AWS Reserved Instances: discounts, types & the Savings Plans question.

01 Key findings

1. An RI is a billing discount, not a reserved server. You commit to an instance configuration for 1 or 3 years and AWS applies a lower rate to matching on-demand usage. The commitment is financial — an unused RI still bills.

2. Standard RIs reach 72%; Convertible RIs trade roughly 10 points for flexibility. A 3-year all-upfront Standard RI tops out near 72% off; the equivalent Convertible sits near 62% in exchange for the right to change instance family.

3. Payment option is a cash-flow choice, not a savings choice. The gap between all-upfront and no-upfront is usually only 2–4 points, rarely enough to justify tying up capital.

4. Savings Plans now match RI depth with more flexibility. Compute Savings Plans apply automatically across families, sizes, regions, and even Fargate and Lambda — the better default for most workloads.

5. Utilisation decides RI value. The discount only lands if running usage matches the reservation. Cover the stable 60–80% of demand with commitments and leave the variable top layer on-demand.

02 How Reserved Instances work

A Reserved Instance is not a reserved server; it is a billing discount applied to matching on-demand usage in exchange for a 1 or 3 year commitment to a specific instance configuration. You commit to an instance family, size, region, and tenancy, and AWS applies the RI's lower rate to any running instance that matches. The discount scales with term length and payment option: a 3-year all-upfront Standard RI reaches the maximum 72% off on-demand, while a 1-year no-upfront Convertible RI sits closer to 30%. Because the commitment is financial rather than capacity, an unmatched RI still bills at its committed rate for nothing.

Regional Linux RIs also carry instance size flexibility: the discount applies across sizes within the same family and generation in proportion to a normalisation factor, so one large reservation can cover two mediums or four smalls. That flexibility narrows the practical gap between Standard and Convertible RIs and lets a buyer commit to Standard without fearing a resize will strand the discount.

03 Standard versus Convertible

The two RI classes trade discount depth against flexibility. Standard gives the deepest rate but locks the instance family; Convertible sacrifices roughly 10 points for the right to exchange into a different family as your architecture changes.

Standard RIs keep an escape hatch: unwanted positions can be sold on the AWS Reserved Instance Marketplace to recover value from an over-commitment. Convertible RIs cannot be resold — only exchanged — so an over-bought Convertible position can be reshaped but not unwound for cash. The choice is a bet on how stable your instance selections are over the commitment term.

04 Discount by term

Two levers set the headline rate: term length and RI class. Three-year commitments roughly double the discount of one-year, and Standard leads Convertible at every term. Approximate maximum discount off on-demand, all-upfront:

05 Payment options

Each RI can be bought under one of three payment structures. The discount difference between them is small; the cash-flow difference is not.

The difference between all-upfront and no-upfront is usually only 2 to 4 percentage points, so the payment choice is more a cash-flow decision than a savings decision. For most enterprises the small extra discount of all-upfront does not justify tying up the capital, and partial or no-upfront preserves flexibility at almost the same rate. The accounting treatment also differs: all-upfront RIs are a prepaid asset amortised across the term, while no-upfront is recognised as incurred — align the option to how finance prefers to treat the spend.

06 Reserved Instances versus Savings Plans

Savings Plans deliver the same compute discount depth as RIs but commit to a dollar-per-hour spend rather than a specific instance configuration. Compute Savings Plans apply automatically across instance families, sizes, regions, and even Fargate and Lambda, which removes the burden of matching RIs to usage. For most workloads this flexibility makes Savings Plans the better default, and AWS has steered new commitments toward them.

Reserved Instances retain two real advantages. Zonal RIs, scoped to a specific Availability Zone, reserve capacity in that zone — a regional RI reserves nothing and is purely a billing discount, and Savings Plans cannot reserve capacity at all. And Standard RIs can be resold on the Marketplace to recover value from an over-commitment. The full comparison is in our Savings Plans versus Reserved Instances analysis and the Savings Plans guide.

07 The utilisation trap

The discount only materialises if the RI is matched by running usage, so utilisation is where RI value is won or lost. An RI that sits unmatched — because the workload it covered was retired or resized — is pure waste, billing at the committed rate for nothing. The lock-in compounds it: a Standard RI cannot change family, and only the Marketplace lets you exit for cash. Mature programmes monitor utilisation continuously, exchange or resell positions that drift out of use, and size commitments to the stable baseline rather than the peak.

A worked example: a fleet with a stable 24/7 baseline of $400,000 a month, a predictable daytime layer of $150,000, and a spiky experimental layer of $80,000. Cover the baseline with a 3-year Savings Plan or Standard RIs at the deepest rate, cover the predictable layer with a renewable 1-year commitment, and leave the spiky layer on-demand or Spot. Zone-critical components take zonal Standard RIs for the capacity guarantee. Run against a self-managed single-instrument mix, this layered structure improves coverage efficiency by a median of about 19 percent.

08 Purchase framework

Four questions decide which instrument fits each slice of demand. Answer them per workload, not per fleet.

09 Our recommendation