The Commitment Discount Model
On-Demand pricing is the most expensive way to run AWS workloads. You pay full price for every hour, with no commitment. AWS rewards commitment with discounts of up to 72%. The two mechanisms for making that commitment are Reserved Instances (RIs) and Savings Plans. Understanding when to use each is a core FinOps skill.
Think of it like a gym membership. On-Demand is paying the drop-in rate every visit. Reserved Instances are buying a specific class package — you get a discount, but only for that exact class. Savings Plans are buying a general membership — you get a discount on any class the gym offers.
1. Reserved Instances (The Specific Commitment)
Reserved Instances (RIs) were AWS's original commitment discount mechanism, introduced in 2009. You commit to a specific instance configuration for 1 or 3 years.
- What you commit to: Instance family, size, Region, tenancy, and OS. Example: "I will use an
m5.largerunning Linux inus-east-1for 1 year." - Payment options: All Upfront (largest discount), Partial Upfront, or No Upfront (smallest discount, but still significant vs. On-Demand).
- Flexibility:
- Standard RIs: Locked to the exact configuration. If you switch to
c5.large, the RI goes unused and you still pay for it. - Convertible RIs: You can exchange them for a different configuration, but the discount is smaller (~54% vs. ~72%).
- Standard RIs: Locked to the exact configuration. If you switch to
- Critical and irreplaceable use case: RIs are the only discount mechanism for Amazon RDS, Redshift, ElastiCache, and OpenSearch. Savings Plans do not cover these services. This is the most commonly missed fact in AWS cost optimization discussions.
2. Savings Plans (The Modern Standard)
Savings Plans, launched in 2019, replaced RIs as the recommended approach for compute workloads. Instead of committing to a specific instance, you commit to a minimum hourly spend.
- What you commit to: "I will spend at least $10.00 per hour on compute, for 1 year."
- How it works: AWS tracks your compute usage in real time. Any usage up to your committed amount gets the discounted rate automatically. Usage above your commitment is charged at On-Demand rates.
- Types:
| Type | Discount | Applies To | Flexibility |
|---|---|---|---|
| Compute Savings Plans | Up to 66% | EC2, Fargate, Lambda — any Region, any family | Highest |
| EC2 Instance Savings Plans | Up to 72% | EC2 only — specific family and Region | Lower |
Compute Savings Plans are the default recommendation for most teams. The slightly lower discount compared to EC2 Instance Savings Plans is almost always worth the flexibility — you can change instance families, move workloads between Regions, and migrate from EC2 to Fargate without losing your discount.
Sizing Your Commitment
The biggest mistake teams make is over-committing. Use the AWS Cost Explorer Savings Plans Recommendations tool, which analyzes your last 7, 30, or 60 days of usage and recommends a commitment amount. The key metric is your baseline (the minimum compute spend you sustain 24/7). Commit to your baseline; let On-Demand cover your peaks.
The Combined Strategy:
* Compute Savings Plans → EC2, Fargate, Lambda (stateless, variable compute)
* Reserved Instances → RDS, Redshift, ElastiCache, OpenSearch (stateful databases)
* On-Demand / Spot → Burst capacity, batch jobs, fault-tolerant workloads
A common interview question is: "We're migrating from EC2 to Fargate. What happens to our Reserved Instances?" The answer is: Standard RIs for EC2 do not apply to Fargate. However, Compute Savings Plans do cover both EC2 and Fargate. This is a strong argument for preferring Compute Savings Plans over EC2 RIs when your architecture is evolving. An interviewer is testing whether you understand the coverage scope of each commitment type.