Reserved Instances vs. Savings Plans: The Right AWS Reservation Strategy

Paying on-demand prices at AWS is like paying the peak daily price at the gas station every day. Reservations are the equivalent of a framework agreement — and save 30-50% without any restrictions on performance or availability.

The Three Reservation Models Compared

Reserved Instances (RIs)

The classic model: you reserve a specific instance family in a specific region for 1 or 3 years. Standard RIs offer the highest savings (up to 72% with 3-year all-upfront), but are tied to the chosen instance type.

Convertible RIs allow switching to equal or higher-value instance types during the term. The savings are slightly lower (up to 66%), but flexibility is significantly higher.

• Best choice for: Stable workloads with a known instance type (e.g., RDS databases, dedicated application servers).
• Caution with: Workloads that change frequently or are planned for Graviton migration.

Compute Savings Plans

The most flexible model: you commit to an hourly amount (e.g., $10/hour) over 1 or 3 years. The discount applies automatically to EC2, Fargate, and Lambda — regardless of instance family, size, region, or operating system.

• Best choice for: Companies with mixed workloads that need flexibility in instance selection.
• Savings: Up to 66% compared to on-demand.

EC2 Instance Savings Plans

A middle ground: commit to a specific instance family in a region, but flexible on size and operating system. Offers slightly more savings than Compute Savings Plans with less flexibility.

• Best choice for: Teams that know their instance family (e.g., m6i) but want to switch between sizes.
• Savings: Up to 72% compared to on-demand.

Our Recommendation: The Layered Approach

In practice, we combine different reservation models:

• Layer 1 — RIs (40-50%): For stable baseline workloads such as RDS, ElastiCache, and dedicated EC2 instances that rarely change.
• Layer 2 — Savings Plans (20-30%): For the variable portion of compute usage that needs to remain flexible.
• Layer 3 — On-Demand + Spot (20-30%): For load spikes and batch workloads.

The Right Timing

Analyze at least 30 days of usage data before reserving. AWS Cost Explorer offers solid guidance under Reservation Recommendations. Start conservatively with 1-year no-upfront — this minimizes risk while still delivering 30-40% savings.

Those who review their reservation strategy annually and adapt it to changing workloads will consistently get the most out of their AWS budget.