Cloud providers offer substantial discounts in exchange for committing to a level of usage over one or three years. The discount is real and the trap is equally real: a commitment made on optimistic growth assumptions becomes a bill for capacity you are not using.
Optimise before committing, not after. Locking in a discount on over-provisioned instances means paying a reduced rate for waste for the next three years, and the right-sizing you do afterwards delivers nothing because the commitment continues regardless. Right-size first, then commit to the reduced baseline.
Commit to the floor, not the average. Almost every workload has a level below which usage never falls. That floor is the safe commitment. The variable portion above it is what on-demand and spot pricing exist for, and paying the premium on the top slice is far cheaper than committing to capacity you use half the year.
Prefer flexible commitment types where available. Options that apply across instance families and regions cost slightly less discount and survive an architecture change; instance-specific reservations give the deepest discount and become worthless the moment you migrate to a different instance type, which over three years is likely.
One year is usually the right term for a growing company. The additional discount for three years is meaningful, but so is the probability that your architecture in 2029 looks nothing like today's. Three-year terms suit stable, well-understood workloads.
Set a review date and put it in a calendar. Commitments expire quietly, renew on default terms, or continue billing for services that have been decommissioned. A quarterly check on utilisation against commitment takes twenty minutes and is the difference between the discount working for you and against you.
