Blog
Azure savings plan explained and when it beats pay-as-you-go
Learn when an Azure savings plan beats pay-as-you-go, how auto-scaling changes break-even utilization, and how to size commitments.
Table of Contents
- How an Azure savings plan works hour by hour
- What "up to 65%" covers
- Where pay-as-you-go still beats a commitment
- Savings plan or reserved instance for Azure Virtual Desktop
- Sizing a commitment around auto-scaling
- How Nerdio Manager supports baseline-first cost planning
- What this means for your Azure desktop budget
- Frequently asked questions about Azure savings plan
The Azure portal is showing you a savings plan recommendation, and finance wants a predictable budget number locked in for next year. An Azure savings plan is most likely to beat pay-as-you-go when your Azure compute usage stays consistent hour after hour. On an auto-scaled Azure Virtual Desktop estate, that consistent floor sits well below the total on your current bill, and the size of the commitment is what determines whether the discount ever pays back.
This guide is for Azure Virtual Desktop owners and FinOps teams deciding how much compute usage they can safely commit.
How an Azure savings plan works hour by hour
You agree to spend a fixed dollar amount per hour on eligible compute across all Azure regions for a one-year or three-year term. An Azure savings plan for compute is a commitment-based discount. Microsoft prices eligible usage at discounted rates in return, so a $5.00-per-hour commitment on a three-year term discounts every hour of eligible usage until that $5.00 is consumed. The plan is a billing construct and it never touches the runtime state of your resources.
Azure calculates and applies the benefit hour by hour in the following:
- The benefit applies automatically each hour: Azure applies it first to the usage with the deepest discount relative to pay-as-you-go, with no manual assignment to individual VMs.
- Each hour is use-it-or-lose-it: Unused commitment in a given hour expires permanently and never rolls over.
- Usage above the commitment bills at pay-as-you-go rates: Overuse at 9 a.m. does not offset an idle 3 a.m.
The plan covers the following eligible services:
- Azure Virtual Machines (what your Azure Virtual Desktop session hosts run on)
- Azure App Service Premium v3 and Isolated v2 plans
- Azure Functions premium plan
- Azure Container Instances
- Azure Dedicated Host
- Azure Container Apps
- Azure Spring Apps for Enterprise
Pay-as-you-go, savings plans, and Azure reserved instances are the three Azure Virtual Desktop pricing options available. Among these, savings plans give you four scope options that range from a single resource group up to shared coverage across the whole enrollment, and you can change that scope after purchase.
A handful of settings are locked in from the moment you buy. The hourly commitment amount, the term length, and the billing frequency cannot change after purchase. Auto-renew defaults to Off, and the all-upfront and monthly payment options total the same amount. Buying one requires an Enterprise Agreement, a Microsoft Customer Agreement, or a Microsoft Partner Agreement.
Those irreversible terms are why the hourly forfeiture rule matters so much. It separates the headline discount from the discount you actually realize, and the gap between the two starts with what that headline rate covers.
What "up to 65%" covers
Microsoft's headline is up to 65% off pay-as-you-go prices. A January 2026 Microsoft example uses a three-year term on an M64dsv2 VM running Ubuntu Linux in East US.
The actual discount you receive lands anywhere in a range from 11% to 65%, depending on four variables. Term length sets the ceiling, since three-year commitments always discount more than one-year ones. VM series and region determine the specific rate, so an M-series VM in East US will price differently than a D-series VM in West Europe. Commitment size then decides how much of your hourly usage the plan actually absorbs before the rest spills over to pay-as-you-go rates.
The headline rate also covers a narrower slice of your bill than it looks. A savings plan excludes software, networking, and storage charges, and on Windows VMs the usage meter splits in two. One meter tracks compute and the other tracks Windows licensing, and the savings plan discounts only the compute meter. The licensing side runs through Azure Hybrid Benefit instead. Microsoft says pairing it with a reserved instance can save up to 80% more than the standard pay-as-you-go rate for Windows Server.
That split matters for Azure Virtual Desktop. Sizing an hourly commitment from your full session-host bill pulls in Windows software meters and other hidden cloud desktop costs, and the result is a commitment larger than the discount can ever cover. A defensible number comes off the compute meter alone.
Even a well-scoped compute number, though, only pays back if you actually consume it. Consumption is where the commitment math tends to break down.
