In 2019, AWS introduced Savings Plans, which have largely superseded RIs for most organizations because they are dramatically simpler to manage.
The key conceptual shift: instead of committing to a specific instance type (like with RIs), you commit to a specific amount of hourly spend — for example, "$70 per hour" — for a 1 or 3-year term. Any usage up to that commitment amount is automatically charged at the discounted Savings Plan rate. Any usage above that amount falls back to the regular On-Demand rate.
This means you never have to think about whether your RI "matches" your current instance type. The discount applies automatically to whatever compute you're running, up to your committed dollar amount.
1. EC2 Instance Savings Plans
- How it works: You commit to spend a certain amount per hour on a specific instance family in a specific region (e.g., "M-family instances in us-east-1"). The discount applies regardless of instance size within that family.
- Flexibility: Low — locked to a family and region, similar to Standard RIs.
- Discount: High (up to ~72%).
- Best for: Stable, predictable workloads where you know the instance family and region won't change.
2. Compute Savings Plans (The Best Practice)
- How it works: You commit to a total hourly spend amount on any compute, anywhere.
- Flexibility: Extremely high. Automatically applies to EC2 instances of any family, in any region, running any OS. Also applies to AWS Fargate (containers) and AWS Lambda (serverless functions).
- Discount: Moderate (up to ~66%).
- Best for: Most organizations. The flexibility means you never have to worry about your commitment becoming misaligned with your infrastructure as it evolves.
For most organizations, Compute Savings Plans are the preferred choice. The ~6% discount gap versus EC2 Instance Savings Plans is almost always worth the operational simplicity and flexibility — especially as teams migrate between instance generations or adopt Fargate and Lambda over time.
Sizing Your Commitment: The Conservative Approach
A common mistake is over-committing. The right strategy is to commit only to your minimum guaranteed baseline — the floor of your usage that you are certain will never drop below that level. Cover the rest with On-Demand or Spot. It is always better to slightly under-commit and pay On-Demand on the margin than to over-commit and pay for unused capacity.