Optimization: Enterprise FinOps Guide pillar. Choosing between Reserved Instances and Savings Plans — or their equivalents across AWS, Azure, and GCP — is one of the highest-impact decisions in cloud cost management. The wrong choice can leave millions of dollars of savings on the table or lock you into commitments for workloads that change. This guide cuts through the complexity with a clear framework for each scenario. For multi-cloud commitment coordination strategies, see the Multi-Cloud Cost Optimization guide.

AWS: Reserved Instances vs Compute Savings Plans

AWS offers the most complex commitment instrument landscape of any cloud provider, with two primary vehicles: EC2 Reserved Instances (RIs) and Savings Plans (SPs). EC2 RIs have been available since 2009; Compute Savings Plans were introduced in 2019 as a more flexible alternative. Understanding the differences — and the scenarios where each performs better — is essential to building an optimal AWS commitment strategy.

AWS Reserved Instances

AWS EC2 Reserved Instances commit to a specific instance type, operating system, tenancy, and region (or availability zone for Zonal RIs). In exchange for this specificity, they offer the highest available discounts: up to 72% savings for a 3-year all-upfront Standard RI compared to on-demand pricing. Standard RIs can be sold on the AWS Marketplace if no longer needed, providing an exit option that Savings Plans do not. Convertible RIs offer lower discounts (up to 66%) but allow modification of instance type, OS, and tenancy during the term — providing flexibility at a cost.

AWS RI Type 1-Year Discount 3-Year Discount Flexibility Marketable
Standard RI (All Upfront)~40%~72%Instance-specificYes
Standard RI (Partial Upfront)~38%~69%Instance-specificYes
Standard RI (No Upfront)~35%~62%Instance-specificYes
Convertible RI (All Upfront)~31%~66%ModifiableNo
Compute Savings Plan~37%~66%Any EC2 + Lambda + FargateNo
EC2 Instance Savings Plan~40%~72%Family-flexibleNo

AWS Compute Savings Plans

Compute Savings Plans commit to a spend rate (e.g., $10/hour) rather than a specific resource configuration. This flexibility spans EC2 instances (any type, any region, any OS), Lambda, and Fargate — making it the most portable commitment instrument on AWS. The trade-off is that the maximum discount for a 3-year Compute SP (approximately 66%) is 6 percentage points lower than a 3-year Standard RI all-upfront (72%). EC2 Instance Savings Plans offer a middle ground: family-level flexibility (any size within an instance family in a specific region) with discounts approaching Standard RI levels.

AWS Strategy Principle

The optimal AWS commitment strategy is typically a layered approach: EC2 Instance Savings Plans (or Standard RIs) for your stable, predictable compute baseline; Compute Savings Plans for your growing or variable EC2 workloads; and on-demand or Spot for burst, development, and unpredictable capacity. Reserve Standard RIs for workloads where you are highly confident of the specific instance type for the full commitment period — and consider the RI Marketplace as a hedge against over-commitment.

Azure: Reserved Instances vs Savings Plans

Azure followed AWS's trajectory: offering Reserved Instances (called Azure Reserved VM Instances) since 2017, then adding Azure Savings Plans for Compute in 2022. Azure's Savings Plans are functionally similar to AWS Compute Savings Plans — a dollar-per-hour commitment that applies flexibly across eligible compute resources.

Expert Advisory

Want independent help negotiating better terms? We rank the top advisory firms across 14 vendor categories — free matching, no commitment.

Azure Reserved VM Instances

Azure Reserved VM Instances commit to a specific VM size and region for 1 or 3 years. Instance size flexibility (the ability to apply a reservation to different sizes within the same family) is enabled by default for most VM families — a more generous flexibility provision than AWS Standard RIs. Discounts reach approximately 65% for 3-year reservations on common VM families. Azure RIs apply to VMs, SQL Databases, Cosmos DB, Redis Cache, App Service, and several other services — extending beyond compute in a way that AWS Standard RIs do not.

Azure Savings Plans for Compute

Azure Savings Plans for Compute commit to a dollar-per-hour spend and apply to VMs, Azure Dedicated Hosts, Container Instances, App Service Premium, and Azure Functions Premium plan. Maximum discounts are slightly lower than Azure RIs (approximately 65% vs 65% for 3-year, with the specific discount varying by VM family and region). The key advantage over RIs is geographic flexibility — Savings Plan benefits apply across all Azure regions, whereas RIs are region-specific (though transferable). For enterprises with workloads distributed across multiple Azure regions, Savings Plans reduce the complexity of managing per-region RI portfolios.

