Reserved Instances vs. Savings Plans: Pre-pay for Cloud Discounts

WHY IT EXISTS: Cloud providers' on-demand pricing is flexible but expensive for constant workloads. To incentivize long-term customer commitment and improve their own capacity planning, providers created discount models for predictable usage. This lets customers lock in lower rates on their baseline compute, and providers secure revenue.

THE MENTAL MODEL: RIs and Savings Plans are financial instruments, not resource reservations. You're not setting aside a specific server; you're pre-purchasing a discount voucher that automatically applies to matching usage. A Reserved Instance is a voucher for a specific instance type in a specific region (e.g., an m5.large in us-east-1). A Savings Plan is a more flexible voucher for a dollar amount of compute spend per hour (e.g., $10/hour), which can apply across different instance families and even regions.

HOW IT WORKS: You commit to either a specific instance configuration (Standard RI) or a dollar amount of hourly spend (Savings Plan) for a 1- or 3-year term. You can pay all upfront, partial upfront, or no upfront, with larger upfront payments yielding bigger discounts. The cloud provider's billing system automatically applies these discounts to any matching on-demand usage. For example, if you have a 5/hour Compute Savings Plan and run 8/hour worth of various EC2 instances, the first 5/hour is covered by the plan's discounted rate, and the remaining 3/hour is billed at the standard on-demand rate.

WHEN TO USE IT: Use these for your absolute baseline, 24/7 workloads. Analyze your usage over several months to find the minimum consistent level of compute you're running. This "always-on" footprint is a safe candidate for commitment. Compute Savings Plans are generally the best starting point due to their flexibility across instance types, families, and regions.

WHEN NOT TO USE IT: Avoid RIs and Savings Plans for spiky, unpredictable, or short-term workloads like development environments, CI/CD runners, or one-off data processing jobs. Using on-demand or Spot Instances is far more cost-effective for these cases. Also, be cautious if you anticipate major architectural changes, as a Standard RI could lock you into an obsolete instance type.

ONE CANONICAL EXAMPLE: A company runs a web application on three m5.large EC2 instances 24/7. Instead of paying the on-demand rate of ~$0.096/hour per instance, they purchase a 3-year Compute Savings Plan. They commit to spending a certain amount per hour which covers their usage. This reduces their effective hourly rate by over 60%. If they later decide to upgrade to m6g.large instances, the Savings Plan discount automatically applies to the new instances, unlike a Standard RI which would not.