Cloud providers offer meaningful discounts for usage committed in advance. The discount is payment for accepting a forecasting risk that the provider would otherwise carry.
The provider is buying predictability
Data centre capacity is built years ahead, and the hardware is purchased regardless of whether anyone uses it.
A customer who commits to a level of usage lets the provider plan with confidence, which is worth a discount that can be a large fraction of the on-demand rate.
The provider is therefore not selling cheaper computing. It is buying certainty, and the discount is the price it pays for that.
The customer takes on forecast risk
A commitment is owed whether or not it is used. If a workload is retired, migrated or optimised away, the obligation continues.
This penalises exactly the improvements teams are otherwise encouraged to make, since reducing consumption below the committed level saves nothing.
Committing to the stable baseline rather than to peak or expected usage keeps the discount while leaving room for the workload to shrink.
Flexible commitments cost more but bind less
Commitments expressed as an amount of spending per hour, rather than as specific machine types, can be applied wherever usage occurs.
They carry a smaller discount than instance-specific reservations, and the difference is the price of retaining freedom to change architecture.
For organisations that expect to modernise their workloads, the flexible form is often cheaper in practice despite the lower headline rate.
Terms interact with the technology cycle
Longer commitments offer larger discounts, and hardware generations arrive faster than the longest terms run.
A three-year commitment to a specific instance family can leave an organisation paying for older hardware while newer types deliver more performance per unit of cost.
Where a provider allows commitments to move to newer generations, this risk largely disappears, which makes that provision worth checking before signing.
Coverage is managed as a portfolio
Mature practice treats commitments like a laddered set of obligations, with staggered expiry dates so that only a portion is renegotiated at any time.
The target is usually to cover the predictable base load while leaving variable demand on flexible pricing.
Reviewing coverage regularly against actual usage catches both under-commitment, which forfeits discount, and over-commitment, which is paid for whether used or not.
Secondary markets exist for some reservation types, allowing an unwanted commitment to be sold on, though pricing there depends on demand for the specific configuration.