Azure Commitment Application Order

Azure applies commitment discounts in a specific order: Azure Hybrid Benefit first, then Reserved VM Instances, then Savings Plans, then Dev/Test pricing. If you hold both RIs and Savings Plans, the RI is applied first — meaning your Savings Plan benefits are consumed only for resources not already covered by an RI. Build your commitment portfolio with this application order in mind, and use Azure Cost Management's reservation utilisation reports to monitor effectiveness.

GCP: Committed Use Discounts

GCP does not offer a "Savings Plan" equivalent — its commitment instrument is the Committed Use Discount (CUD), which commits to a specific number of vCPUs and GB of memory for 1 or 3 years in a specific region. CUDs are more restrictive than AWS or Azure Savings Plans: they cannot flex across machine types or regions. However, GCP's automatic Sustained Use Discounts (SUDs) — applied without any commitment for resources running more than 25% of a calendar month — provide a baseline of 20–30% savings that AWS and Azure on-demand customers do not receive.

For a detailed breakdown of GCP CUD mechanics, SUD interaction, and BigQuery-specific pricing, see the GCP Cost Optimization Guide. The key cross-provider insight is that GCP's effective on-demand price (after automatic SUDs) is often competitive with or lower than AWS and Azure on-demand rates — changing the incremental value calculation for CUDs vs RIs/SPs on competing clouds.

Complete Discount Rate Table

Instrument Cloud 1-Year Best 3-Year Best Flexibility Level Payment Options
Standard RIAWS~40%~72%Instance-lockedAll/Partial/No upfront
Convertible RIAWS~31%~66%ModifiableAll/Partial/No upfront
Compute Savings PlanAWS~37%~66%Any EC2+Lambda+FargateHourly commitment
EC2 Instance SPAWS~40%~72%Family-flexibleHourly commitment
Reserved VM InstanceAzure~36%~65%Size-flexible (family)Upfront or monthly
Savings PlanAzure~35%~65%Any eligible computeHourly commitment
Committed Use DiscountGCP~37%~55%Machine-type lockedMonthly billing
Sustained Use DiscountGCPAuto: ~30%Auto: ~30%Automatic, no commitmentAutomatic

Flexibility Analysis

Flexibility has a cost — it is measured in foregone discount points. The flexibility premium (the discount you give up for more flexible commitment instruments) varies by cloud and instrument, but typically runs 4–8 percentage points. Understanding when flexibility is worth paying for is the core judgment call in commitment strategy.

Free Resource

Get the IT Negotiation Playbook — free

Used by 4,200+ IT directors and procurement leads. Oracle, Microsoft, SAP, Cloud — all covered.

Flexibility matters most in four scenarios: when infrastructure footprint is actively changing (migrations, application modernisation, cloud consolidation); when specific instance types may be deprecated or superseded by newer generations; when workloads span multiple regions and regional RI purchases would fragment coverage; and when the organisation lacks the FinOps maturity to manage granular per-instance RI portfolios efficiently. Flexibility matters least — and should be foregone for maximum discount — when infrastructure is stable, long-running, well-understood, and likely to persist in its current form for 3+ years.

Not sure which commitment instruments fit your workload profile?

Independent cloud cost advisors can model your exact environment and recommend an optimal strategy.

Get Expert Help →

Commitment Risks and Mitigation

Commitment risk — the risk of paying for reserved capacity that you don't use — is the primary downside of all commitment instruments. Under-utilisation occurs when committed resources are decommissioned, migrated, or scaled down before the commitment expires. Over-commitment at purchase is the most common cause, driven by optimistic infrastructure projections, technical migrations that never complete on schedule, and organisational changes (M&A, divested business units) that change workload profiles.

Risk Mitigation Strategies

Use conservative (P10) usage projections when sizing commitments — commit to the volume you are confident of running, not the volume you plan to run. For AWS, prefer Compute Savings Plans and EC2 Instance Savings Plans over Standard RIs when workload configurations might change; the flexibility premium is cheap insurance. For Azure RIs, enable instance size flexibility to allow the reservation to cover different sizes within the same family. For GCP CUDs, use 1-year terms for growing workloads and 3-year terms only for fully stable production infrastructure.

The AWS RI Marketplace provides an important mitigation for Standard RI over-commitment — unused Standard RIs (with at least one month remaining) can be listed and sold to other AWS customers. Selling RIs at a discount recovers capital and eliminates ongoing commitment charges. This option is not available for Savings Plans, Convertible RIs, or any Azure/GCP commitment instruments.

