The Principle: Commit to Save
AWS on-demand pricing is deliberately high for the flexibility it offers. In exchange for a 1 or 3-year commitment, AWS offers substantial discounts. Two mechanisms coexist: Reserved Instances (RI), the older system, and Savings Plans (SP), introduced in 2019 and now recommended in most cases.
Comparison Table
| Criterion | Standard RI | Convertible RI | Compute SP | EC2 Instance SP |
|---|---|---|---|---|
| Max discount (3 yr) | 72% | 54% | 66% | 72% |
| Instance family flexibility | ❌ | ✅ | ✅ | ❌ (same family) |
| Region flexibility | ❌ | ✅ | ✅ | ❌ (same region) |
| Covers Lambda/Fargate | ❌ | ❌ | ✅ | ❌ |
Sizing Strategy
The golden rule: only cover stable baseline consumption with commitments. Peak load must remain on-demand or Spot.
- Analyse 3 months of history to identify the floor consumption (baseline)
- Cover 70–80% of this baseline with Savings Plans (not 100% — leave a margin)
- Remainder on-demand or Spot depending on criticality
- Reassess every 6 months
Pitfalls to Avoid
- Over-committing: buying more RI/SP than your actual consumption — you pay for non-existent resources
- Committing before rightsizing: if you rightsize after buying RIs, the RIs no longer apply to new differently-sized instances
- Ignoring the RI Marketplace: you can resell unused Standard RIs on the AWS marketplace
- Forgetting RDS: RDS Reserved Instances work the same way and can represent 30% additional savings
Conclusion
For most workloads in 2025, Compute Savings Plans on a 1-year, no-upfront basis offer the best flexibility/savings balance. Reserve EC2 Instance Savings Plans and Standard RIs for stable, well-known components. Always rightsize before committing.