Azure cloud cost management: commitment shape and waste matter more than the rate.

On demand Azure rates are largely fixed, so arguing them rarely pays. The savings come from choosing the right commitment for each workload and removing spend that should not exist.

What drives Azure cloud costs for enterprise buyers?

Two things drive most Azure overspend: the shape of your commitments and the resources that should not be running at all. On demand rates are largely fixed, so time spent arguing them rarely pays back. Azure rewards customers who match committed pricing to steady usage and clean up the rest.

Microsoft documents the tools and commitment options in the Cost Management documentation, and current rates sit on the Azure pricing page. How much discount you capture depends on how closely each commitment fits the workload underneath it.

The three things that change an Azure bill

• Commitment coverage. Reservations and savings plans placed on the steady base load, and nowhere else.
• Waste recovery. Shutting down idle resources and resizing oversized ones.
• Workload placement. The right size, the right region and the right service tier for the job.

Why the list price is the wrong fight

Microsoft publishes Azure rates globally, so an account team has little room to move one meter for one customer. Its room sits in the commitment: the MACC size, the incentives attached and the eligible spend definition. Negotiate there, and put engineering time into removing usage you do not need.

How do Azure reservations, savings plans and the MACC differ?

Reservations commit to a specific resource type, usually in one region, for one or three years at the deepest discount. Savings plans commit to an hourly spend across eligible compute, with more flexibility and a smaller discount. The MACC is a pledge on total Azure spend over several years, in return for discounting and incentives.

When a reservation is the right commitment

Reservations suit workloads that will not change shape during the term, as set out in the reservations documentation. The deeper discount is payment for certainty, so only buy one where you have it.

• Predictable base load. Databases, domain controllers and virtual machines that run all day, every day.
• Stable region and family. The resource type will not move to a new series or region before the term ends.
• Long horizon. A three year term only for the workloads you are most certain about.
• Payment choice. Upfront and monthly payment cost the same in total, so monthly payment keeps cash free at no premium.

When a savings plan fits better

Savings plans fit variable compute where the resource mix changes but overall spend stays stable. They trade some discount for freedom to shift across instance types and regions. The commitment is a dollar amount per hour, and any hour you do not use expires without rolling over.

Two rules shape how you size one. Azure applies reservation benefits first and the savings plan second, to whatever eligible usage is left. Savings plans also cannot be canceled or refunded once bought, so size the hourly amount to the low point of your eligible compute usage. Our reservations and savings plans comparison covers the choice service by service.

What changes for reservation exchanges in 2027

Microsoft still allows exchanges of compute reservations under a grace period that runs until further notice. From February 1, 2027, reservations bought for services that savings plans also cover, such as Virtual Machines, App Service and Azure SQL Database, will no longer be exchangeable. Reservations bought before that date keep one final exchange.

Refunds are capped at $50,000 in a rolling 12 month window per billing profile or enrollment. Treat every new compute reservation from 2027 as fixed for its full term, and buy it only after the cleanup described below.

How much Azure spend goes to idle resources, and how do you recover it?

In the Azure environments we reviewed, 15 to 30 percent of compute spend went to oversized or idle resources with no named owner. That spend has to come out before you commit to anything. A reservation bought on top of waste turns a monthly cost you could cut into a three year obligation.

Where the idle spend usually hides

• Machines stopped from inside the operating system. A VM shut down from the guest OS shows as Stopped and is still billed for compute. Only Stopped (deallocated) ends the compute charge.
• Leftovers from deallocation or deletion. Managed disks and networking charges continue after a VM is deallocated, and orphaned disks often outlive the VM itself.
• Oversized virtual machines. Sizes picked for a migration peak and never revisited.
• Development and test running around the clock. Environments used during office hours but billed for all 730 hours of the month.

Worked example: clean up before you commit

Say you run $200,000 a month of virtual machine compute at pay as you go rates, and 20 percent of it, $40,000, sits on idle or oversized machines. Assume a 40 percent average reservation discount for the illustration. The table compares reserving everything as it stands with removing the waste first.

The reserve as is route looks respectable on a slide because it shows a 40 percent saving. The cleanup route saves 52 percent of today's bill, and it avoids paying $864,000 over three years for machines that do no useful work.

How should you size an Azure MACC?

Size the MACC to the spend floor you are confident will occur, and let forecast growth land outside it. The Microsoft Azure Consumption Commitment trades a multi year spend pledge for discounting and Marketplace incentives.

Sizing the commitment to a floor

• Commit the floor. The spend you are confident will occur regardless of new projects.
• Keep forecast headroom. Growth lands outside the locked commitment, where it costs you nothing if it slips.
• Count Marketplace carefully. Qualifying Marketplace spend counts toward the commitment, but only for offers badged Azure benefit eligible and bought through the Azure portal on a subscription tied to the MACC billing account.
• Check the milestones. A MACC can carry interim milestone targets as well as an end date, and each one can trigger a shortfall charge.

Worked example: forecast commitment against floor commitment

Say your Azure spend runs at $6 million a year and the business forecasts $8 million and then $10 million. The account team proposes a three year MACC of $24 million, matching the forecast. Growth then slips, and actual spend comes in at $6 million, $6.5 million and $7 million.

Microsoft applies a shortfall charge in the form of Azure prepayment credit, and usage covered by that credit does not count toward any MACC.

You can spend that credit on Azure later, but you pay the $4.5 million at the end date whether or not you need the capacity. Set it against the extra discount the larger commitment earned before deciding the bigger number was worth signing.