Selection Framework: Which Instrument to Choose

Scenario A
Stable, long-running production workloads with known instance types
Use AWS Standard RIs (3-year all-upfront) or EC2 Instance Savings Plans for maximum discount. Use Azure Reserved VM Instances (3-year). Use GCP 3-year CUDs. The stability justifies locking in specific resource configurations for the deepest available discounts. Model conservatively to avoid over-commitment, then layer additional commitment as confidence grows.
Scenario B
Growing workloads or active cloud migration
Use AWS Compute Savings Plans (1-year) — the flexibility across instance types, regions, and Lambda/Fargate protects against architecture changes. Use Azure Savings Plans for Compute (1-year). Use GCP 1-year CUDs for established workload components, leave growth on SUD + on-demand. Review and expand commitments annually as the infrastructure stabilises.
Scenario C
Multi-region footprint or uncertain instance type trajectory
Azure Savings Plans are particularly valuable here — their cross-region flexibility eliminates the need to manage per-region RI portfolios. AWS Compute Savings Plans are similarly region-agnostic. For GCP, SUD coverage handles variable workloads automatically; consider regional CUDs only for anchor production workloads you're confident will remain in specific regions.
Scenario D
Low FinOps maturity / small cloud team
Favour flexible instruments — AWS Compute Savings Plans, Azure Savings Plans — that require less ongoing management than granular RI portfolios. The 4–6% discount penalty is worth the operational simplicity. As FinOps maturity develops, layer in RIs for the stable core to recapture the discount delta. Consider a cloud cost optimisation advisor to review your commitment portfolio annually.

Stacking Strategies: Combining Instruments

The highest-performing commitment strategies stack multiple instruments to cover different workload tiers. A typical AWS strategy: Standard RIs for stable baseline instances (e.g., the 200 m6i.xlarge instances that have run 24/7 for 18 months and will continue indefinitely), EC2 Instance Savings Plans for the moderate-growth tier (instances that may change size but will stay within the family), and Compute Savings Plans for new services, Lambda, and Fargate. The Azure equivalent: RIs for known stable VMs plus Savings Plans for growth capacity, with Azure Hybrid Benefit maximised across all Windows and SQL commitments.

See our dedicated guide on Azure Reserved Instances for Azure RI stacking with Azure Hybrid Benefit and MACC, and the AWS Cost Optimization guide for detailed AWS RI and Savings Plan stacking strategies.

Frequently Asked Questions

Can I convert Standard RIs to Savings Plans?
No — Reserved Instances and Savings Plans are separate instruments and cannot be directly converted. If you have Standard RIs you no longer want, you can list them on the AWS RI Marketplace (if they meet eligibility criteria — typically at least one month remaining and purchased through the regular RI purchase flow). Once sold, you can purchase Savings Plans for the equivalent coverage. The conversion process involves marketplace timing and potential pricing gaps, so plan the transition carefully rather than doing it opportunistically.
Which commitment instrument is best for Kubernetes workloads?
For Kubernetes on AWS EKS, AWS Compute Savings Plans are generally preferred over Standard RIs — the flexibility to cover Fargate and changing EC2 instance types (as you rightsize nodes or adopt new generations) is valuable in dynamic K8s environments. For Azure AKS, Azure Savings Plans for Compute work similarly. For GCP GKE, Sustained Use Discounts provide automatic baseline coverage; CUDs can be applied to the underlying Compute Engine node pools for stable cluster configurations. See the Kubernetes Cost Optimization guide for detailed cluster-level cost strategies.
What happens to unused Reserved Instances or Savings Plans?
Unused commitment benefits are wasted — you pay the committed rate whether or not you consume the corresponding resources. For AWS Standard RIs, unused capacity can be partially recovered via the RI Marketplace. For Savings Plans (AWS and Azure) and Azure RIs, there is no marketplace option — unused commitments represent sunk costs. This is why conservative sizing is critical: commit to your P10 floor usage, not your P50 or P90. Monitor utilisation weekly and adjust new commitment purchases based on actual utilisation rates of existing instruments.
How do Savings Plans interact with Spot Instances?
On AWS, Savings Plans do not apply to Spot Instances — Spot pricing is already deeply discounted (60–90% below on-demand) and operates through a separate capacity market mechanism. Similarly, Azure Savings Plans and RIs do not apply to Azure Spot VMs. GCP CUDs do not apply to Spot VMs (Spot pricing is separate from CUD/SUD coverage). The practical implication: your commitment instruments cover your on-demand and reserved compute baseline; Spot Instances provide an additional, independent cost reduction layer for interruptible workloads on top of your commitment strategy.

Optimise Your Cloud Commitment Strategy

Connect with an independent cloud cost advisor who can model your workload and design an optimal RI/Savings Plan mix across AWS, Azure, and GCP.