One page on every AWS commitment: what each Savings Plan and reserved instance actually discounts, how the hourly commitment is applied, how to size a purchase you will not regret, how benefits are shared across an organization, and how to keep utilization near 100 percent. The numbers below are what we check on customer payer accounts every week.
Updated 24 September 2026 · 12 min read · by the TruCost.Cloud FinOps team
Every AWS commitment is the same bargain: you promise to pay for a certain amount of usage for one or three years, and AWS charges that usage at a lower rate. What differs is what you commit to. A Savings Plan commits to an amount of spend per hour, measured at the discounted rate, and follows your usage across the resources the plan covers. A reserved instance (or reserved node, or reserved capacity) commits to a specific thing: an instance class in a region, an engine, a number of read units.
Three mechanics matter more than any discount percentage:
AWS documents the families in Savings Plans types; the Database family was announced in December 2025 with its own pricing page.
| Family | Covers | Flexibility | Maximum discount | Terms and payment |
|---|---|---|---|---|
| Compute Savings Plans | EC2 (any family, size, region, OS, tenancy), Fargate, Lambda | Highest: follows a workload from one family or region to another, and from EC2 to containers or serverless | Up to 66% | 1 or 3 years; all, partial or no upfront |
| EC2 Instance Savings Plans | One EC2 instance family in one region (any size, OS, tenancy) | Within the family and region only | Up to 72% | 1 or 3 years; all, partial or no upfront |
| SageMaker Savings Plans | SageMaker instance usage (notebooks, training, inference) | Across families, sizes and regions within SageMaker | Up to 64% | 1 or 3 years; all, partial or no upfront |
| Database Savings Plans (December 2025) | Aurora, RDS (all engines), DynamoDB, ElastiCache, DocumentDB, Neptune, Keyspaces, Timestream, DMS: instances, serverless and throughput, generation 7 and newer | Across engines, instance families, regions and from provisioned to serverless | Up to 35% serverless, 20% provisioned instances, 18% on-demand throughput, 12% provisioned capacity | 1 year only, no upfront only |
The rule of thumb: the more flexible the plan, the smaller the discount. A Compute Savings Plan is the right default for a baseline that will change shape; an EC2 Instance Savings Plan pays a few more points for a family you know you will keep. Savings Plans never cover storage, data transfer, licences billed separately, or (for Database Savings Plans) anything on generation 6 and older. The Database Savings Plans deep dive covers the eligible engines, the discount tiers and the console purchase flow in detail.
Reservations predate Savings Plans and still win in three situations: when you want a capacity reservation as well as a discount, when a service has no Savings Plan (OpenSearch, Redshift), and when you want to pay up front for a database, since Database Savings Plans are no-upfront only.
| Situation | Buy | Why |
|---|---|---|
| Stable compute baseline that changes shape (new families, Graviton migration, some containers) | Compute Savings Plan | The discount follows the workload; nothing to manage when architecture changes |
| Stable usage of one EC2 family in one region for years | EC2 Instance Savings Plan, or Standard RI if you may need to resell | A few more points of discount for a commitment you are sure of |
| You need guaranteed capacity in one Availability Zone | Zonal Standard RI (or On-Demand Capacity Reservation plus a Savings Plan) | Savings Plans never reserve capacity |
| Databases that will be upgraded, migrated between engines or moved to serverless within the year | Database Savings Plan | Cross-engine, cross-region, provisioned-to-serverless; one year, no upfront |
| A database you will keep exactly as it is for three years and can pay for up front | RDS reserved instance | Deeper discount and up-front payment options the Savings Plan does not offer |
| Provisioned DynamoDB tables with steady throughput | DynamoDB reserved capacity | Far deeper than the 12 percent Database Savings Plan rate |
| OpenSearch, Redshift | Reserved instances / nodes | No Savings Plan exists for these services |
| Bursty or interruptible capacity above the baseline | Nothing: Spot for interruptible work, on-demand for the rest | A commitment you cannot fill every hour is a liability with a payment schedule |
For any specific instance, the AWS instance pricing calculator shows on-demand, every reserved and Savings Plan option and spot side by side at the hours you run it, with the break-even.
Utilization is how much of the commitment you used: a $10 an hour plan with $9 of eligible usage is 90 percent utilized. Coverage is how much of your eligible spend a commitment covered: if $70 of a $100 hour of compute was discounted by plans, coverage is 70 percent.
The trap is optimizing coverage. We regularly see accounts at 95 percent coverage and 70 percent utilization, which means 30 percent of the commitment is paid for and unused every hour. Coverage is a vanity metric unless utilization backs it up. Targets that hold up in practice: utilization at 95 to 100 percent on the baseline commitment (anything sustained below 90 percent warrants an audit), and coverage wherever it lands once every commitment is fully used, typically 70 to 85 percent of eligible spend, with the rest deliberately left on on-demand and Spot. The utilization deep dive walks through the hourly mechanics and the four-step audit.
A commitment beats on-demand only if the instance runs enough hours. The break-even is the number of hours a month at which the commitment's effective hourly cost (upfront amortized plus any hourly fee) equals the on-demand cost for the same hours. For a typical one-year no-upfront EC2 Instance Savings Plan the break-even sits around 450 to 500 hours a month; a three-year all-upfront reservation needs the instance running essentially all month for three years. Business-hours workloads, at roughly 200 hours a month, rarely earn back any commitment, which is why they belong on schedules, not plans.