Where pay-as-you-go still beats a commitment
Pay-as-you-go wins when hourly utilization falls below the plan's break-even threshold. A low floor relative to peak is a warning sign. Microsoft's own Partner Center documentation shows how quickly realized savings shrink when the commitment doesn't match usage. In its worked example, a $0.10-per-hour commitment applied against a VM with a $0.3264 pay-as-you-go rate and a $0.22381248 savings plan rate (a 31.43% discount) covers 0.4468 hours of usage per hour. The remainder bills at full pay-as-you-go, and the net effective savings on total spend comes out to roughly 14.04%.
Utilization determines whether the commitment breaks even within the term, according to the FinOps Foundation's break-even model. At a 33% discount on a 12-month term, a commitment running at 100% utilization breaks even in month 8. At 80% utilization, break-even slips to month 10, and only two months of net savings remain. In that same example, below roughly 67% utilization the commitment never breaks even inside the term.
Pay-as-you-go stays the cheaper option in several recognizable situations:
- Spiky workloads with a low floor burn commitment during quiet hours: Spikes bill at pay-as-you-go anyway.
- Seasonal patterns can waste a commitment in either direction: A commitment sized for the academic year or a retail peak wastes money in the trough; sized for the trough, it leaves the peak uncovered.
- Aggressively auto-scaled Azure Virtual Desktop estates can leave little compute to absorb a commitment: When auto-scaling stops and deallocates session hosts overnight, compute consumption drops toward zero.
- Workloads that won't last the term favor pay-as-you-go: Microsoft's states: "If you expect to use the resource for less than a year, you may be better off with pay-as-you-go pricing."
Nerdio Manager for Enterprise's patented auto-scaling can deallocate idle session hosts, establishing a cleaner compute baseline before teams evaluate Microsoft's native purchase recommendations.
You can quantify your waste today in Azure Cost Management. Amortized cost data filtered on ChargeType = UnusedSavingsPlan surfaces the forfeited dollars line by line. That line item is the number nobody wants to explain at the Monday finance review.
Savings plan or reserved instance for Azure Virtual Desktop
For a stable floor that remains fully utilized, both instruments beat standard pay-as-you-go rates. They differ in what you lock and what you can undo. Azure reserved instances sit at the rigid end of that trade.
|
Dimension |
Azure savings plan |
Azure Reserved VM Instances |
|
Commitment |
Fixed dollars per hour across eligible compute, all regions |
Specific VM family in a specific region |
|
Maximum discount |
Up to 65% (3-year term) |
Up to 72% (3-year term) |
|
Flexibility |
Any eligible compute service, any region |
Instance size flexibility within the same size group and region |
|
Cancellation |
None; all purchases are final |
Refundable, capped at $50,000 USD per billing profile in a rolling 12-month window |
|
Unused benefit |
Hourly commitment expires unused |
Discount wasted for hours the matching VM isn't running |
Microsoft measures both ceilings against 24/7 pay-as-you-go pricing, and both assume you consume the discounted usage in full. The 65% figure comes from a single three-year M64dsv2 example and the 72% figure from a single three-year M64ls example, each on one VM size running continuously for 36 months. Auto-scaling reduces committable compute in environments that are not running 24/7, narrowing the realized savings delta from either instrument.
Per Microsoft "all savings plan purchases are final." Reservations can be refunded within the cap, and you can trade eligible existing reservations in for a savings plan, though Microsoft does not allow the reverse.
Starting February 1, 2027, reservations purchased after that date lose exchange eligibility for services covered by savings plans, including Azure Virtual Machines. Eligible active reservations purchased before the deadline keep one final exchange. That change narrows the flexibility edge reserved instances have traditionally held.
Microsoft recommends reserved instances for workloads that "run continuously and are highly stable" and savings plans for dynamic workloads that shift across instance families or regions. The two stack. Azure applies reservation benefits first.
Sizing a commitment around auto-scaling
Right-sizing comes before any purchase decision. Azure auto-scaling reshapes your consumption curve, and a commitment sized before that curve settles will overshoot your real baseline.
The number worth committing to is the always-on floor that remains after auto-scaling, right-sizing, and the recommendation engine's look-back window have all had a chance to catch up.
