Blog
Azure Reserved Instances: how they work, what changed, and when they save you money
Learn how Azure Reserved Instances work, what's changing in 2026, and how teams can maximize savings on cloud desktop compute costs.
Table of Contents
- What are Azure Reserved Instances?
- How does the discount apply across your VMs?
- How does instance size flexibility stretch one reservation?
- What changed for Azure Reserved Instances on July 1, 2026?
- Reserved Instances or Azure Savings Plans: which should you choose?
- Can you exchange, cancel, or refund a reservation?
- How do reservations fit Windows 365 and Azure Virtual Desktop?
- Where does Nerdio Manager fit?
- Frequently asked questions about Azure Reserved Instances
Azure Reserved Instances let you commit to one or three years of Azure compute in a specific region in exchange for a lower hourly rate. Microsoft measures that discount against a VM running 24/7 at list price, so how much you actually save depends on how consistently your workloads run.
This guide walks Managed Service Provider (MSP) owners, FinOps practitioners managing Azure spend, and enterprise teams running cloud desktops through when a reservation pays off. It covers Windows Cloud, which encompasses both Windows 365 and Azure Virtual Desktop.
What are Azure Reserved Instances?
An Azure Reserved Instance is a billing commitment rather than a server. You agree to keep paying for a set amount of Azure compute for one or three years, and in return Microsoft discounts the matching usage on your bill. It never starts a resource on its own.
Microsoft now files Reserved Instances under the broader label Azure Reservations, but the mechanics are the same: you pick a term, and Microsoft applies the discount automatically, every hour, to any resource that matches. According to Microsoft Learn, reservations can reduce your resource costs by up to 72% versus pay-as-you-go, while Microsoft's own pricing page gives the practical range as 36% to 72%, depending on the VM series, region, and term.
The 72% figure applies to VMs running 24/7 at list price. If your workloads already power down and deallocate overnight, the real savings shrink, and an oversized reservation turns into a bill you pay whether the VMs run or not. Whether a reservation pays off comes down to how many hours your VMs run.
A reservation buying decision depends on:
- Terms
One-year and three-year terms are standard, and three-year terms carry the deeper discount. - Payment
Upfront or monthly. Per Microsoft Learn (cited previously), "the total cost of up-front and monthly reservations is the same," with no extra fee for paying monthly. - Coverage
A VM reservation covers compute only. Azure still bills storage, networking, and Windows licensing separately. Windows licensing savings come from Azure Hybrid Benefit, which stacks on top of a reservation. - Eligible services
Beyond virtual machines, you can reserve capacity for Azure Blob storage, Azure Data Lake Storage Gen2, SQL Database compute, and Azure Cosmos DB throughput, plus Virtual Machine Scale Sets, Azure Kubernetes Service, Azure Batch, and Service Fabric on the VM side.
Together, these four choices decide how closely a reservation fits the workload you are trying to cover.
How does the discount apply across your VMs?
Azure applies the discount automatically, every hour, to any running VM that matches the reservation and sits inside its scope. Scope is the boundary you set when you buy the reservation. It decides which VMs the discount can reach. Azure started with two scope options and now offers four:
- Single resource group - Applies only to matching resources in one resource group.
- Single subscription - Applies to matching resources anywhere in the selected subscription.
- Shared - Applies to matching resources across all eligible subscriptions in the billing account, such as an Enterprise Agreement (EA) or eligible pay-as-you-go subscriptions. This is the scope most relevant to MSPs and enterprises with multiple subscriptions.
- Management group - Applies to matching resources in subscriptions that sit in both the management group and the billing scope.
When more than one scoped reservation matches a VM, Azure applies them in a fixed order: single resource group first, then single subscription, then management group, then shared.
The discount is use-it-or-lose-it. If no matching resource runs in a given hour, that hour's reserved quantity is gone; it does not carry forward. Where the discount lands also depends on VM state.
- A stopped VM is still billing for compute, so it still uses up the reservation's hours.
- A deallocated VM stops billing for compute, so there is nothing for the discount to land on. Azure re-applies it to another matching VM in scope; if none exists that hour, the reserved hours go to waste.
Monitor usage after you buy. The Azure Portal shows usage percentages, and reservation usage alerts email you when a reservation is running underused. Nerdio Manager for MSP gives MSPs the same usage visibility and right-sizing recommendations before they commit client spend to a reservation.
How does instance size flexibility stretch one reservation?
