AWS tells you to stop buying Reserved Instances. The EC2 user guide opens with "We recommend Savings Plans over Reserved Instances." Two things in AWS's own documentation survive that recommendation: a zonal Reserved Instance reserves physical capacity in an Availability Zone, and a Standard Reserved Instance can be sold to someone else. No Savings Plan does either. Everything else in the Reserved Instances vs Savings Plans comparison favours the Savings Plan.
What is the difference between Reserved Instances and Savings Plans?
A Reserved Instance commits you to a configuration. A Savings Plan commits you to an amount of money per hour. The Savings Plans user guide defines the commitment as "using a specified amount of compute power (measured per hour) for a one or three year period", while the EC2 guide on applying Reserved Instance discounts describes a Reserved Instance as "not physical instances, but rather a billing discount that is applied to the running On-Demand Instances in your account".
AWS publishes most of this comparison itself, across four pages nobody reads together. Consolidated, with the quota row from the Capacity Reservations comparison table and the exchange and resale rows from the Reserved Instance offering classes page:
| Standard RI | Convertible RI | EC2 Instance Savings Plan | Compute Savings Plan | |
|---|---|---|---|---|
| You commit to | An instance configuration | An instance configuration | $/hour | $/hour |
| Headline saving | Up to 72% | Up to 66% | Up to 72% | Up to 66% |
| Covers Fargate and Lambda | No | No | No | Yes |
| Crosses Regions | No | No | No | Yes |
| Crosses instance families | No | By manual exchange | No | Yes |
| Reserves capacity | Zonal scope only | Zonal scope only | No | No |
| Exchangeable | No | Yes | No | No |
| Resellable | Yes | No | No | No |
| Applied to usage | First | Second | Third | |
| Purchase quota | 20 per Region or per AZ, per month | "No limit." | ||
The savings columns come from the AWS comparison of Compute Savings Plans and Reserved Instances, which prints them as a matched pair: Compute Savings Plans "provide savings up to 66% off On-Demand, similar to Convertible RIs", and EC2 Instance Savings Plans "offer savings up to 72% off of On-Demand, similar to Standard RIs". The discount is the same. The reach is not. Only the Compute plan crosses Regions, and only the Compute plan reaches past EC2: "Compute Savings Plans automatically reduce your cost on EC2 instance usage, Fargate, and Lambda."
Does a zonal Reserved Instance reserve capacity that a Savings Plan cannot?
Yes, and this is the first of the two gaps. The page on regional and zonal Reserved Instance scope puts it in a table: "A zonal Reserved Instance reserves capacity in the specified Availability Zone", while "A regional Reserved Instance does not reserve capacity." The Savings Plans FAQ is asked the question directly and answers "No, Savings Plans does not provide a capacity reservation."
AWS's Capacity Reservations page draws the same line in a four-column table. Capacity Reservations and zonal Reserved Instances share one merged cell, "Capacity reserved in a specific Availability Zone." Regional Reserved Instances and Savings Plans share the other: "No capacity reserved." The price of that guarantee is every kind of flexibility at once. Zonal Reserved Instances get no instance size flexibility, because AWS restricts it: "Instance size flexibility is only supported for Regional Reserved Instances." They get no Availability Zone flexibility either, since "the Reserved Instance discount applies to instance usage in the specified Availability Zone only". They are also rationed. The Reserved Instance quotas page allows 20 new zonal reservations "per Availability Zone per month", so "in a Region with three Availability Zones, the default quota is 80 new Reserved Instances per month". Savings Plans have no equivalent ceiling.
One limitation is documented and almost never repeated. A zonal Reserved Instance's discount crosses account boundaries but its capacity does not. The applying-discounts page says "Zonal Reserved Instances reserve capacity only for the owning account and cannot be shared with other AWS accounts", yet its own Scenario 4 has a zonal Reserved Instance in one linked account paying for usage in another. The half you bought it for stays put; the half you did not travels. The rest of that sharing behaviour is in what consolidated billing actually saves.
If capacity assurance is the real requirement, AWS points at a third product instead: "you can allocate On-Demand Capacity Reservation (ODCR) for your needs and your Savings Plans will apply." That is the only route to a capacity guarantee and Region-wide discount flexibility at once, and it is priced in what an unused EC2 Capacity Reservation costs.
Can you sell a Savings Plan the way you sell a Reserved Instance?
No. The Reserved Instance Marketplace exists to facilitate "the sale of unused Standard Reserved Instances from AWS customers and third-party sellers". There is no Savings Plans marketplace, and the Savings Plans comparison page states flatly that Savings Plans "can't be cancelled during the term".
