AWS Savings Plans, mapped. Compute, SageMaker, and EC2 shapes across the commitment ladder.

Savings Plans are not a cost saving. They are a forecast bet. This hub provides the buyer side framework that protects the bill before the EDP conversation closes the next three years.

AWS Savings Plans are the most important contractual lever on the AWS bill. Once a Savings Plan is signed, the next three years of compute cost are locked into a curve the buyer side can no longer move.

Sized well, the discount carries the cloud business case. Sized badly, it sets the bottom of the savings ceiling for the next renewal.

This hub covers the three plan families, the math, and the buyer side framework for sizing the commitment. Read alongside the AWS advisory practice, the EDP negotiation guide, the EDP commitment calculator, and the Reserved Instances versus Savings Plans comparison.

The three Savings Plan families

Compute Savings Plan

The Compute Savings Plan applies across EC2, Fargate, and Lambda. It is the most flexible family. The discount sits around twenty five percent at one year and over fifty percent at three year all upfront.

EC2 Instance Savings Plan

EC2 Instance plans lock a specific instance family in a specific region. Less flexible than Compute. Higher discount in the same band, typically up to sixty six percent at three year all upfront.

SageMaker Savings Plan

SageMaker plans cover ML compute, training jobs, and inference endpoints. Useful for established ML platforms. Risky for early stage ML estates because the workload shape is still unstable.

Sizing the commit

Baseline workload

The first sizing question is the baseline. Find the steady state compute hour count across the trailing six months. That number is the floor.

Elastic remainder

• Spike workloads. Leave on demand. Do not commit.
• Predictable batch. Cover with Compute Savings Plan.
• Steady production. Cover with EC2 Instance Savings Plan in the right family and region.
• Dev and test. Schedule and right size before committing.

Renewal math

The Savings Plan does not renew automatically. It expires. The buyer side question at renewal is whether to roll or rebuild the commit shape.

Reserved Instance overlap

Coverage overlap rules

RIs and Savings Plans apply in priority order. RIs cover first. Savings Plans cover the remainder. The combined coverage cannot exceed actual use.

When to mix

• RIs for database workloads that benefit from convertible exchange.
• Compute Savings Plans for the elastic remainder.
• EC2 Instance Plans for legacy steady workloads in one family.
• SageMaker Plans for established ML estates only.

Common commit traps

Oversizing

The most common error is oversizing the Compute Savings Plan. The discount is real, but the wasted spend on uncovered hours often outpaces the savings.

Region lock

• Single region commit. Lower flexibility, higher discount.
• Multi region estate. Use Compute family across regions.
• Disaster recovery posture. Confirm DR region eligibility before commit.
• Migration in flight. Defer commit until target region is stable.

What to do next

1. Pull the trailing six month compute hour count and identify the steady baseline.
2. Classify workloads into spike, predictable batch, steady production, and dev or test.
3. Right size the steady baseline and confirm the family and region mix.
4. Select the right plan family for each bucket of baseline spend.
5. Layer reservations underneath the Savings Plan for database workloads that need convertibility.
6. Negotiate the commit shape with AWS as part of the EDP, not afterwards.
7. Set utilization alerts at seventy and ninety percent of plan coverage.
8. Brief the AWS advisory practice and run the commitment proposal through a buyer side review.

Frequently asked questions

Are Savings Plans cancellable?

No. The commit is binding for the full term. Spend obligations cannot be exited mid term. The buyer side question is therefore conservative sizing, not aggressive coverage.

Can we combine Reserved Instances and Savings Plans?

Yes. RIs apply first, Savings Plans cover the remainder. The combined coverage cannot exceed actual compute use.

Should we commit three year all upfront?

Only for workloads that are demonstrably steady. The discount is deepest but the cash exposure and the inflexibility compound. Most enterprises mix terms and payment options.

How do Savings Plans interact with the EDP?

A Savings Plan signed inside the EDP can count toward the EDP commitment and deepen the discount tier. Sign the plan after the EDP and the leverage is gone.

What is the difference between Compute and EC2 Instance Savings Plan?

Compute covers EC2, Fargate, and Lambda across families and regions. EC2 Instance locks one family in one region for a deeper discount.

Should we move SageMaker workloads to a Savings Plan?

Only for established ML estates with stable training and inference patterns. Early stage ML workloads change shape too quickly to commit.

How often should we review Savings Plan coverage?

Monthly for the first quarter after signing. Quarterly thereafter. Coverage gaps and excess coverage both leak money.