Savings Plans replaced EC2 RIs for most use cases but RDS, ElastiCache, Redshift and OpenSearch RIs remain the primary commit vehicle for the database layer. The current correct portfolio is narrower, deeper and more disciplined than it was five years ago — and the median enterprise still over-buys EC2 RIs while under-deploying RDS RIs.
Reserved Instances (RIs) predate Savings Plans by a decade and have a specific role that SPs have not replaced. RIs still deliver capacity reservations alongside the discount, still cover services that SPs do not reach (RDS, ElastiCache, Redshift, OpenSearch, DynamoDB), and still offer the convertible exchange mechanic that has no SP equivalent. The change is that for plain EC2 compute, Savings Plans almost always win on flexibility for similar discount, and the EC2 RI portfolio should be allowed to roll off rather than refreshed.
The current correct RI portfolio sits in three places. RDS Reserved Instances for the production database layer — there is no equivalent SP, the discounts are 30–55%, and the workload is structurally stable. ElastiCache, Redshift and OpenSearch RIs where applicable. And a thin layer of EC2 Convertible RIs for very specific cases where the capacity reservation matters more than the optionality — typically large regulated workloads that need guaranteed instance availability in a specific AZ.
Standard RIs cannot be modified. Convertible RIs can be exchanged for different instance families, OS, tenancy or scope as long as the new RI has equal or greater value. The discount difference between Standard and Convertible has narrowed in recent years, and the practical optionality of Convertible is usually worth more than the 3–5 point discount gap. The rule for most enterprises: if you are buying RIs at all, buy Convertible unless you have an extremely specific operational reason to lock the instance family.
The post-SP playbook is different. We have seen enterprises hold $4M+ of expired EC2 RIs that should have been Compute SPs.
RDS Reserved Instances are the most consistently under-deployed RI category in our engagement portfolio. The reasons are operational: database teams have less visibility into commit programs than EC2 teams, the choice of engine and instance class is more conservative, and procurement teams often treat RDS as if it were a managed service exempt from commitment optimisation. The discount math is identical to EC2: 30–55% off on-demand for 1- or 3-year commits, with payment options that mirror EC2.
The right RDS RI portfolio mirrors the production database footprint with a thin buffer for growth. Across our portfolio, the median RDS RI coverage is 38% — when it should sit between 75% and 90%. The gap typically represents $200K–$2M per year in foregone discount, depending on RDS footprint size. The work to close the gap is operational, not commercial: identify the stable RDS instances, model the right RI mix, automate the renewal cadence.
RIs you no longer need can be sold on the AWS RI Marketplace at a discount to your remaining commitment value. The discount needed to actually move on the marketplace is usually 18–35% off remaining value, depending on instance family and term remaining. The marketplace is not a primary strategy — it is the cleanup mechanism for the inevitable mismatches. Customers who treat it as a backstop write off the right RIs without panicking. Customers who treat it as a savings strategy lose money buying speculative inventory.
Commitment sizing model, RI vs SP decision matrix, marketplace strategy and the renewal timeline.
The deepest RI portfolios in our engagement portfolio share a single characteristic: a named owner with authority over both the financial commit decision and the engineering instance choice. Where these two decisions sit in different organisations — finance owns the commit, engineering owns the workload — the RI portfolio under-performs by 18–34% even when both teams are operationally competent. The financial team buys what was historically used. The engineering team optimises for current workload without regard to outstanding RI inventory. The two clocks drift apart until the next quarterly review. Closing that gap is less a procurement exercise than a license cost reduction discipline — one owner, one model, one renewal cadence.
For plain EC2, mostly yes — Compute SPs deliver similar discount with more flexibility. For RDS, ElastiCache, Redshift, OpenSearch and DynamoDB where no SP equivalent exists, RIs remain the primary commit vehicle.
Convertible unless you have a specific operational reason to lock the instance family. The 3–5 point discount gap is usually worth less than the optionality to exchange.
75–90% of the stable database footprint, with the remaining buffer covering growth and instance class drift. Median observed coverage sits at 38%, which leaves $200K–$2M of discount on the table for most enterprises.
Yes, on the AWS RI Marketplace. Expect a 18–35% discount to remaining value to actually move inventory. Treat as a cleanup mechanism, not a primary strategy.
RIs apply first, SPs second. The combination produces strange artefacts at the margin. Generally do not stack RIs and SPs on the same workload class — pick one vehicle per workload type.
We have advised on cloud contracts from $2M to $120M annually across AWS, Azure and GCP. The leverage is in the structure, not the rate card.
Weekly compliance intelligence for IT leaders.