The exit is narrower than it sounds, and it inverts what people assume: the flexible reservation is the one you cannot sell.
| Reserved Instance Marketplace rule | What it means |
|---|---|
| "Only Amazon EC2 Standard regional and zonal Reserved Instances can be sold in the Reserved Instance Marketplace." | Convertible Reserved Instances "can't be sold in the Reserved Instance Marketplace". Exchange is their only exit. |
| "AWS charges a service fee of 12 percent of the total upfront price of each Standard Reserved Instance you sell in the Reserved Instance Marketplace." | A $4,000 listing nets $3,520. |
| "You can sell up to $50,000 in Reserved Instances" | Lifetime, not annual. After the 12% fee an account can never net more than $44,000. |
| "You can sell up to 5,000 Reserved Instances" | Also lifetime. Below $10.00 average upfront, this cap binds first. |
| "There must be at least one month remaining in the term of the Standard Reserved Instance." | The last month of a bad commitment is not recoverable. |
| "the bank you specify must have a US address" | A non-US business cannot use this exit. AWS India accounts are excluded by name. |
| "Only the AWS account root user can register an account as a seller." | Not delegable to an IAM role. |
AWS states those as lifetime limits that "can't be increased", so the marketplace is an escape hatch for a mistake, not a strategy. It is also explicit about what you give up: "After the sale, you've given up the capacity reservation and the discounted recurring fees."
Savings Plans do have one exit, and it contradicts the user guide. The FAQ says: "Any Savings Plans that has an hourly commitment of $100 or less, purchased in the past 7 days and in the same calendar month can be returned." The comparison page's note says a Savings Plan "can't be cancelled during the term". Both are current AWS pages. Read together, a Savings Plan is cancellable for about a week and then never. And $100 an hour is not a small plan: over a one-year term, which AWS defines as 31,536,000 seconds and so 8,760 hours, that is $876,000 of commitment inside the return window.
What can a Convertible Reserved Instance be exchanged for?
Another Convertible Reserved Instance, in the same Region, of equal or higher value. The exchange page allows "a different configuration, including instance family, operating system, and tenancy", with "no limits to how many times you perform an exchange, as long as the new Convertible Reserved Instance is of an equal or higher value than the Convertible Reserved Instances that you are exchanging". Region is fixed: "You can't exchange a Convertible Reserved Instance for a Convertible Reserved Instance in a different Region."
That equal-or-higher rule is a one-way ratchet. A Compute Savings Plan already applies across families, sizes, operating systems, tenancies and Regions with no transaction at all; a Convertible Reserved Instance reaches the same place by a manual exchange that can only ever increase your commitment. AWS's own worked example shows the cost of that without naming it. It exchanges "1 x $35-list value Convertible Reserved Instance" for units worth $10 each, prints "$35/$10 = 3.5", then notes "It's not possible to purchase half reservations; therefore you must purchase an additional Convertible Reserved Instance to cover the remainder". Four units at $10 is $40 of list value for $35 surrendered: a 14.29% forced overshoot, on which "you pay the true-up cost for the fourth reservation".
The penalty tracks the remainder, not the size of the unit, which means a bigger target instance can be the cheaper exchange. Same $35 reservation, five target unit values:
| Target unit list value | $35 / unit | Units you must take | List value received | Forced overshoot |
|---|---|---|---|---|
| $8 | 4.375 | 5 | $40 | 14.29% |
| $10 (AWS's example) | 3.5 | 4 | $40 | 14.29% |
| $12 | 2.917 | 3 | $36 | 2.86% |
| $18 | 1.944 | 2 | $36 | 2.86% |
| $20 | 1.75 | 2 | $40 | 14.29% |
That is arithmetic on AWS's example, not a table AWS publishes, and it is the number to check before an exchange: a $12 target costs a fifth of what a $10 target costs for the same reservation.
Are Reserved Instances or Savings Plans applied first to EC2 usage?
Reserved Instances, always. The page on how Savings Plans apply to usage opens its billing section with one sentence: "Savings Plans apply to your usage after the Amazon EC2 Reserved Instances (RI) are applied." The comparison page says the same thing from the other side: "Savings Plans doesn't apply to spot usage or usage covered by RIs." So does the consolidated billing guide, which describes AWS "aggregating all usage that's not covered by Reserved Instances, and applying the Savings Plans rates starting with the highest discount". Among plans the order continues: "EC2 Instance Savings Plans are applied before Compute Savings Plans because Compute Savings Plans have broader applicability."
