How Big Should Your Next Cloud Commitment Be?

A larger cloud commitment saves more at today’s prices but can cost more if prices or usage fall early in its term. Before you approve one, find the fall that would make a smaller commitment the cheaper choice.

Picture next quarter’s rate optimisation report for an illustrative portfolio. Coverage has slipped from 70.0 to 67.3 per cent, the effective savings rate is up from 25.0 to 26.4 per cent, and the monthly bill is 8 per cent higher. Usage has not changed. Your provider raised compute list prices by 10 per cent across the board. The savings rate improved anyway. Neither coverage nor the savings rate tells you what another price change would do to the bill, or whether the commitment awaiting your signature is the right size.

A commitment fixes what you owe

A rate-optimisation commitment fixes a payment obligation and may or may not protect the unit price. The FinOps Foundation’s playbook calls the effective savings rate the return on investment for discount instruments: savings divided by what the same usage would cost on demand. Because it is measured against the provider’s price list, a rise widens the gap on every line whose payment stays flat.

Reported to a board without the bill beside it, the savings rate can make a price rise look like an improvement. Coverage does not show the price exposure a commitment creates. When a floating-rate commitment is sized at or just below the usage within its scope, that exposure turns lopsided: a rise passes through in full, and a cut saves at most what you already spend above the commitment.

Commitments do not all hold the price

AWS says the prices you pay for usage stay the same through a Savings Plan’s term. An Azure savings plan also fixes an hourly spend, but its discounted rates are repriced when Azure changes pay-as-you-go prices.

Azure reserved VM instances lock in a price. Google’s compute flexible CUDs keep the same discount percentage, so the committed spend stays fixed while the price you pay follows list. Google’s Flexible Savings Plans, a separate commitment for generative AI and related products, work on a monthly window instead of an hourly one.

Recent rises came product by product

AWS is reported to have raised Capacity Block prices for several GPU instance types by about 20 per cent from 1 July 2026, and existing blocks kept the price set when they were bought. Google doubled North American egress through CDN Interconnect and peering from 1 May 2026. It says discounts off list may take the new price, while fixed-price contracts generally hold until renewal. A Microsoft advisory posted on Microsoft Q&A by a partner, with no public release yet, announces rises of 8 to 17 per cent for some VM and Storage services in seven regions from 1 February 2027.

Rises and falls do not net out

The portfolio behind the opening figures has $100,000 a month of commitment-eligible compute at on-demand prices. Line A has a fixed-price commitment at 40 per cent off. Line B has a floating-rate commitment of $21,000 a month at an assumed steady 30 per cent off, scoped to line B’s workloads and enough for all their current usage. Line C has none, and line B’s commitment cannot reach it. Usage holds steady and all prices move together.

A rise costs this portfolio $6,000 a month, split evenly between lines B and C, and an equal fall saves $3,000, all of it on line C. Line A’s flat payment lifts the savings rate. Without it, the combined savings rate for lines B and C falls from 15.0 to 13.6 per cent.

A smaller commitment, sized to cover $25,000 of line B’s $30,000, would take coverage to 65 per cent. The larger commitment is $1,500 a month cheaper at base and after a rise, and $1,500 dearer after a 10 per cent fall. A rise adds $3,000 a month to line B under either sizing, so neither size protects line B against it, while line A’s fixed price holds line A at $24,000. The gap between the two sizes moves only when line B’s prices or eligible usage fall: it narrows, then reverses once the fall passes 5 per cent.

Fixed prices have a cost too. When AWS cut P5 prices by up to 45 per cent in June 2025, the new Savings Plan rates applied to plans effective after 4 June. Any commitment, fixed or floating, also stays payable if the workload shrinks.

A cut has to come early to favour the smaller commitment

Over a 12-month term, the larger commitment saves $18,000 against the smaller one if prices hold. In this example the break-even is a cut of 5 per cent in line B’s prices, in effect from month 1 and held for the year. Anything larger favours the smaller commitment. Delay the cut and the size needed rises, because the smaller commitment has fewer months left to win back the $1,500 a month it gave up before the cut. It needs more than 10 per cent from month 7 and more than 15 per cent from month 9. From month 10 no cut is large enough, because its lead stops growing at $3,500 a month once a cut reaches about 17 per cent. A sustained fall in line B’s eligible usage works the same way, and the two combine. What counts is how far line B’s on-demand value falls below its level at signing, through price, usage or both.

The most the larger commitment can end the year behind is $42,000, and that takes a cut or usage fall of about 17 per cent or more from the first month. So the larger commitment risks up to $42,000 to save $18,000. Neither figure tells you how likely a cut is; the CFO judges that, on the evidence.

What goes in the approval paper

For each proposal, recalculate the forecast at list prices 10 per cent higher and lower, once as proposed and once with a credible smaller alternative. Size that alternative to the usage left after planned rightsizing and migrations out of scope. Treat the 10 per cent as an illustrative stress, and test announced changes at their own size and date on the services and usage they touch. Measure headroom inside each floating commitment’s scope rather than across the portfolio: this portfolio shows 70 per cent coverage while line B has no unused commitment headroom at baseline. Where a reservation and a savings plan cover the same workloads, AWS and Azure apply the reservation first, so once on-demand usage above both has gone, a further fall hits the savings plan. On Azure you can widen a savings plan’s scope, up to your enrolment or billing profile, without changing its term. Commitment left unused after a fall can then apply to other eligible usage inside the wider scope. Run floating-rate lines on hourly usage with the forecast’s growth, and split the period at each expiry. If line A renews after three months at the same 40 per cent off the new list, twelve times the monthly figures understates the 12-month sensitivity by $21,600 each way.

A negotiated discount off list can move with list too, depending on the contract, and if it rests on a committed spend total, check the fall case for a shortfall. Route each provider’s price notices (AWS, Google, Azure) to finance. Azure shows price changes only to subscription owners, contributors or a custom role, and AWS publishes its notices through SNS, so someone has to subscribe a finance address. If the gap between the two commitments over the term, in the base case or after a fall in price or usage, clears finance’s materiality threshold, the figures go in the paper.

The decision

The additional savings from a larger commitment have a break-even point against future price and usage changes, and finance should know that point before approving the purchase. Compare the proposal with a credible smaller commitment over the same term, and identify the size and timing of a price or usage reduction that would reverse the decision. Where rightsizing or a migration is already planned, build it into the forecast. Where a reduction is uncertain, weigh its likelihood and its cost against the savings available today, using the evidence you already hold: renewal dates, contract terms and the provider price notices finance receives. Approve the commitment whose economics and downside the business is prepared to accept, and record the break-even assumptions for the next renewal review.