Azure Virtual Machine Cost in 2026

Ask Azure what a virtual machine costs and it will give you an hourly rate to six decimal places. Ask your finance team what your Azure VMs cost and they will give you a number two or three times larger. Both are correct, and the gap between them is the subject of this guide. The hourly rate is the price of compute; the invoice is the price of compute plus the disk it boots from, the IP address it answers on, the bandwidth it sends, the operating system license it may or may not need, and the hours it spent running while nobody was using it.

This guide lays out what Azure virtual machines really cost in 2026, in US dollars for the East US region, size by size. It covers the four pricing models and what each one saves, the charges that sit outside the hourly rate, a worked example with the full bill, and the order of operations for cutting it. Every figure is drawn from Azure's official Linux VM pricing and Windows VM pricing pages as of late September 2026; rates vary by region and change, so confirm yours in the Azure Pricing Calculator before budgeting.

Key Takeaways

• Common Azure Linux VMs in East US cost roughly $30 to $280 a month always-on, from a B2s to a D8s v5; memory-optimized E-series and GPU sizes cost more.
• Windows VMs carry a license premium of roughly 50 to 90 percent over Linux on the same size; Azure Hybrid Benefit removes it for customers who own licenses.
• Four pricing models apply: pay-as-you-go, Reservations (up to 72 percent off), Savings Plans (up to 65 percent), and Spot (up to 90 percent, interruptible).
• Disks, public IPs, and bandwidth typically add 30 to 50 percent on top of the VM rate, and they are where most surprises come from.
• Cut costs in order: remove idle resources, right-size, schedule shutdowns, apply Hybrid Benefit, then commit. Committing first locks waste in.

How Azure Prices a Virtual Machine

Azure bills a VM by the second, with a one-minute minimum, at an hourly rate determined by four things: the size (vCPU, memory, and family), the operating system (Linux is the base rate; Windows and some Linux distributions add a license charge), the region (US regions are generally the cheapest, with others running 5 to 20 percent higher), and the pricing model (pay-as-you-go, reserved, savings plan, or Spot). The compute rate covers the virtual CPU and memory only. Everything the VM needs to function, its OS disk, data disks, public IP, and network egress, is billed as a separate line. This is the model behind every Azure pricing decision, and understanding it is what separates the rate card from the invoice.

Azure VM Costs by Size (East US, 2026)

The table below shows pay-as-you-go rates for the most commonly deployed sizes, running 730 hours a month (always-on), on Linux and Windows. Azure's full catalog of families is described on its VM series page. Figures are rounded and exclude disks, IPs, and bandwidth.

Three patterns stand out. B-series burstable VMs are dramatically cheaper because they run below full CPU most of the time and bank credits for bursts, which makes them ideal for dev, test, and lightly loaded services but wrong for sustained compute. Windows roughly doubles the cost of the D-series, because the Windows Server license is bundled into the rate. And within a family, price scales almost exactly with vCPU count, so a D8s v5 costs twice a D4s v5, which means right-sizing down one step halves the compute line.

The Four Pricing Models and What They Save

The pay-as-you-go rates above are the ceiling. Azure offers three ways to pay less, and they are complementary rather than competing.

Two rules keep these from backfiring. First, reservations and savings plans do not stack with each other on the same usage, but both stack with Hybrid Benefit, so a reserved Windows VM with Hybrid Benefit applied can land 80 percent or more below its pay-as-you-go rate. Second, commit only after right-sizing: a 3-year reservation on a VM that is twice the size it needs locks in the waste for three years at a discount. Our guides to how reservations and savings plans work and the pay-as-you-go, reserved, and spot pricing models cover the mechanics in depth.

The Costs Outside the Hourly Rate

This is where the rate card and the invoice part ways. Every VM carries several separately billed components, and together they commonly add 30 to 50 percent to the compute line.

