TL;DR:

• Azure Reservations/Savings Plans and AWS RIs/Savings Plans price their discounts on commitment length: roughly mid-30s to mid-40s percent off on Azure, 30 to 50 percent off on AWS
• Teams under-commit not from ignorance but from rational caution: a 1- or 3-year contract is a bet that today's usage still looks the same when the term ends
• Get that bet wrong and you're paying for capacity nobody's using anymore, which is why so many programs plateau under-committed on both clouds
• Guaranteed commitments let you buy the same discount tiers with terms as short as 30 days and a rebate if usage drops, on Azure and AWS alike
• That turns "stay on-demand" or "commit long and hope" into a third option: capture most of the discount without betting on your infrastructure three years out

Every FinOps team eventually hits the same fork in the road. Your Azure or AWS bill is creeping up, someone in finance asks why you're not using Reservations or Savings Plans, and the math on paper looks obvious: commit to a term, cut your rate by 30-50%, done.

Then someone asks the follow-up question that actually matters: commit to what, for how long, and what happens if it's wrong?

That's the catch: on both Azure and AWS, the discount scales with the length of the commitment. Sign up for longer and save more, whether it's an Azure Reservation or an AWS Reserved Instance. A 1-year Azure Reservation nets meaningfully more than on-demand. A 3-year AWS Reserved Instance can cut compute costs by roughly half. But the same contracts that reward commitment punish change, and change is the one constant in cloud infrastructure. Teams re-platform, right-size, move workloads between regions, and adopt new instance families, often faster than a 1- or 3-year contract can keep up with.

The bet you're actually making

When you buy a 1-year or 3-year commitment on either cloud, you're not just buying a discount. You're betting that your usage a year (or three) from now will still look enough like your usage today for the commitment to pay off. Get that bet right, and you save real money. Get it wrong, because you migrated or modernized off that instance family, a product line got sunset, or a team shrank, and you're paying for capacity nobody's using anymore.

That risk pushes FinOps teams toward a defensive posture: under-commit to stay safe, and leave savings on the table that a more confident team would have captured. It's a rational response to an irrational amount of risk being asked of a single line item on a budget.

Where the discount tiers actually land

Roughly, and depending on instance family, region, and payment option:

• Azure: 1-year Reservations run in the mid-30s percent off on-demand; Savings Plans land a bit lower depending on flexibility. Multi-year and higher-upfront options push toward the mid-40s.
• AWS: 1-year Reserved Instances and Savings Plans typically land in the 30-40% range off on-demand; 3-year, no-upfront terms can approach 50%.

Those aren't small numbers. On a seven- or eight-figure annual cloud bill, the gap between on-demand and a well-utilized 3-year commitment is often the single biggest lever a FinOps team has. It's also the single riskiest one, because the further out the term, the harder it is to guarantee the usage will still be there to justify it.

The question worth asking before your next commitment 

Before committing to any Reservation, Savings Plan, or Reserved Instance, it's worth pressure-testing three things:

1. How confident are you in this usage profile 12-36 months out? Not "probably fine": actually confident, based on a roadmap, not a trend line.
2. What's the cost if you're wrong? Model the downside, not just the upside. A 46% discount that goes unused for eight months of a three-year term can net out worse than doing nothing.
3. Is there a way to get the discount without taking on the full length of the bet? This is the part most teams don't know to ask, because for years the honest answer was no. Cloud providers are only priced in discounts for commitments of a year or more.

That last question is exactly where the market has started to move. A newer category of tools now lets teams buy the same Reservations, Savings Plans, and Reserved Instances they'd buy directly from Azure or AWS, but with the underutilization risk insured: terms as short as 30 days, with cash back if usage drops before the commitment period ends. The mechanics are simple: you still get cloud-provider-grade discounts, you're just no longer the one holding the bag if your infrastructure changes faster than a 3-year contract can.

For teams on Azure, AWS, or both, that changes the calculus. Instead of choosing between "stay on-demand and pay full price" or "commit long and hope," there's a middle path that captures most of the discount without needing to predict your infrastructure three years into the future.

Cloud discounts were never the hard part. Confidently committing to them, on either cloud, has been. That's the part worth solving before your next renewal.

‍

If you’re leaning on Azure’s Reservation Exchanges as your own safety net for exactly this kind of risk, it’s worth knowing that safety net has an expiration date:

‍Azure Is Ending Reservation Exchanges in 2027

‍