The cloud commitment decision is one of the few in IT where finance and technology have to make the same call together. This is the framework that works.
Unlike many technology decisions, cloud questions aren’t purely an IT call. The commercial implications can last for years, making it just as important to finance leaders as it is to infrastructure teams.
Get it right, and you save 30–50% on the workloads it covers. Get it wrong, and you're locked into a multi-year contract for capacity you no longer need, at prices the rest of the market has moved past.
Most decisions go wrong not because the maths is complicated – it isn't – but because the question is framed badly.
Before deciding whether to commit, businesses need to understand the options available to them. Each commitment model offers a different balance of cost savings, flexibility and risk. The right choice depends less on the size of the discount and more on the nature of the workload you're trying to support.
The three instruments
Not all cloud discounts work in the same way. The key is understanding how much certainty you have about future demand and how much flexibility you want to retain. These are the three most common commitment models available today:
Reserved instances (RIs). A commitment to a specific Virtual Machine (VM) Stock Keeping Unit (SKU), in a specific region, for either one year or three years, in exchange for a discount of 30–72% depending on term length and payment option (upfront vs monthly). Best suited to: stable, predictable workloads that won't change shape or location for the term of the commitment.
Savings plans. A commitment to a fixed hourly spend (dollars per hour) across a flexible portfolio of compute services. Discount is slightly smaller than RIs – usually 20–65% – but the flexibility is significant. Best suited to: workloads where the volume is predictable but the underlying SKU or service may change.
Commit deals (EA, MCA, CSP-level). A higher-level commitment, usually annual, made at the subscription or tenancy level. Best suited to: large estates where the total spend is predictable even if the individual workload mix is fluid.
These three instruments are not exclusive. Mature estates use all three at different layers of the cost model.
The four questions before any commitment
Cloud commitments are often sold on the size of the discount. But the real question is whether the workload is predictable enough to justify giving up flexibility. Before making any commitment, businesses should answer four key questions about utilisation, predictability, risk and future infrastructure plans.
What's the utilisation history of the workload? Twelve months of usage data is the minimum baseline. Six months is not enough. Three months is gambling. The right question isn't "what's the average utilisation?" It's "what's the lowest 30-day utilisation in the last 12 months, and could we live with a commitment sized to that?" Reservations and savings plans break even at around 70–80% utilisation. Below that, the discount is cancelled by the underutilisation cost.
How predictable is the workload? Stable workloads (databases, identity services, ERP systems, virtual desktops) are good commit candidates. Variable workloads (dev/test, batch processing, analytics, ML training, customer-facing apps with seasonal spikes) generally aren't. A useful rule of thumb: if your workload has changed shape, size or region in the last six months, don't commit yet.
How locked-in are you happy to be? Three-year reservations carry the biggest discount. They also carry the biggest risk. The cloud market has moved significantly in the last 36 months. The capacity you committed to in 2024 may be wrong for your business in 2027. The CFO question to ask the IT leader: "if we walked away from this commitment in 18 months, what would it cost us?" Then ask the same question of every other commitment proposal that follows.
What's the strategic direction of the platform? This is the question most businesses skip and most regret. If you're considering repatriating some workloads in the next 24 months – and 87% of UK businesses are at least exploring that question – then committing to large RI portfolios at the same time is a strategic conflict. The CFO should always ask: "are we committing to spend on infrastructure we may not still be using by the time the commitment ends?"
The decision framework
Use this as a starting point. Adjust for your own organisational appetite.
Steady-state, fully predictable, 18+ months of consistent demand history → 3-year reserved instance. Maximum discount, lowest flexibility, only safe with high confidence in the workload.
Predictable but with some likely change of shape → 1-year savings plan. Solid discount, much more flexibility. Most mid-market estates should default here.
Volume predictable but service mix uncertain → 1-year savings plan, smaller commitment value. Locks in spend without locking in technology choices.
Large total estate, mixed workload portfolio → annual commit deal at the agreement level. Best handled through a CSP partner who can negotiate, administer and adjust the commitment over the year.
Anything unpredictable or strategically unstable → no commitment. Pay as you go. The flexibility is worth more than the discount.
Where hybrid changes the maths
This is the part the major hyperscalers won't volunteer.
If you operate a hybrid estate, you can use Virtual Data Centres (VDC) – a no-contract sovereign IaaS platform – as the predictable floor under your cost model. The steady-state workloads run on VDC with consumption-based pricing and no commitment. The variable workloads run on Azure with appropriately-sized commitments.
The maths becomes considerably more favourable. You commit less. You discount more. You retain more flexibility. The CFO conversation gets significantly easier.
One worked example
A mid-market UK customer is running approximately £20,000 per month of steady-state Azure consumption, plus £8,000 per month of variable workloads.
Without optimisation: pay-as-you-go, £28,000 per month, £336,000 annually.
With a Cloud Service Provider (CSP) -led optimisation: £15,000 per month of steady-state moves to VDC at approximately £9,000 per month (saving £6,000 per month gross). £5,000 per month of remaining steady-state Azure spend goes on a 1-year reserved instance portfolio at ~55% discount (saving £2,750 per month). £8,000 per month variable spend stays pay-as-you-go.
New monthly run-rate: ~£18,250 per month. Annual saving: ~£117,000.
The numbers vary by workload. The pattern doesn't.
What to ask before your next commitment
If the answer to question 4 is "I don't know," the optimisation conversation hasn't started yet.