AWS offers two primary service-level commitment mechanisms for reducing compute costs: Savings Plans and Reserved Instances (RIs). Both require committing to use a defined level of compute capacity for 1 or 3 years in exchange for discounts of up to 72% compared to On-Demand pricing. But they work differently, suit different environments, and require different management strategies.
This guide is part of our cloud cost optimisation series and sits alongside our AWS EDP negotiation playbook — the EDP provides a top-line discount on total spend, while Savings Plans and RIs provide service-level discounts that stack on top. Understanding both is essential for an optimised AWS commercial strategy.
For a broader comparison including Azure MACC and GCP Commit, see our complete reserved instances vs savings plans guide.
For most enterprises: Compute Savings Plans for flexible workloads, targeted EC2 RIs for stable workloads where instance type is fixed, and stacking both on top of an EDP for maximum effective discount. The optimal split is typically 60–70% Compute Savings Plans, 20–30% EC2 RIs, with 10–20% On-Demand buffer.
Understanding the Two Mechanisms
AWS Savings Plans
Savings Plans commit you to a minimum level of compute spend (measured in $/hour) rather than specific instance types. Introduced in 2019 as AWS's response to the inflexibility of RIs, Savings Plans automatically apply to eligible compute usage regardless of instance family, region, or OS — within the scope of the plan type.
There are three Savings Plan types:
| Plan Type | Applies To | Flexibility | Max Discount vs OD |
|---|---|---|---|
| Compute Savings Plans | EC2, Lambda, Fargate | Any instance family, region, OS, tenancy | ~54% (3yr, no upfront) |
| EC2 Instance Savings Plans | EC2 in specific family and region | Flexible across size, OS, tenancy | ~72% (3yr, no upfront) |
| SageMaker Savings Plans | SageMaker instance usage | Any instance family, region | ~64% (3yr) |
Reserved Instances
RIs are commitments to use a specific instance type in a specific region, for 1 or 3 years. In exchange for this lock-in, AWS offers its highest EC2 discounts. RIs can be Standard (locked to instance family and region) or Convertible (can be exchanged for different instance types).
| RI Type | Convertible? | Max Discount vs OD | Can Sell on Marketplace? |
|---|---|---|---|
| Standard RI (1yr) | No | ~40% (no upfront) | Yes |
| Standard RI (3yr) | No | ~62% (partial upfront) | Yes |
| Convertible RI (1yr) | Yes | ~31% (no upfront) | No |
| Convertible RI (3yr) | Yes | ~54% (partial upfront) | No |
Head-to-Head: Discount Rate Comparison
The following comparison uses m5.xlarge in us-east-1 as a reference instance type. On-Demand price as of Q1 2026 is approximately $0.192/hour.
| Option | Term | Payment | Effective Hourly | Discount vs OD |
|---|---|---|---|---|
| On-Demand | N/A | N/A | $0.192 | Baseline |
| Compute SP | 1yr | No Upfront | ~$0.131 | ~32% |
| Compute SP | 3yr | No Upfront | ~$0.088 | ~54% |
| EC2 Instance SP | 1yr | No Upfront | ~$0.116 | ~40% |
| EC2 Instance SP | 3yr | No Upfront | ~$0.079 | ~59% |
| Standard RI | 1yr | No Upfront | ~$0.116 | ~40% |
| Standard RI | 3yr | Partial Upfront | ~$0.073 | ~62% |
| Standard RI | 3yr | All Upfront | ~$0.070 | ~64% |
Note: Exact rates vary by instance type, region, OS, and AWS pricing updates. Always verify against current AWS pricing pages before committing.
The headline finding: Standard RIs with all-upfront payment offer the highest absolute discounts, but Compute Savings Plans offer nearly comparable savings with dramatically more flexibility. The difference between a 3-year Compute SP (54%) and a 3-year All-Upfront Standard RI (64%) is ~10 percentage points — in exchange for complete flexibility vs. full lock-in.
Flexibility: The Real Differentiator
For enterprises with evolving infrastructure — migration projects, modernisation initiatives, or workload variability — flexibility is often more valuable than the incremental discount of a locked RI.
Compute Savings Plans: Maximum Flexibility
A Compute SP applies to any EC2 instance regardless of family (m5, c5, r5, graviton), size, region, OS, or tenancy. It also covers Lambda and Fargate usage. If you migrate from x86 to Graviton, move a workload from us-east-1 to eu-west-1, or switch from EC2 to Fargate containers, your Compute SP continues to apply. This makes Compute SPs the default choice for organisations managing complex, changing infrastructure.
EC2 Instance Savings Plans: Regional Flexibility Only
EC2 Instance SPs lock you to a specific instance family within a specific region, but allow flexibility across size, OS, and tenancy. For example, a commitment to the m5 family in us-east-1 covers m5.large, m5.xlarge, m5.2xlarge — any size, any OS, any tenancy. This is significantly more flexible than a Standard RI (which locks to a specific size) while offering better discounts than a Compute SP.
Standard RIs: Maximum Discount, Minimum Flexibility
Standard RIs lock to a specific instance type, size, region, and tenancy. They offer the best discounts but the worst flexibility. If your workload changes, you are left with unused RI capacity that you are still charged for. Mitigations include: selling unused Standard RIs on the RI Marketplace (for 1-year terms primarily); converting to Convertible RIs if available for your instance family; and using RI sharing across accounts in your Organisation.
