Cloud Commitment Strategies: Reserved Instances, Savings Plans, and CUDs

Navigate cloud provider commitment discounts to reduce compute costs by matching commitments to measured steady-state usage. Covers AWS Reserved Instances vs Savings Plans, Azure Reservations, GCP Committed Use Discounts, coverage analysis, and the financial modeling needed to make smart commitment decisions.

Cloud commitment strategies—Reserved Instances, Savings Plans, and Commitment Use Discounts—are the primary levers cloud operators use to reduce compute spend without rearchitecting workloads. They matter when you have baseline, predictable usage that can amortize a commitment, or when you want to lower the effective hourly rate while keeping the flexibility to shift instance types or regions. This page covers how to review existing commitments, select the right model for a given workload, purchase and modify instruments, and the practical failure modes that surface in production.

1. Surveying Existing Commitments

Before purchasing or renegotiating, you need a clear picture of what’s already locked in. Most providers expose commitment inventory through their CLI.

AWS CLI example:

aws ec2 describe-reserved-instances
aws ec2 describe-savings-plans

The output lists reserved-instance-id, instance-type, region, start, end, and the discount relative to on-demand. Across providers, the pattern is similar: a list of active commitments, each scoped to a region or account, with a remaining term and a discount rate.

Common blind spot: Commitments are often region-scoped by default. A Reserved Instance purchased in us-east-1 has no effect in eu-west-3. Savings Plans add a scope field (Region vs AvailabilityZone) that determines portability. Teams that assume cross-region applicability routinely find unused discounts gathering dust in the wrong region.

2. Choosing the Right Model for a Workload

Three structural patterns dominate the market. Understanding which matches your workload profile is the difference between a discount that compounds and one that becomes a sunk cost.

Concrete CLI check (AWS):

aws ec2 describe-reserved-instances-offerings \
  --instance-type m5.large \
  --region us-east-1

The response includes offering-id, duration, fixed-price, recurring-charge, and offering-type (standard vs convertible). For Savings Plans:

aws savings-plans describe-savings-plans

Lists active plans with plan-id, commitment, term, scope, and state.

Pattern note: The effective hourly rate is a fixed proportion of the on-demand rate, set at purchase time and immutable without modifying or canceling the commitment. Nothing in the CLI or API lets you retroactively adjust the rate.

3. Purchasing a Commitment

The purchasing flow follows a consistent pattern across clouds, though the CLI flags and UI paths differ.

  1. Profile actual usage: Pull hourly metrics for the target instance family from your billing export (e.g., AWS Cost and Usage Report in BigQuery, or the provider’s equivalent export pipeline). Calculate the 30-day average hourly spend.
  2. Select term and payment option: Longer terms and upfront payment options reduce the effective hourly cost, though they commit you for a longer period. No-upfront or monthly options preserve cash flow at a higher effective rate.
  3. Choose scope: Region-scoped commitments cover any matching instance in the region; AZ-scoped commitments are cheaper but tie you to a specific availability zone.
  4. Execute the purchase: Use the provider’s purchase endpoint or CLI.

AWS CLI purchase examples:

Standard RI:

aws ec2 purchase-reserved-instances-offering \
  --instance-count 10 \
  --offering-id ri-0a1b2c3d4e5f6a7b8 \
  --instance-type m5.large \
  --region us-east-1

Compute Savings Plan:

aws savings-plans create-savings-plan \
  --plan-type Compute \
  --commitment 600 \
  --term-years 3 \
  --payment-option AllUpfront \
  --region us-east-1

The CLI returns a savings-plan-id and a start-time. The commitment becomes active immediately; the discounted rate begins applying to matching usage in the next billing cycle.

Generic failure point: If the instance family you’ve committed to undergoes a rename or deprecation by the provider (e.g., m5 → m6i with a new microarchitecture), existing RIs may not auto-map. The CLI will not apply the discount to the new family, and you’ll see full on-demand charges until you either purchase new RIs or modify the existing ones (if the provider supports it).

4. Modifying and Cancelling Commitments

Workloads evolve. How you adjust a commitment depends on the model.

Concrete CLI modify example (AWS):

aws ec2 modify-reserved-instances \
  --reserved-instance-ids ri-1a2b3c4d5e6f7g8h9 \
  --instance-count 5

If the modification is invalid (e.g., crossing families), the CLI responds with:

An error occurred (InvalidParameterCombination) when calling the ModifyReservedInstances operation: Cannot modify reserved instance to a different instance type family.

Sharp edge: Modification windows are sometimes silent. The CLI may accept a request that appears to succeed but the discount only applies to new launches after the modification, not to already-running instances. You must restart or replace instances for the new commitment to take effect.

5. Sharp Edges – What Silently Fails or Catches Teams

This section documents the practical gotchas that surface after deployment, not theory.

6. Practical Workflow Checklist

When a cost-optimization ticket lands in your queue, run through these steps in order. They’re ordered to avoid the most common rework cycles.

  1. Inventory: Pull active commitments.
   aws ec2 describe-reserved-instances
   aws savings-plans describe-savings-plans

Note scope, instance-family, start, end, and utilization metadata.

  1. Usage profiling: Export the last 30 days of hourly compute spend per instance family from your billing data store. Identify families where average hourly usage constitutes a substantial portion of the instance’s on-demand capacity.
  1. Model decision:
  1. Purchase: Execute the CLI command matching your model. Double-check scope and payment-option before confirming. For Savings Plans, set the commitment to the 30-day average, not the peak.
  1. Verify: In the next billing cycle, check the commitment utilization metric. If utilization is low after 30 days, re-evaluate the commitment amount or switch models.
  1. Document: Record commitment-id, term, scope, instance-family, and expected engagement pattern in your cost-allocation tagging schema (e.g., CostCenter, Environment, CommitmentModel). Future teams need this context to avoid repeating the same assumptions.

This page was rewritten on 26 October 2026. It replaced a templated version that was not specific to its own title. How these pages are written and checked. If anything here is wrong, tell us at hello@thegarnetwiki.com and we will correct it.