Three AWS pages, one order, and it settles a question people get backwards. A Savings Plan cannot strand a Reserved Instance: the reservation is served first and keeps its discount. The plan is what the order puts at risk. What low utilization and low coverage each mean is worked through in Savings Plans coverage vs utilization.
Will a Savings Plan bought on top of a Reserved Instance go unused?
It will, for exactly the usage the Reserved Instance already covers. What follows is hypothetical, on AWS's own rates from the applying page, which states "The rates in these examples are illustrative only."
Suppose you run four r5.4xlarge Linux instances and already hold
four matching regional Reserved Instances. AWS's example table puts the
On-Demand rate at $1.00 an hour and the Compute Savings Plans rate at $0.70.
Sizing a plan on those four instances gives a $2.80 an hour commitment. Because
Reserved Instances apply first, all four are covered before the plan is
consulted, and the plan has nothing left to apply to. AWS's Scenario 4 shows
the mechanism on the same instance type: "First, the RI covers two of the
r5.4xlarge instances." The commitment does not wait for it, because "Each
hour's commitment can only be used within that hour and cannot be carried
over":
| Hypothetical stranded commitment | Hours | Cost of the unused commitment |
|---|---|---|
| One hour | 1 | $2.80 |
| One month, at AWS's 24 x 30 convention | 720 | $2,016.00 |
| One-year term (31,536,000 seconds) | 8,760 | $24,528.00 |
| Three-year term (94,608,000 seconds) | 26,280 | $73,584.00 |
The shape matters more than the amount: a plan sized against usage a reservation already covers is not a smaller discount, it is a pure loss, and it lasts the whole term. Size the commitment against usage that is not already reserved, then read the result out of the effective-cost columns described in amortized vs unblended cost.
An idle Reserved Instance is no better, to be clear. The Reserved Instance billing page says they "are billed for every clock-hour during the term that you select, regardless of whether an instance is running". Both instruments are use-it-or-lose-it by the hour. Only one of them can be sold afterwards.
Do Reserved Instances or Savings Plans save more on EC2?
Neither, and two AWS pages disagree about it. The EC2 Reserved Instances overview says Savings Plans "offer lower prices (up to 72% off On-Demand pricing), just like Reserved Instances", which reads as though every Savings Plan reaches 72%. The Savings Plans comparison table on the other page gives 72% to EC2 Instance Savings Plans and 66% to Compute Savings Plans, the same split as Standard against Convertible Reserved Instances. The FAQ is the one that states it cleanly: Savings Plans "offer the same savings as RIs, but with additional flexibility."
Same savings with more flexibility is the accurate version, which means the decision is not about the discount rate. It is about what happens when you are wrong. A Compute Savings Plan absorbs a change of family, size, operating system, tenancy, Region, or a move to Fargate or Lambda without any transaction. A Reserved Instance requires a modification, an exchange, a sale, or nothing.
One asymmetry does favour large Standard buyers. Reserved Instance discount pricing tiers still exist: "To qualify for a discount, the list value of your Reserved Instances in the Region must be $500,000 USD or more", and they are "currently not applicable to Convertible Reserved Instance purchases". The Savings Plans types page states the opposite property: "Savings Plans prices do not change based on the amount of hourly commitment."
When should you still buy a Reserved Instance in 2026?
Three cases, and they are narrow. When you need a capacity guarantee in a named Availability Zone and want the discount attached to it in one purchase: buy a zonal Standard Reserved Instance. When the commitment crosses the $500,000 regional list value that triggers a pricing tier no Savings Plan can match. And when the ability to list it on the marketplace is worth more than the flexibility you give up, at a 12% fee and a $50,000 lifetime cap.
Everything else points the other way, including AWS. The FAQ tells existing holders "as your RIs expire we encourage you to sign up for Savings Plans as they offer the same savings as RIs, but with additional flexibility", and the EC2 guide adds that "Reserved Instances do not renew automatically". Expiry is the cheapest moment to switch, because nothing has to be exchanged or sold.
Either way the commitment is irreversible in the direction that matters: "After you purchase a Reserved Instance, you cannot cancel your purchase", and a Savings Plan's terms "can't be changed after purchase". Size against the floor of your usage rather than its average, subtract what is already reserved, and verify the outcome in the Cost and Usage Report rather than on a recommendation screen. Which report answers which question is covered in Cost Explorer vs the Cost and Usage Report.