Are Savings Plans always better than Reserved Instances?

Not always. Compute Savings Plans give the most flexibility (any instance family, region, OS, tenancy, and Fargate/Lambda) at a slightly lower max discount. Standard RIs can discount a specific instance family slightly more and can be sold on the Reserved Instance Marketplace, but they are rigid. For most teams a Compute Savings Plan is the safer default; RIs make sense for a stable, well-understood baseline you won't re-architect.

Should I buy 1-year or 3-year commitments?

1-year (No Upfront) is the low-risk default — meaningful discount, limited downside if usage changes. 3-year gives the deepest discount but only fits a baseline you are confident will still exist and run the same way in three years. Mixing terms across a coverage target is common: 3-year for the rock-solid base, 1-year for the layer above it.

Does migrating to Graviton affect my commitments?

Yes, and it is why flexibility matters. A Compute Savings Plan keeps applying when you move a workload from x86 to Graviton (Arm) instances, so you capture both the commitment discount and the Graviton price/performance gain. A Standard RI locked to an x86 family would be stranded by that migration.

AWS · Rate optimization · Updated June 2026

AWS Savings Plans vs Reserved Instances: Which to Buy in 2026

By the CloudFinOpsKit team — building the AWS tool's commitment & rate-optimization checks. 8 min read.

Paying on-demand for a workload that runs 24×7 is the single most expensive habit in AWS — effectively a 30–70% premium over the committed rate. Commitments fix that, but the menu confuses people: Compute Savings Plans, EC2 Instance Savings Plans, Standard RIs, Convertible RIs, three terms, three payment options. Here's the decision, distilled.

The three instruments, side by side

Instrument	Covers	Max discount*	Flexibility
Compute Savings Plan	EC2 (any family/region/OS/tenancy) + Fargate + Lambda	~66%	Highest — auto-applies as you change family, size, region, even to Graviton
EC2 Instance Savings Plan	A chosen instance family in a chosen region (any size/OS)	~72%	Medium — locked to family + region, flexes on size/OS
Standard RI	Specific family/region (size-flexible within family)	~72%	Low — can't change family; sellable on the RI Marketplace
Convertible RI	Exchangeable for other families	~54%	Medium — exchange (not cancel)

*Indicative 3-year, All Upfront. Actual discount varies by family/region/term/payment; always verify against the live Cost Explorer recommendation.

The real trade-off: discount vs flexibility

The extra few points an EC2 Instance SP or Standard RI gives you over a Compute SP come at the cost of lock-in. And lock-in is exactly what bites: the most common commitment write-off we see is a 3-year Standard RI bought against a workload that got re-architected — or moved to Graviton — two months later. The commitment stranded, the discount wasted.

For most teams the Compute Savings Plan is the right default: a couple of points less discount, but it keeps applying through size changes, family changes, region moves, and the migration everyone's doing right now —

The Graviton angle (don't skip this)

Moving x86 workloads to Graviton (Arm) typically improves price/performance ~20–40%. A Compute Savings Plan follows you onto Graviton, so you stack both savings. A family-locked RI does not. Sequence it right: migrate to Graviton first where you can, then commit — committing to the architecture you'll actually run avoids buying a discount for instances you're about to retire.

How much coverage? A risk-bounded strategy

Don't cover 100% — usage moves, and uncovered commitment is wasted money. Cover your stable baseline, leave the spiky top on-demand (or Spot). A practical model:

Find the floor. In Cost Explorer, look at the last 3 months of on-demand compute — the trough is your safe baseline.
Target 60–80% coverage of that baseline, with ≥95% utilization. Below 95% utilization you're paying for commitment you're not using.
Layer terms: 3-year for the rock-solid base, 1-year No Upfront for the layer above it where you're less certain.
Stress-test the downside: model usage falling 10/20/30% and confirm worst-case stranded commitment stays within your risk tolerance before you buy.

Payment options

All Upfront > Partial > No Upfront on discount, but No Upfront keeps cash free and still captures most of the saving. If cash flow matters more than the last few percent, No Upfront is a perfectly good default.

See your commitment opportunity from your own data. The CloudFinOpsKit AWS Tool reads the authoritative AWS Cost Explorer Savings Plans & RI purchase recommendations, surfaces under-utilized existing commitments, and shows the realized amortization on your bill — read-only, no agents, alongside 70+ other Well-Architected checks. It never invents a number; the recommendations come from AWS's own engine.

New to the cadence? The free AWS Cost Review Checklist includes the monthly commitment-coverage review.

FAQ

Can I cancel a Savings Plan or RI?

No — commitments can't be cancelled mid-term. Standard RIs can be sold on the RI Marketplace; Convertible RIs can be exchanged; Savings Plans can't be undone. That's why right-sizing coverage and stress-testing the downside matter.

Do Savings Plans cover Fargate and Lambda?

Compute Savings Plans do — EC2, Fargate, and Lambda. EC2 Instance SPs and RIs cover EC2 only.

What about RDS, Redshift, ElastiCache, OpenSearch?

Those have their own Reserved-style commitments (Reserved Nodes / Reserved Instances per service); Savings Plans don't cover them. Review them separately.