Per Microsoft's size flexibility guidance, instance size flexibility lets one reservation cover several different VM sizes, not just the exact size you bought. It is the default when you buy a Reserved VM Instance, and it applies across every size in the same flexibility group. The alternative, capacity priority, reserves capacity for one specific size; you can only use it with single-subscription scope, and turning it on gives up size flexibility.
Flexibility groups are subsets within a VM series, and the boundaries are strict. A reservation for Standard_DS3_v2 applies to Standard_DS1_v2 through Standard_DS4_v2, but not to the DSv2 High Memory group (Standard_DS11_v2 and up). A Standard_D1 reservation does not apply to Standard_DS1 VMs, because D-series and DS-series are separate groups.
Inside a group, each size carries a ratio that reflects its relative compute footprint, and the math works in both directions. A reservation for Standard_DS4_v2 (ratio 8) plays out like this.
| Running VMs | Ratio math | Result |
| Eight Standard_DS1_v2 (ratio 1 each) | 8 × 1 = 8 | Discount covers all eight VMs |
| Two Standard_DS2_v2 (ratio 2) + one Standard_DS3_v2 (ratio 4) | 2 + 2 + 4 = 8 | Discount covers all three VMs |
| One Standard_DS5_v2 (ratio 16) | 8 ÷ 16 = 0.5 | Discount covers half that VM's compute cost |
| One Standard_DS5_v2 + a second DS4_v2 reservation | 8 + 8 = 16 | Both reservations cover the whole VM |
One change to note (from the previously cited size flexibility guidance): the ratio CSV file stopped receiving updates on May 9, 2026, and Microsoft will remove it on August 30, 2026. Ratio data now lives in the Catalogs API (the ReservationsAutofitGroup and ReservationsAutofitRatio properties) or in PowerShell.
What changed for Azure Reserved Instances on July 1, 2026?
As of July 1, 2026, Microsoft stopped selling and renewing Azure Reserved VM Instances for a list of older VM series. If you already hold a reservation on one of the affected series, it runs to the end of its term, but it will not auto-renew and you cannot buy more.
So take inventory of your reservations. Flag anything on a series Microsoft now labels legacy, and map each one to an exchange or a trade-in before its term expires. For MSPs holding reservations across dozens of client subscriptions, this is real work. A reservation that quietly fails to renew flips that compute back to pay-as-you-go rates on the next invoice.
Reserved Instances or Azure Savings Plans: which should you choose?
Choose a reservation when the workload is stable and stays on one VM family in one region. Choose an Azure Savings Plan when your usage moves across sizes, regions, or services. A savings plan commits you to a fixed hourly dollar amount across eligible compute rather than to a specific VM, and you trade some discount for that flexibility.
| Metric | Reserved VM Instances | Savings Plan for Compute |
| Maximum discount | Up to 72% vs. pay-as-you-go | Up to 65% vs. pay-as-you-go |
| Commitment type | Specific VM family + region | Hourly dollar spend across eligible compute globally |
| Flexibility | Size flexibility within series group; region-locked | Applies across regions, series, and OS automatically |
| Exchanges and refunds | Available (see next section) | Not available |
Both maximums come from Microsoft's Reserved VM pricing and Savings Plans pricing pages. Your real discount depends on the VM series, region, term, and how consistently the workload runs.
Microsoft's own decision framework draws the same line: reserve when a workload runs continuously and stays on one instance type, family, and region, and pick a savings plan when the workload is dynamic or evolving, uses multiple instance families or compute services, or shifts between regions.
You can hold both. When a reservation and a savings plan both match a resource, Azure applies the reservation first, because it is the more specific commitment with the deeper discount. To bring down cost in an estate you already run, Microsoft recommends working in this order:
- Right-size existing resources to keep invoices defensible
- Exchange underused reservations
- Trade in underused reservations for savings plans
- Buy new reservations for stable workloads
- Buy new savings plans for the remaining baseline
If you want to model the savings before committing either way, run the estate through a cost estimator first.
Can you exchange, cancel, or refund a reservation?
Yes, within limits. You can exchange a reservation, trade it in for a savings plan, or cancel it for a refund. Here are the current rules from Microsoft's exchange and refund policy.