Reading the MACC terms before you sign

Check shortfall treatment, eligible spend definitions and term flexibility. Microsoft sets out the program through its licensing resources, and the fine print on eligibility decides how achievable the commitment is. Azure credits from Microsoft do not decrement a MACC. A prepayment decrements it once, at purchase, and the usage it later covers does not count again.

The wider renewal position sits in our EA 2027 series, and the server side, including Azure Hybrid Benefit for SQL Server, in the SQL Server licensing guide. For negotiation detail on the commitment itself, see our MACC negotiation guide.

Why we advise against signing the biggest MACC on offer

The standard advice is to sign the largest MACC you can to earn the deepest discount. We disagree. In roughly half of the Azure environments we reviewed, an aggressive MACC set to growth forecasts turned into a shortfall risk when the growth slipped, and the shortfall erased the discount.

The better course is to size the MACC to a spend floor you can prove from billing history, layer reservations and savings plans on the proven base load, and keep headroom for the forecast you have not yet earned.

The deepest discount on a commitment you miss costs more than a shallower one you fill.

What have we seen in 20 to 35 Azure cost reviews?

Across roughly 20 to 35 Azure environments we reviewed between 2024 and 2025, the spend problem almost always came down to commitment shape and waste. The unit rate was almost never the cause. Three patterns came up again and again.

• Idle drag. Oversized and idle resources with no owner took a large share of compute spend, at the level described above.
• Commit mismatch. Reservations and savings plans covered the wrong workloads, which left steady use running on demand and bursty use sitting on commitments it could not fill.
• MACC overreach. Spend commitments set to growth forecasts created shortfall exposure when the growth slipped, in about 1 in 2 of the environments.

Act on ownership first. In almost every review, the idle resources belonged to projects that had ended or teams that had reorganized, and no one had the job of switching them off. They kept billing until a named person was made accountable for each subscription.

What will the Microsoft account team say about Azure commitments?

Expect the account team to push toward a bigger and longer commitment. These are the lines we hear most often, with the reply we recommend.

• "A larger commitment gets you a better discount tier." Ask for the offer priced at your floor and at the forecast side by side, with the shortfall treatment for each. Then compare the extra discount in dollars against the shortfall you would face if growth slipped.
• "Marketplace spend counts, so the number is easy to reach." Ask which of your current software vendors sell Azure benefit eligible offers, and confirm you can move those purchases into the Azure portal on the MACC billing account. Count only those.
• "Reservations can always be exchanged later." That holds for compute today under the grace period. From February 1, 2027, new reservations on savings plan eligible services lose it. Savings plans cannot be canceled at all.
• "Move everything to a savings plan for flexibility." The savings plan discount is lower than the reservation discount for the same VM. Keep reservations for stable databases and virtual machines, and use the savings plan for the variable layer above them.

Which Azure commitment terms should you ask for?

Ask for terms that limit the cost of being wrong about growth. Microsoft will not grant all of them, but each one you get reduces shortfall exposure.

1. Shortfall roll forward. The right to carry an unmet balance into an extended term or the next agreement, instead of an invoice at the end date.
2. Back loaded milestones. Annual targets that rise with the migration plan, or a single end date target, so a slow first year does not trigger a charge.
3. A written eligible spend definition. A list of the services and Marketplace offers that decrement the MACC, attached to the agreement, so eligibility cannot change under you mid term.
4. Term extension right. The option to extend the end date by a fixed period if spend is close to the commitment but short of it.
5. Discount held for the full term. The negotiated discount applies to all eligible spend until the end date, including growth above the commitment.

How do you check your own Azure cost position?

Start in Cost Management and Azure Advisor. Both are included with Azure and show most of what you need before a negotiation.

Reports and settings to pull

• Cost analysis by tag and resource group. Shows which spend has an owner. Untagged spend is where idle drag hides.
• Azure Advisor cost recommendations. Flags underused virtual machines to resize or shut down, and suggests reservation and savings plan purchases from your own usage history.
• Reservation and savings plan usage. The Reservations and Savings plans pages show how much of each commitment is used. A low figure means commit mismatch. Check it every month.
• The MACC tile. Under Credits + Commitments for an Enterprise Agreement, or Benefits for a Microsoft Customer Agreement, it shows the commitment, remaining balance and the invoiced events that decremented it.
• VM power state. List machines in the Stopped state that are still allocated, and disks with no VM attached.
• Azure Policy. The built in policy that requires a tag on resources stops new untagged spend from appearing.

Our guide to Azure cost optimization tools covers each report in more detail, and our tagging strategy guide shows how to assign owners at scale.

What to do next

1. Twelve months before the end date: tag and assign owners. Tag every resource with a cost center and a named owner, so every line of spend has someone who can switch it off.
2. Over the next three months: recover idle spend. Identify idle and oversized resources and remove or resize them before committing to anything. The cleanup needs months to show in billing data.
3. Six months out: agree the base load. Measure the steady base load that is safe to commit from the cleaned data, with finance and engineering agreeing the number.
4. Three months out: price the floor and the forecast. Ask Microsoft for both offers side by side, with shortfall, milestone and eligible spend terms in writing. Do not wait for the expiry alerts Microsoft sends at 90, 60 and 30 days.
5. One month out: size the MACC to a floor. Commit to a floor you can prove from billing history, leave the growth forecast outside it, and sign only once the contract terms you asked for are in the paper.
6. After signature: match commitments to workload shape. Cover the proven base load with reservations and savings plans chosen for how each workload behaves over the full term.

Frequently asked questions