Payment options trade cash for discount. All upfront gives the deepest rate, partial upfront a few points less, no upfront the least, but the gap is smaller than most people expect (usually 3 to 7 points) and the no-upfront option keeps the money and the flexibility. Amortization matters for reporting: AWS Budgets and Cost Explorer can show net amortized cost, which spreads an upfront payment evenly across the term, so a team's monthly cost reflects the commitment it uses rather than a spike in the month of purchase. The Budgets article covers the net unblended and net amortized metrics added in April 2025.
By default a commitment bought anywhere in an AWS Organization benefits the whole organization, applied to the highest-discount usage first. That maximizes the discount and breaks accountability: the platform team buys a plan, a data team's cluster consumes it, and the chargeback report shows the wrong team saving money.
Since late 2025, Reserved Instance and Savings Plans group sharing lets you define groups with Cost Categories and choose how benefits flow:
Each account belongs to one group; existing commitments are not changed (see discount sharing in the Billing user guide), only where their benefit lands; the setting is in Billing preferences (configured in us-east-1, applied globally) and takes effect for the current month. Prioritized is the right default: it keeps the accountability and only sacrifices the spillover when a group over-buys. Restricted exists for compliance boundaries and grant-funded accounts. The group sharing deep dive covers the setup and the FinOps implications. If you do not use group sharing, the honest alternative for chargeback is to allocate the amortized commitment cost to teams by the usage that consumed it, which is what allocation tools do from the Cost and Usage Report.
The Commitments module reconciles Savings Plan and reserved instance utilization and coverage to Cost Explorer to the cent, per account and per business unit under your allocation rules, shows the expiry calendar, and sizes the next purchase from hourly usage rather than monthly totals. Amortized commitment cost is allocated to the teams whose usage consumed it, so chargeback reflects reality even without group sharing. See Commitments on the platform page.
A Savings Plan commits to an hourly spend and follows your usage across the resources it covers (families, sizes, regions, and for Compute plans, Fargate and Lambda). A reserved instance commits to a specific instance class in a region or Availability Zone. Reservations can be a few points cheaper and can reserve capacity; Savings Plans need no management when the architecture changes.
You pay the full hourly commitment anyway. Commitments are evaluated every hour and unused amounts do not roll over. That is why utilization, not coverage, is the number to watch.
Only within 7 days of purchase, in the same calendar month, if the hourly commitment is $100 or less, and at most 10 returns per management account per year. There is no marketplace for Savings Plans. Standard EC2 reserved instances can be sold on the Reserved Instance Marketplace instead.
Launched in December 2025, they cover Aurora, RDS, DynamoDB, ElastiCache, DocumentDB, Neptune, Keyspaces, Timestream and DMS with one commitment, across engines and regions, for generation 7 and newer. Discounts are up to 35 percent on serverless, 20 percent on provisioned instances, 18 percent on on-demand throughput and 12 percent on provisioned capacity. Only a one-year, no-upfront term is offered.
Not on the same usage. A reservation applies first to the instance it matches; a Database Savings Plan then covers eligible usage that has no reservation. Many organizations keep reservations on their most stable databases and use the plan for everything in flux.
95 to 100 percent on the baseline commitment. Sustained utilization below 90 percent means part of the commitment is paid for and unused every hour and warrants an audit. Coverage then lands wherever full utilization puts it, typically 70 to 85 percent of eligible spend.
By default a plan bought in any account applies first to that account and then to any account in the organization, on the highest-discount usage first. Reserved Instance and Savings Plans group sharing, added in late 2025, lets you keep benefits inside a Cost Category group (prioritized, with spillover, or restricted, without).
It depends on the option. A one-year no-upfront EC2 Instance Savings Plan typically breaks even around 450 to 500 hours a month; a three-year all-upfront reservation needs near-continuous use. The instance pricing calculator shows the break-even for any type at your hours.
The articles this guide draws on, each on one narrower question.

Learn how to improve AWS Savings Plan utilization, avoid unused commitment, size plans correctly, and monitor coverage to reduce AWS cloud costs.

Learn how AWS Database Savings Plans cut database costs by up to 35% with flexible discounts across Aurora, RDS, DynamoDB, ElastiCache, and more. Perfect for Fi…

Learn how new AWS Reserved Instances & Savings Plans Group Sharing brings precise cost control, accountability, compliance, and accurate ROI tracking for enterp…

In today’s fast-paced, cloud-driven world, businesses are constantly striving to strike the perfect balance between performance and cost. As cloud adoption skyr…

AWS Cost Explorer now provides purchase recommendations for Amazon DynamoDB reserved capacity

Managing cloud costs on AWS can feel like trying to solve a puzzle—just when you think you’ve got the pieces in place, they shift again. Cloud budgeting has oft…
Thirty minutes with a FinOps engineer, on your AWS data if you like. No slideware.
Tell us a little about your AWS setup and we will come back within one business day — or pick a slot straight away.