Right-size before you commit
Microsoft's own cost management guidance puts right-sizing before any purchase. Find the underused resources, shut them down or resize them, and treat the commitment as the last step. Discounts reduce rates, not waste, and the commitment should be sized to the clean baseline that remains.
For Azure Virtual Desktop that sequence has a specific shape. Azure auto-scaling reshapes your consumption curve. Pooled multi-session hosts can be stopped and deallocated overnight and on weekends, and the always-on floor that remains is the only usage worth committing to.
Sizing a commitment before Azure auto-scaling goes live uses a consumption curve that auto-scaling will reshape.
Commit to the floor you can keep utilized
Once the baseline stabilizes, the next question is how much of it to lock in. The FinOps Foundation's sizing guidance follows the same logic as Microsoft's. Its success measures put steady-state utilization for resource-based commitments such as reserved instances at an upper waterline around 80%, and treat a break-even inside nine months as the threshold for a defensible purchase.
Both point the same direction. Commit to the floor you can keep utilized, and leave peak usage on pay-as-you-go. A $5-per-hour commitment against $3 of hourly consumption leaves $2 unused every hour of the term.
Watch the recommendation engine's look-back window
Azure's native recommendation engine handles the arithmetic. It builds candidate hourly commitments from your on-demand usage over the last 7, 30, and 60 days, runs hundreds of simulations against those candidates, and surfaces only the ones that produce net savings. But it models the usage it can see.
If auto-scaling went live two weeks ago, a 60-day look-back still contains the older, fatter curve. Microsoft partly guards against this by running a separate three-day pass and giving you the lower of the three-day and 30-day recommendations, so a recent drop in usage doesn't turn into an overcommitment. Even with that guardrail, the number worth committing to is the floor you can measure after the new pattern has accumulated, with the scaled burst layer left on pay-as-you-go.
Repeating that analysis across a dozen host pools, each running its own schedule and scaling behavior, is where a management layer earns its keep.
How Nerdio Manager supports baseline-first cost planning
Nerdio Manager applies the same baseline-first order. Its patented auto-scaling deallocates idle session hosts first, and Nerdio Advisor then uses measured utilization to produce right-sizing and optimization recommendations. This leads to a cleaner compute baseline for teams to assess before they commit budget.
Nerdio Advisor includes a cost calculator for TCO comparisons across Windows 365, Azure Virtual Desktop, and physical desktops, plus a right-sizing tool that analyzes usage to prevent overprovisioning. The commitment instruments are Microsoft's, Nerdio Manager is the management and optimization layer that shrinks the usage baseline you end up committing to.
Equitable Bank saw 74% compute savings per month from auto-scaling, and TechTarget's Enterprise Strategy Group found up to 55% Azure Virtual Desktop cost savings with Nerdio Manager compared with Azure Virtual Desktop alone.
Because savings plans cover only the eligible Azure compute services listed earlier, Windows 365 Cloud PC licensing falls outside this hourly commitment math. Nerdio Advisor produces right-sizing and optimization recommendations per Cloud PC and per host pool from measured utilization, without applying Azure Virtual Desktop commitment math to Windows 365.
For Windows 365 endpoints, Nerdio Manager extends Microsoft Intune management for Cloud PCs, including application and policy operations. Its Unified Application Management can deploy applications in about 30 seconds, compared with up to three hours through native Microsoft Intune delivery.
For the many enterprises running both products, that means one set of recommendations to act on rather than two cost models to reconcile by hand.
What this means for your Azure desktop budget
The portal recommendation becomes answerable once right-sizing and auto-scaling have reshaped consumption. From that cleaner baseline, a savings plan covers an estate that shifts across VM families and regions, a reservation covers a stable VM footprint, and the scaled burst layer stays on pay-as-you-go.
Sequenced that way, finance gets a fixed commitment plus a variable layer instead of a guess about next year's usage.
The full budget picture also includes Microsoft licensing and Nerdio Manager pricing, or the cost of any other third-party management layer, alongside the Azure compute number.
If you want to automate auto-scale schedules and track utilization from one management layer, you can get a demo to see how Nerdio Manager works across your Windows 365 and Azure Virtual Desktop environment, or try it free in your Azure tenant.
Frequently asked questions about Azure savings plan
Resources
Learn more about Nerdio Manager