- Exchanges are still open
Microsoft's planned January 2024 end date for compute reservation exchanges was extended until further notice, with at least six months' warning before it ends. Compute reservations bought during this grace period keep the right to one more exchange after the grace period closes. - Exchange rules
An exchange stays within the same product family. You can change term, region, series, SKU, quantity, and payment frequency, and Microsoft allows partial-quantity exchanges. The new reservation's total commitment must equal or exceed what remains on the original, and its term restarts from the exchange date. Exchanges carry no penalty and no annual limit, and exchange-related refunds do not count against the refund cap. - Cancellations
Refunds are capped at $50,000 USD in a rolling 12-month window per billing profile or enrollment (per customer for Cloud Solution Provider accounts). Microsoft charges no early termination fee today, though it reserves the right to add a 12% fee in the future. It calculates the refund on the lower of purchase price or current price, and only reservation order owners can process one. - Trade-ins
You can trade in a reservation for a savings plan with no time limit. It does not work in reverse: a savings plan cannot be exchanged for anything.
Microsoft excludes Red Hat plans, SUSE Linux plans, and pre-purchase plans from both exchanges and refunds.
How do reservations fit Windows 365 and Azure Virtual Desktop?
Reservations apply to Azure Virtual Desktop, and they only pay off on the session hosts that run around the clock. Windows 365 uses flat-rate per-user pricing, so its cost is predictable by design; you manage it through license right-sizing rather than reservations. Many enterprise customers run Windows 365 and Azure Virtual Desktop together, each for different workloads.
Azure Virtual Desktop pricing runs on consumption-based Azure compute, and Microsoft's documentation on Azure Virtual Desktop costs names both levers: autoscale to scale session hosts to demand, and Azure savings plans or Azure reserved VM instances to bring the compute rate down.
Those two levers can work against each other if you buy a reservation before you know your always-on floor.
Auto-scaling deallocates session hosts off-hours, which is exactly what you want for the bill. But a deallocated host releases its reservation discount. If no other matching VM runs that hour, you lose the reserved hours while the reservation keeps billing. In short, reserving capacity that your auto-scaling routinely shuts off means paying for compute that is not running.
Work in this order before you commit reservation spend:
- Auto-scale and right-size first. Find your true always-on floor before committing to anything. Microsoft recommends basing reservation purchases on consistent base usage, so validate your scaling behavior first with Azure auto-scaling.
- Reserve only the 24/7 floor. Domain controllers, persistent personal desktops, and always-on multi-session pooled hosts that never deallocate are reservation candidates. Variable multi-session capacity is not.
- Cover variable capacity with a savings plan or pay-as-you-go. Savings plans flex across series and regions, which fits desktop capacity that moves up and down.
That split reserves your stable compute and keeps your variable desktop capacity flexible.
Where does Nerdio Manager fit?
Nerdio Manager pinpoints the stable, always-on capacity that is safe to reserve before you commit money to it. Reservations are Microsoft technology, and reservation planning is one cost-management workflow inside Nerdio Manager, a cloud desktop management platform for Microsoft Windows Cloud environments.
For MSPs: cleaner reservation calls across client tenants
Nerdio Manager for MSP manages Windows 365, Microsoft Intune, and Azure Virtual Desktop from one multi-tenant console, surfacing reservation and savings plan recommendations alongside cross-client usage. Seeing real usage across every client makes the buy decision cleaner and reduces the chance of reserving capacity a client will not use. One platform covers desktops and endpoints across both Windows Cloud paths through an Intune for MSPs service, keeping Intune a core managed service that supports margin.
For enterprise teams: reserve only your true always-on floor
The reliable way to size a reservation is to strip off-hours waste with auto-scaling first, then reserve the floor that never turns off. Nerdio Manager for Enterprise runs Windows 365, Microsoft Intune, and Azure Virtual Desktop from one console and drives that sequence for you.
For Azure Virtual Desktop, Nerdio Manager uses patented auto-scaling to deallocate idle session hosts and switch their OS disks to cheaper storage tiers while they sit idle. Equitable Bank achieved 74% compute savings per month, Penn State reported a 71% reduction in Azure Virtual Desktop spend while adding 1,000+ users, and Sage reported $1.5 million in annual savings. Once auto-scaling clears that off-hours waste, your genuine always-on floor is the only capacity worth committing to for one or three years.
From the same console, Nerdio Advisor right-sizes Windows 365 Cloud PCs and flags Flex (previously called Frontline) license conversions for users who never overlap, such as swapping three $31/month enterprise licenses ($93 total) for one $62/month Flex license that covers all three. Unified Application Management deploys applications across Cloud PCs and Azure Virtual Desktop hosts, and Nerdio Manager can create, back up, and restore Microsoft Intune policies from that same console.
Get a demo to see how Nerdio Manager identifies your true always-on floor and sizes reservations against real Azure Virtual Desktop usage, or try it free in your Azure tenant.
Frequently asked questions about Azure Reserved Instances
Resources
Learn more about Nerdio Manager