Payment Options: Upfront vs No-Upfront
Both RIs and Savings Plans offer three payment structures: All Upfront, Partial Upfront, and No Upfront. The financial trade-off is cash flow vs. cost.
| Payment Option | Cash Flow Impact | Effective Discount (vs No Upfront) | Best For |
|---|---|---|---|
| All Upfront | High — full cost paid at purchase | Additional 2–5% vs No Upfront | Organisations with surplus capital and cost-minimisation focus |
| Partial Upfront | Medium — ~50% upfront, balance monthly | Additional 1–2.5% vs No Upfront | Balanced approach; most common for 3-year commitments |
| No Upfront | Low — monthly charges only | Baseline | OpEx-focused organisations; preserving CapEx budget |
For most enterprise organisations, No Upfront 3-year Compute Savings Plans represent the best risk-adjusted starting point. The additional discount from All Upfront rarely justifies the capital cost when you factor in the opportunity cost of capital and the risk of workload change during the term.
Stacking Strategy: The Optimal Portfolio
Best-in-class AWS cost management does not choose between Savings Plans and RIs — it layers them. The cloud commitment portfolio strategy for AWS typically follows a 70/30 rule:
Cover your baseline, predictable compute spend with Compute SPs. Target p60–p70 of your trailing 90-day average compute consumption. This gives you broad coverage with maximum flexibility for workload changes.
For workloads where instance type is stable and instance family will not change — database servers, legacy applications, dedicated infrastructure — Standard RIs provide higher discounts. Target specific, well-understood workloads only.
Maintain 10–20% of average consumption as On-Demand to absorb seasonal spikes, development workloads, and unexpected growth. Never commit 100% of consumption — utilisation rates below 85% on committed capacity erode savings.
This portfolio approach typically achieves 45–55% effective discount on compute spend — compared to 25–35% for organisations using Savings Plans alone without RIs for stable workloads.
Stacking Savings Plans and RIs with an EDP
Both Savings Plans and RIs stack with AWS EDP discounts. The mechanics: Savings Plans and RIs apply first (reducing the effective hourly rate for covered usage), then the EDP discount applies to the resulting bill. The effective combined discount is not simply additive, but the stacking is real and significant.
Example: EC2 m5.xlarge, 3-year Compute SP (54% discount) + 15% EDP discount:
On-Demand: $0.192/hour → After SP: $0.088/hour → After EDP: $0.075/hour
Effective total discount: ~61% vs On-Demand
For organisations with $5M+ EDP commitments, Savings Plans and RIs are the primary mechanism by which you drive actual cost per unit down. The EDP provides the top-level programme discount; Savings Plans and RIs do the heavy lifting on compute optimisation. For a full treatment of the AWS commitment portfolio, see our AWS cost optimisation strategies guide.
Four Scenario Verdicts
Profile: Fast-growing SaaS company, $1M/year AWS spend, containerised workloads on ECS/Fargate, agile infrastructure team.
Verdict: 100% Compute Savings Plans, 1-year term, No Upfront
Reasoning: Workloads change too fast to lock into RIs. Compute SPs cover EC2, Fargate, and Lambda with full flexibility. 1-year term reflects the growth uncertainty. Review quarterly and increase coverage as spend stabilises.
Profile: Financial services firm, $10M/year AWS spend, large fleet of m5/r5 instances running legacy applications, stable workloads, 3-year migration roadmap.
Verdict: 60% Compute SPs (3yr) + 30% Standard RIs (3yr) + 10% On-Demand
Reasoning: Legacy workloads have predictable instance needs justifying Standard RIs. Compute SPs cover newer workloads and migration flexibility. Parallel EDP negotiation recommended at this spend level.
Profile: Technology company, $5M/year AWS spend, significant SageMaker usage, GPU instances (p3/p4), variable AI/ML job volumes.
Verdict: SageMaker Savings Plans + Compute SPs for EC2 + Spot for non-critical jobs
Reasoning: SageMaker SPs provide dedicated discounts on ML instance usage. Compute SPs cover EC2 baseline. GPU RIs are high-risk due to rapid instance type evolution — avoid Standard GPU RIs. Spot Instances for batch training jobs can add 50–80% savings on non-critical AI workloads.
Profile: Global conglomerate, 50+ AWS accounts, $30M/year total spend, central FinOps team, business unit accounts with diverse workloads.
Verdict: Centralised Compute SPs (70%) + Targeted Standard RIs via RI sharing (20%) + On-Demand (10%) + EDP negotiation
Reasoning: Compute SPs managed centrally in the payer account apply automatically across all linked accounts. Standard RIs for specific stable workloads can be shared across accounts. At $30M spend, EDP negotiation should target 18–22% programme discount. Combined effective discount of 55–65% on compute is achievable.
Managing Your Commitment Portfolio
Commitments require active management. The common mistake is purchasing Savings Plans or RIs and then leaving them unreviewed for the duration of the term. Best practices for ongoing management:
Monthly utilisation review
Monitor SP and RI utilisation rates in AWS Cost Explorer. Target >85% utilisation. If utilisation consistently falls below 80%, you are over-committed — reduce at the next renewal. If consistently above 95%, you may be under-committed and leaving savings on the table.
Coverage tracking
Track your SP/RI coverage rate — the percentage of eligible usage covered by commitments. Target 70–85% coverage. 100% coverage indicates over-commitment risk; below 60% indicates sub-optimal savings.
Rightsizing before committing
Never commit to RIs or SPs for over-provisioned instances. Run rightsizing analysis first to identify instances that can be downsized — Graviton migration, instance family changes, vertical autoscaling. Committing to inefficient baselines locks in waste. See our cloud waste reduction guide for the methodology.
RI Marketplace for exit
If you have Standard RIs that are no longer needed, the AWS Reserved Instance Marketplace allows resale of eligible 1-year and 3-year Standard RIs. This provides an exit mechanism that Savings Plans lack — factoring into the decision for organisations with higher uncertainty.
Cloud Cost Optimisation Playbook
Download our expert guide covering Savings Plans strategy, RI portfolio optimisation, EDP negotiation, and FinOps governance for AWS enterprises.