AWS Reserved Instances vs Savings Plans in 2026: Which Commitment Discount Should You Actually Buy?

TL;DR

• Reserved Instances lock in a specific instance family (and optionally a specific region/AZ) for 1 or 3 years, for the deepest possible discount — up to ~72% off on-demand.
• Savings Plans commit you to a dollar-per-hour spend level instead, and apply automatically across whatever you run — more flexible, with a slightly smaller maximum discount for the most flexible tier (Compute Savings Plans).
• Default to Compute Savings Plans unless you’re certain your instance family and region won’t change for the life of the commitment — most teams’ usage shifts more than they expect over 1-3 years.
• Commitments are only as good as the usage data behind them. Model them against your real baseline, not a forecast — sized wrong, a “savings” tool becomes a cost you’re stuck with for a year.

The question behind the question

Nobody asks “Reserved Instances or Savings Plans” until they’ve already decided to commit to something — usually because finance has asked why the AWS bill keeps climbing when on-demand pricing alone should be predictable. By that point the real question isn’t which product AWS named more cleverly. It’s: how much do we actually know about our own usage, and how much are we willing to bet on that knowledge holding steady for a year or three?

That’s the right question, because both products solve the same underlying problem — trade commitment for discount — but they make you commit to different things. Get that part wrong and a savings tool turns into the exact kind of locked-in waste it was supposed to prevent.

Side-by-side comparison

What you’re actually committing to

Reserved Instances (RIs) commit you to a specific EC2 instance family, in a specific region, for 1 or 3 years. Standard RIs give the deepest discount (up to ~72% off on-demand) but can’t be changed to a different instance family — though they can be resold on the AWS RI Marketplace if your needs change. Convertible RIs trade some discount depth (up to ~54%) for the ability to exchange into a different instance family, OS, or tenancy mid-term.

Savings Plans commit you to a dollar-per-hour spend level instead of a specific instance type. The discount applies automatically to whatever compute you run, as long as you’re spending at least that much per hour. EC2 Instance Savings Plans are scoped to one instance family in one region (matching Standard RI’s discount depth, up to ~72%) but stay flexible on instance size, OS, and tenancy within that family. Compute Savings Plans are scoped to nothing at all — the commitment applies across any EC2 instance family, region, and tenancy, plus Fargate and Lambda — at a slightly lower maximum discount (up to ~66%) in exchange for that flexibility.

When this matters: the tighter the scope you commit to, the deeper the discount — and the more painful it is if your usage doesn’t match the shape you locked in.

Coverage

Reserved Instances exist per-service — EC2 RIs, and separate reservation products for RDS, ElastiCache, Redshift, and OpenSearch, each bought and managed independently.

Savings Plans (Compute Savings Plans specifically) cover EC2, Fargate, and Lambda under one commitment. They don’t extend to RDS or other managed services — those still need their own Reserved Instance if you want a commitment discount there.

When this matters: if your spend is concentrated in EC2 plus serverless (Fargate/Lambda), one Compute Savings Plan can cover most of your commitment-eligible spend in a single purchase. If a meaningful chunk of your bill is RDS or another reserved-eligible managed service, you’ll be buying that commitment separately either way.

Payment options and term length

Both products offer the same three payment structures — No Upfront, Partial Upfront, and All Upfront — with more upfront cash buying a marginally deeper discount. Both also offer the same term lengths: 1 year or 3 years.

When this matters: payment structure is mostly a cash-flow decision, not a technical one. Term length is the bigger bet — see the FAQ below.

What happens if your usage changes

Standard RIs can be resold on the AWS RI Marketplace if you no longer need them — not instant, and not guaranteed to recover full value, but it’s an exit. Convertible RIs can be exchanged for a different configuration directly through AWS.

Savings Plans have no resale or exchange mechanism. If your usage drops below your committed dollar-per-hour level, you simply stop getting full value from the commitment for the remainder of the term — you keep paying for it either way. This is the trade-off for their flexibility on what you run: there’s no flexibility on whether you keep paying.

Decision matrix

Criterion	Choose Reserved Instances	Choose Savings Plans
Certainty about instance family	High — you know exactly what you’ll run	Low to medium — mix may shift
Coverage needed	EC2 only, or a specific reserved-eligible service (RDS, etc.)	EC2 + Fargate + Lambda under one commitment
Priority	Maximum discount depth (Standard RI)	Flexibility, simpler management
Exit if usage changes	Standard RI: resell on RI Marketplace. Convertible RI: exchange directly	None — you keep paying for the remainder of the term
Management overhead	Higher — track per-instance-family utilization	Lower — one commitment covers a spend level
Best first commitment for most teams	Only if usage is already stable and well-understood	Yes — Compute Savings Plans, 1-year, no upfront

Real-world example

We covered the sequencing question in detail in FinOps in 90 Days, but the specific commitment decision is worth pulling out on its own. In that engagement, a B2B SaaS startup had grown its AWS bill from $8K to $26K a month with no tagging, no rightsizing history, and no clear picture of what was driving the increase. The first month went entirely to visibility and cleanup — killing an orphaned data pipeline, terminating idle instances, rightsizing the rest. Only in month 2, once there was real post-cleanup usage data to model against, did a commitment purchase happen: a one-year Compute Savings Plan covering the steady-state baseline, which saved an additional $1.8K a month versus on-demand — on top of the cleanup and rightsizing savings already in hand.

That ordering matters more than the product choice. A commitment placed in month 1, against pre-cleanup usage, would have locked in a baseline that was about to shrink — turning a savings mechanism into a cost the team would have carried for a year regardless.

Trade-offs and what we’d avoid

• Don’t commit before you optimize. Modeling a commitment against usage that still has obvious waste in it just discounts the waste — clean up first, commit second.
• Don’t default to 3-year terms to chase the bigger discount. Unless you have a full year of stable, post-optimization data behind you, a 1-year term is the safer first move. See the FAQ above.
• Don’t assume Reserved Instances are automatically cheaper. Standard RIs and EC2 Instance Savings Plans have the same maximum discount depth (~72%) for the same scope of commitment — the real trade-off is flexibility, not price.
• Don’t treat a commitment purchase as a one-time decision. Usage shape drifts as teams ship. Re-model quarterly against real utilization, not the assumptions you made when you first bought the commitment.
• Don’t forget Savings Plans have no exit. If there’s real uncertainty about whether your usage will hold, that lack of an exit ramp is the single biggest reason to under-commit slightly rather than over-commit.

What to do next

1. If you haven’t optimized yet — do that first. Why Kubernetes clusters waste 40% of their budget and How much can you save on your AWS bill? cover the cleanup and rightsizing work that should happen before any commitment purchase.
2. If you’re ready to model a commitment — see Cloud Cost Optimization for how we model commitment shapes against real usage curves rather than forecasts, and re-model them quarterly.
3. If you want the full 90-day sequence — read FinOps in 90 Days for the week-by-week plan this example came from, or book a 30-minute call to talk through your specific bill.

Related reading: How much does 24/7 SRE really cost? — if commitment-based savings are freeing up budget, this breaks down where a lot of teams reinvest it next.