Reserved Instances vs. Savings Plans: An Honest Comparison

AWS introduced Savings Plans in 2019 and has positioned them as the successor to EC2 Reserved Instances for most use cases. The AWS documentation emphasizes Savings Plans' flexibility — they apply across instance families and regions, cover Fargate and Lambda in addition to EC2, and do not require committing to specific instance types or sizes. For organizations that are migrating workloads or changing instance families frequently, this flexibility is genuinely valuable.

But Savings Plans are not strictly better than Reserved Instances. For organizations with stable workloads that have completed right-sizing and have a clear picture of their committed compute baseline, Reserved Instances can provide equal or better discount rates with more predictable coverage. The correct choice depends on three variables: workload stability, commitment horizon, and the mix of services you need to cover. This article analyzes each.

Discount Rate Comparison: When RIs Win

For EC2 instances in stable configurations (same instance family, same region, same operating system for 1-3 years), 3-year No Upfront Reserved Instances provide a 60-62% discount relative to On-Demand pricing. Compute Savings Plans provide a maximum discount of 66% for the most popular configurations but average 54-58% across all instance families. EC2 Instance Savings Plans (locked to a specific instance family) provide up to 72% discount for specific configurations.

The comparison is not clean because the savings rate depends heavily on which specific instance types you are committing to. For common workloads running on m5 or c5 instances in us-east-1, the discount rates between 1-year All Upfront RIs and 1-year Compute Savings Plans are within 3-5 percentage points of each other, and the Savings Plan flexibility advantage outweighs the small discount difference.

For less common configurations — Graviton instances (r6g, m6g), instances in less popular regions, or Windows instances — Reserved Instance discounts can be 8-12 percentage points higher than equivalent Savings Plans discounts. For organizations running significant workloads in these configurations, the RI discount advantage is material enough to justify the flexibility trade-off.

Coverage Mechanics: The Critical Difference

The operational difference between Reserved Instances and Savings Plans that most organizations underestimate is how coverage is applied. Reserved Instances cover specific instance configurations: a 1-year m5.2xlarge RI covers m5.2xlarge instances running Linux in the specified region and AZ. If you right-size those instances to m5.xlarge in six months, the RI coverage no longer applies and you are paying both the RI commitment and On-Demand for the m5.xlarge.

Compute Savings Plans apply to any EC2 compute usage up to the committed dollar amount per hour. If you commit to $10/hour of Compute Savings Plans and your m5.2xlarge instances are right-sized to m5.xlarge, the Savings Plan automatically applies to the smaller instance. No RI modification required, no coverage gap, no wasted commitment.

This is the most cited argument for Savings Plans, and it is valid for organizations in the middle of a right-sizing program. As discussed in our article on right-sizing methodology, completing a right-sizing cycle takes 3-6 months for a large account. Buying Reserved Instances before right-sizing is complete locks in commitment at the current (over-provisioned) instance size. Savings Plans provide a natural hedge against the coverage mismatch that results from right-sizing a committed workload.

The Right Sequencing: Right-Size First, Then Commit

The most common FinOps mistake we see is purchasing Reserved Instances or Savings Plans before completing right-sizing. The commitment is made at current instance sizes, which are over-provisioned. After right-sizing reduces the fleet, the commitments no longer align with actual usage. The result is either unused RI capacity (which you pay for regardless) or a Savings Plan utilization rate below 80% (which means you committed to more than you are using).

The correct sequencing is: complete the right-sizing cycle, verify the resulting instance fleet runs stably for 4-6 weeks at the right-sized configuration, then analyze the committed compute baseline and purchase commitments against that baseline. This sequencing delays the discount benefit by 3-4 months but produces commitments that are well-utilized for the full commitment term.

For a practical commitment strategy after right-sizing: use Compute Savings Plans to cover 70-80% of your steady-state compute baseline. The 20-30% uncovered by Savings Plans handles natural variance in the fleet without wasting committed capacity. Purchase 1-year terms unless you have high confidence in 3-year stability — the discount premium for 3-year terms is approximately 15-20%, but the risk of technology changes (new instance families, architecture migrations to containers) making the 3-year commitment inefficient is non-trivial.

Reserved Instance Marketplace: The Overlooked Liquidity Option

One RI advantage that is rarely discussed: the AWS Marketplace allows you to sell Reserved Instances before their term expires. If you commit to 1-year Standard RIs and your workload changes significantly after 6 months, you can sell the remaining commitment on the RI Marketplace at a discount, recovering a portion of the commitment cost.

Savings Plans cannot be sold on the Marketplace. If you commit to a 1-year Savings Plan and your compute usage drops significantly (a product is shut down, a large customer churns), you are committed to paying the full hourly rate for the remaining term regardless of whether you use the capacity.

The RI Marketplace liquidity option matters for organizations with uncertain commitment horizons — early-stage startups, companies undergoing significant product pivots, or companies with lumpy growth patterns. For these organizations, the flexibility of Savings Plans on paper is partially offset by the inflexibility of the Savings Plan commitment itself. A Standard RI provides both the flexibility to change instance configurations (via modification) and an exit option (via Marketplace sale) that Savings Plans do not provide.

Multi-Service Coverage: Where Savings Plans Win Clearly

For organizations running significant workloads on AWS Fargate or Lambda in addition to EC2, Compute Savings Plans provide the clearest advantage. Fargate and Lambda are covered by Compute Savings Plans and are not covered by EC2 Reserved Instances. If your organization is migrating workloads from EC2 to Fargate as part of a containerization program, Compute Savings Plans provide coverage throughout the migration without requiring separate commitment instruments for EC2 and Fargate.

The Lambda coverage in Savings Plans is primarily relevant for high-volume Lambda deployments — the discount applies to compute duration charges (GB-seconds), not to invocation charges. For Lambda deployments with hundreds of millions of monthly invocations, the savings are material. For typical Lambda usage in a microservices architecture, the discount on Lambda charges is a small component of total Savings Plan value.

Practical Decision Framework

The choice between Reserved Instances and Compute Savings Plans should be made workload-by-workload rather than as an organization-wide policy. For workloads that are: stable for 1-3 years, running on non-mainstream instance families where RI discounts are substantially higher, or subject to potential architectural changes that an RI Marketplace sale could hedge — Reserved Instances are the better instrument. For workloads that are: in the middle of right-sizing, migrating between instance families, or split between EC2 and Fargate — Compute Savings Plans provide better coverage coverage with less operational management overhead.

In practice, most mature engineering organizations end up with a mix: Compute Savings Plans covering the baseline compute tier that spans multiple services and may change instance families over time, and Standard Reserved Instances for specific stable workloads (database instances, long-running analytics clusters) where the configuration is unlikely to change and the RI discount advantage over Savings Plans is measurable.