What you actually trade away
AWS gives you a ladder of commitment instruments, and each rung trades discount for freedom. Compute Savings Plans hold their rate while you move across instance family, size, region, operating system, and across EC2, Fargate, and Lambda. EC2 Instance Savings Plans lock you to a family in a region for a slightly deeper rate. Standard Reserved Instances lock tighter still. Convertible Reserved Instances let you exchange for a different type but not refund. Term adds a second axis on top of scope: one year or three year, with the three year rate roughly 8 to 15 points deeper depending on the instrument.
So a buyer faces two questions at once. How tightly do I want to be scoped, and how long do I want to be locked. The flexible, short answers cost a little discount. The rigid, long answers pay the most per hour but expose you to stranded commitment if the workload changes.
How to value the flexibility you keep
The cleanest way to decide is to put a number on the risk. Estimate the probability that the covered workload changes shape inside the term, through a rearchitecting effort, a migration, a shrink, or a retirement. Multiply that probability by the committed spend you would strand. Compare it against the extra discount the longer or tighter instrument would have saved. If the expected stranded cost is larger than the saving, you are buying a discount you cannot afford to keep.
This is why coverage should follow a defensible forecast rather than the deepest rate on the rate card. Commitments discount roughly 20 to 72 percent against on demand, and the temptation is to maximise that number. The discipline is to commit only the part of your estate you are confident will still exist, in the same shape, when the term ends.
A worked example
Indicative figures, verified against the client's billing data, anonymized. A Fortune 500 retailer runs a steady base of EC2 compute worth 200,000 USD a month on demand and is choosing how to cover it. The team expects a platform rebuild in roughly 18 months that will move about 40 percent of this workload to Graviton and Fargate.
| Coverage choice | Discount captured | Stranded risk at the rebuild | Indicative net position |
|---|---|---|---|
| Three year standard Reserved Instances on the full base | Deepest | High: 40 percent locked to old families | Saves most per hour, strands roughly 29,000 USD a month after the rebuild |
| Three year Compute Savings Plan on the full base | Deep | Low: discount follows the move to Graviton and Fargate | Captures most of the discount and survives the rebuild intact |
| One year Compute Savings Plan, renewed | Shallower by about 11 points | None | Pays more per hour but carries zero stranded risk through the change |
The deepest rate, three year standard Reserved Instances, looks best until the rebuild, then strands roughly 29,000 USD a month of commitment locked to families the workload has left. The flexible three year Compute Savings Plan is the strongest choice here: it keeps most of the discount and the discount simply follows the workload onto Graviton and Fargate. The one year plan is the safe floor for the part of the estate the team is least sure about.
A simple decision rule
Split the estate by confidence, not by size. Put genuinely steady, long lived workloads on three year terms and accept the tighter scope where the family is settled. Put anything you might rearchitect, migrate, or retire on a one year term or a flexible Compute Savings Plan, where the discount moves with the workload. Layer the two so the committed floor is always capacity you are certain to use, and let the uncertain top ride on shorter, flexible coverage. That is how you capture most of the discount without buying a lock you cannot honour.
Your next step
The instrument you choose should fall out of a forecast, not a rate card. Read the full method in the AWS cost optimization guide, then compare terms directly in one year versus three year commitments on AWS and frame the risk with the risk adjusted approach to AWS commitments. To turn the plan into verified savings, our AWS cost optimization service models your coverage against a forecast you can defend, and you can request a free trial with zero provider commissions.
Put a defensible number on your cloud spend.
No provider in the room, no published price list. Tell us your footprint and we will scope the savings against your billing data — we reduce your cloud spend or we reimburse our service fee.
The Cloud Spend Navigator: what changed in cloud pricing, commitments, and FinOps — no vendor spin.