The two that surprise people most are disks and licensing. A managed disk bills on the size you provisioned, so a 512 GB Premium SSD costs the same whether it holds 20 GB or 500, and disks left behind when a VM is deleted keep billing indefinitely. Licensing is larger still: on a D4s v5, the Windows charge is roughly $128 a month, nearly matching the $140 compute rate, and SQL Server licensing can exceed both combined. These are the Azure-specific versions of the leaks we cover in where cloud waste hides, and finding them requires seeing cost by resource, not just by subscription.

A Worked Example: The Full Bill for One Production VM

Take a single D4s v5 running a Windows application server in East US, always-on, with a 256 GB Premium SSD OS disk, a 512 GB Premium SSD data disk, one static public IP, and 500 GB of outbound traffic a month.

At pay-as-you-go rates the bill is roughly: $268 for the Windows VM (of which about $128 is the Windows license), about $38 for the 256 GB disk, about $77 for the 512 GB disk, about $4 for the IP, and about $35 for the 400 GB of bandwidth beyond the free allowance. That is around $422 a month, and the compute rate alone would have suggested $140. The VM itself is a third of the real cost.

Now apply the levers. Hybrid Benefit removes the $128 license charge, taking the VM to $140. A 3-year reservation cuts that compute to roughly $45. Right-sizing the data disk to what it actually holds, say 256 GB, saves another $39. The same server now costs about $205 a month, less than half the original, with no change to the application. The compute discount was real, but Hybrid Benefit and the disk delivered as much as the reservation did, which is why the order of operations matters.

How to Cut Azure VM Costs, in Order

1. Delete what is not running anything

Stopped-but-allocated VMs still bill for compute; deallocate them. Unattached managed disks, orphaned public IPs, and old snapshots bill forever. Azure Advisor flags most of these, and this step is risk-free.

2. Right-size using real utilization

Advisor and Azure Monitor show CPU and memory utilization over time. Most estates find a large share of VMs running below 20 percent utilization; stepping each down one size halves its compute cost. For spiky, low-average workloads, move to the B-series and let it burst.

3. Schedule shutdowns for non-production

A dev or test VM that runs only 8 hours on weekdays costs about a quarter of one that runs 730 hours. Azure's auto-shutdown and start/stop automation make this a configuration change, and non-production is where it applies to most of the fleet.

4. Apply Azure Hybrid Benefit everywhere it fits

If you own Windows Server or SQL Server licenses with Software Assurance, apply them. On Windows-heavy estates this single step often saves more than the reservation program, and it takes minutes per VM.

5. Only then, commit

With usage honest, cover the always-on baseline with reservations, variable-but-steady compute with a savings plan, and run interruptible work on Spot. Buy conservatively and grow coverage as utilization proves out. The full playbook is in our Azure cost optimization guide, and the tagging that makes VM costs attributable to teams is in our Azure tagging guide.

How Azure VM Costs Compare to AWS

For teams weighing the two, Azure and AWS price comparable general-purpose instances within a few percent of each other on Linux; a D4s v5 at $0.192 an hour sits close to an AWS m6i.xlarge. The larger differences are structural: Azure Hybrid Benefit has no direct AWS equivalent for reusing owned Windows licenses, which can make Azure meaningfully cheaper for Windows-heavy estates, while AWS Savings Plans cover a broader set of services. Our AWS vs Azure pricing comparison works through the trade-offs in detail.

The Bottom Line

An Azure virtual machine costs what its hourly rate says only if you ignore everything attached to it. The compute line is real but partial; disks, IPs, bandwidth, and above all licensing routinely double it, and idle hours multiply it. The good news is that every one of those factors is a lever. Remove what is idle, size what remains to what it uses, shut non-production down at night, apply the licenses you already own, and then commit the steady remainder. Done in that order, the same workload commonly runs at half its original cost. Keeping it there means seeing VM cost by resource and by team continuously, which is what Opslyft's cost visibility and cost control are built to do across Azure and every other cloud you run.