If you're running significant workloads on AWS, you've probably noticed that on-demand pricing adds up quickly. The good news? AWS offers two powerful discount mechanisms: Reserved Instances (RIs) and Savings Plans. The challenge? Knowing which one to use, and when.
After analyzing hundreds of AWS accounts, we've seen organizations leave thousands of dollars on the table by choosing the wrong commitment type—or worse, avoiding commitments altogether out of fear of making the wrong choice.
Understanding the Basics
Reserved Instances
Reserved Instances have been around since 2009 and offer discounts of up to 72% compared to on-demand pricing. Here's how they work:
- Instance-specific: You commit to a specific instance type (e.g., m5.xlarge) in a specific region
- Term options: 1-year or 3-year terms
- Payment options: All upfront, partial upfront, or no upfront
- Scope: Regional or Availability Zone-specific
The key characteristic of RIs is their specificity. When you purchase an RI for an m5.xlarge in us-east-1, that's exactly what you get. If your needs change, you're stuck with that commitment.
Savings Plans
Introduced in 2019, Savings Plans offer similar discounts with more flexibility:
- Commitment-based: You commit to a dollar amount per hour (e.g., $10/hour)
- Two types: Compute Savings Plans and EC2 Instance Savings Plans
- Flexibility: Automatically applies to any matching usage
- Term options: Same 1-year or 3-year terms
The key advantage of Savings Plans is flexibility. Your commitment automatically adjusts as your usage patterns change.
The Decision Framework
When should you choose one over the other? Here's our framework:
Choose Reserved Instances When:
- Stable, predictable workloads: Databases, always-on applications
- Specific capacity requirements: You need guaranteed capacity in a specific AZ
- Marketplace liquidity matters: You may want to sell unused RIs
- Legacy systems: Older applications that won't change instance types
Choose Savings Plans When:
- Variable workloads: Applications that scale or change instance types
- Multi-service usage: You use Lambda, Fargate, or other compute services
- Modernization in progress: You're moving toward containers or serverless
- Simplicity preferred: You want automatic discount application
Real-World Scenarios
Scenario 1: The Stable Database
A company running a production PostgreSQL database on r5.2xlarge instances that hasn't changed in two years.
Recommendation: Standard Reserved Instance
Why: The workload is stable, predictable, and unlikely to change. The RI provides maximum discount for minimum risk.
Scenario 2: The Growing Startup
A startup with rapidly changing workloads, experimenting with different instance types, and planning to adopt EKS.
Recommendation: Compute Savings Plan
Why: Maximum flexibility to adapt as the infrastructure evolves. The discount applies automatically regardless of instance changes.
Scenario 3: The Hybrid Approach
An enterprise with both stable core infrastructure and dynamic application workloads.
Recommendation: Combination strategy
- RIs for known, stable workloads (databases, core services)
- Savings Plans for variable application tiers
- On-demand for truly unpredictable burst capacity
Common Mistakes to Avoid
1. Over-committing
We frequently see organizations purchase commitments for 100% of their current usage. This leaves no room for optimization or changing needs.
Best Practice: Start with 70-80% coverage of your baseline usage.
2. Ignoring Utilization Data
Many teams purchase commitments based on current instance counts without analyzing actual utilization. If your instances are oversized, you're committing to waste.
Best Practice: Right-size first, then commit.
3. Analysis Paralysis
Some organizations are so afraid of making the wrong choice that they stick with on-demand pricing indefinitely.
Best Practice: Even a suboptimal commitment usually beats on-demand. Start small and iterate.
4. Forgetting About Convertible RIs
Standard RIs offer better discounts but less flexibility. Convertible RIs can be exchanged for different instance types.
Best Practice: Consider Convertible RIs for workloads that might need different instance families.
The Math: A Real Example
Let's look at a concrete example with m5.xlarge in us-east-1:
| Option | Hourly Cost | Annual Cost | Savings |
|---|---|---|---|
| On-Demand | $0.192 | $1,682 | - |
| 1-Year RI (No Upfront) | $0.121 | $1,060 | 37% |
| 1-Year Savings Plan | $0.122 | $1,069 | 36% |
| 3-Year RI (All Upfront) | $0.076 | $666 | 60% |
| 3-Year Savings Plan | $0.077 | $675 | 60% |
The discount difference between RIs and Savings Plans is minimal. The real question is flexibility vs. specificity.
Key Takeaways
- RIs offer maximum discount for predictable, specific workloads
- Savings Plans provide flexibility for changing environments
- Hybrid approaches often work best for complex organizations
- Start conservatively and increase commitment over time
- Right-size before committing to avoid locking in waste
What's Next?
Understanding your options is the first step. The next step is analyzing your actual usage patterns to determine the right mix for your organization.
Our free waste scanner can identify not only what commitments would save you money, but also which resources should be right-sized before you commit. Combined with our RI/SP coverage analysis, you'll have a clear picture of your optimization opportunities.
Ready to find your optimal commitment strategy? Start your free scan and get personalized recommendations for your AWS environment.
