Commitment discounts are the largest single lever in cloud cost, and the least discussed honestly. The published advice assumes a stable organisation committing to its own baseline.
Across a portfolio, that assumption fails in a specific way: the entity signing the commitment is not the entity that decides whether the workload still exists in eighteen months.
The risk you are actually taking
A commitment is a bet that a level of usage will persist. In a single company, you control both sides. In a portfolio, an operating company can be sold, can migrate, can pivot, or can simply decide to run something differently, and none of those decisions route through whoever bought the commitment.
So the question is not what the total baseline is. It is which part of that baseline is stable for reasons nobody is likely to revisit.
Commit to the floor, not the average
The approach that held up was to commit only to consumption that would survive almost any individual decision. In practice that meant shared platform infrastructure, long-lived data storage, and the compute underneath services with contractual obligations attached.
Commit to what will still be running if the most volatile tenant disappears tomorrow.
That is deliberately conservative and it leaves discount on the table. It also means a commitment never became a stranded cost that group finance was carrying for a company that had changed direction. Given the difference in consequences, we would take that trade again.
Prefer flexible instruments where they exist
Both major providers offer commitments that apply to spend rather than to a specific machine shape. Those are worth a meaningful amount of discount to give up, because they survive a tenant changing instance family, region or architecture without you renegotiating anything.
In an estate you fully control, locking to specific shapes for the extra few percent can be right. Where the workload owner may re-architect without telling you, flexibility is not a luxury, it is the thing that keeps the commitment useful.
Ladder the terms
Rather than committing everything at once for three years, we laddered: a base layer on longer terms for the genuinely permanent, and shorter overlapping commitments above it that expire at different times.
This gives you a decision point every few months rather than one large cliff, and it means a change in the portfolio only ever affects the next tranche rather than the whole position.
Decide who carries it, in writing
The question that causes trouble later is who bears the loss if committed capacity goes unused. Group buys the commitment for the portfolio benefit, but the unused portion originated with one company changing its plans.
This has to be agreed before anything is signed, and it is a commercial conversation rather than a technical one. Teams that leave it implicit end up litigating it retrospectively, at exactly the moment when someone is already unhappy about a decision that was made for good reasons.
And do it last
Commitment is the final step of a cost programme. Buying against a baseline you are about to reduce locks in the waste you were about to remove, and unlike every other mistake in this discipline, it is one you cannot simply reverse next month.
