Commitments without regret: Savings Plans vs RIs vs Spot
The Lizrd team · · 2 min read
Commitment discounts are the most tempting lever in cloud cost work: sign up, pay less per hour, done. They’re also where teams do the most quiet damage — because a commitment locks in whatever you commit to, including the waste. Buy a three-year Reserved Instance for an over-provisioned fleet and you’ve just made the waste non-negotiable for three years.
The tools aren’t the problem. Committing before you’ve measured is.
Know what each instrument actually buys
They’re not interchangeable, and treating them as one “discount” is how regret happens:
- Reserved Instances — commit to a specific instance family/region for 1–3 years. Deepest discount, least flexible. Right for a stable, well-understood baseline you’re confident won’t change shape.
- Savings Plans — commit to a dollars-per-hour spend level, not a machine. More flexible across families and, for the compute plan, across EC2/Fargate/Lambda. Right for a baseline whose amount is stable even if the mix moves.
- Spot — no commitment at all; deep discount in exchange for interruption. Right for the elastic, fault-tolerant layer on top.
Rightsize first, commit second — always
This is the rule that prevents almost all commitment regret: never commit to a resource you haven’t rightsized. A commitment on an oversized instance compounds the mistake — you’re now locked into paying a discounted rate for capacity you never needed.
The correct order is boring and non-negotiable: measure utilization, rightsize to real demand, let the new baseline settle, then commit to what’s left. Do it backwards and every rightsizing you’d want to do later is blocked by a contract.
Layer coverage, don’t max it
Coverage isn’t a number to maximize; 100% commitment coverage means you’ve bet that nothing will ever shrink. Think in layers against your usage curve:
- The flat baseline — the capacity that’s on 24/7, month after month — carries RIs or Savings Plans.
- The predictable daytime bulge — a Savings Plan sized to the trough of that bulge, not its peak.
- The spiky top — Spot and on-demand, uncommitted, free to scale to zero.
Commit to the part of the curve you’re confident is permanent, and leave the rest flexible. Under-committing costs you a little discount; over-committing costs you the ability to optimize.
The whole strategy depends on seeing the baseline
Every decision above rests on one thing: an honest, continuous view of what your steady-state usage actually is — after rightsizing, broken down by family and account, projected forward. Most teams commit off a single month’s snapshot and a spreadsheet, which is how you end up locked into last quarter’s mistakes.
That’s the visibility Lizrd is built to give you: it reads utilization and billing together, shows the rightsized baseline before you commit, and flags where existing commitments are covering waste instead of demand. Get the picture right and commitments become what they should be — a discount on capacity you were always going to use.