Azure VM Pricing: How It Works and How to Reduce Costs

Virtual machines can often be the single biggest line item on your Azure bill. They’re also one of the most complicated to price.

You have dozens of VM families, multiple sizing options within each family, regional variations, disk costs, licensing considerations… it can be a bit of a headache.

The upside to all this is flexibility. There’s almost certainly a combination of VM type, size, and pricing model that’s a brilliant fit for your workload. So how do you figure it out?

We’re going to break down how Azure VM pricing actually works. You’ll find out what goes into the different costs, and how to avoid paying more than you need to.

Let’s take a look at the different components that contribute to your monthly bill. A VM isn’t just a single charge; it’s several costs bundled together.

• Compute is the core cost: the processing power and memory of the VM itself. This is what most people think of as “the VM price” and it’s typically the largest component.

• Storage covers your disks. Every VM needs at least an OS disk, and most production workloads have additional data disks. Disk costs vary dramatically depending on the performance tier you choose.

• Networking includes data transfer charges (particularly for data leaving Azure) and costs for public IP addresses if you need them. For VMs that stay within Azure and don’t shift huge amounts of data, networking costs are often modest.

• Licensing is another part to consider. A Windows Server VM costs more than an identical Linux VM because you’re paying for the Windows licence. Add SQL Server and the licence cost can exceed the compute cost.

• Additional services like Azure Backup, monitoring, and disaster recovery add to the total. These are optional (but often essential) for production workloads.

Throughout this guide, we’ll mention prices. These are approximately accurate for the UK as of late 2025, but Azure pricing changes regularly, so treat them as ballpark figures rather than quotes. Remember to check the official Azure VM pricing page for exact numbers.

Azure offers dozens of VM families, each optimised for different workloads. Here’s a quick overview of the main options:

General purpose VMs offer a balanced ratio of CPU to memory and suit most workloads:

• B-series (burstable): Ideal for variable workloads that don’t need constant CPU; accumulates credits during quiet periods.

• D-series: The Azure workhorse for consistent, predictable performance. Suits most business applications.

⠀Compute optimised VMs provide high CPU-to-memory ratios:

• F-series: Best for CPU-bound workloads like batch processing where you need more processing power than RAM.

⠀Memory optimised VMs suit applications needing lots of RAM:

• E-series: Good for in-memory databases and analytics.

• M-series: Up to 4TB RAM for demanding workloads like SAP HANA.

⠀Storage optimised VMs deliver high disk throughput:

• L-series: Designed for NoSQL databases and data warehousing with high-speed local NVMe storage.

⠀GPU VMs serve machine learning, rendering, and graphics-intensive virtual desktops:

• NC-series: Popular for ML inference and training.

• ND-series: High-end deep learning with A100 GPUs.

• NV-series: Visualisation and remote graphics.

⠀High performance compute VMs handle simulations and modelling:

• HB-series: High memory bandwidth with InfiniBand support for tightly coupled parallel workloads.

Example pricing scenarios

Pricing varies significantly based on VM size, operating system, and disk configuration. Here are two example setups using D-series VMs (Windows, pay-as-you-go, UK South region) to give you a sense of scale:

Budget: Small web server or dev environment

• D2s v3 (2 vCPUs, 8GB RAM): £115.24/month

• P10 Premium SSD (128GB): £18.10/month

• Total: ~£133/month

Performance: Application server or database

• D8s v3 (8 vCPUs, 32GB RAM): £460.94/month

• P30 Premium SSD (1TB): £124.13/month

• Total: ~£585/month

These are baseline costs. Linux VMs are cheaper (no Windows licence), and commitment discounts like Reserved Instances or Savings Plans can reduce your costs by 25% or more. For current pricing across all VM families, check Microsoft’s Azure VM pricing page.

Within each VM family, you choose a size that determines the number of vCPUs, amount of RAM, and other characteristics. Understanding the naming convention helps you make sense of the options.

A name like Standard_D8s_v3 breaks down as:

• Standard: The VM tier (as opposed to Basic, which is deprecated)

• D: The family (general purpose)

• 8: The number of vCPUs

• s: Support for Premium SSD storage

• v3: The version/generation

Pricing within a family scales roughly linearly with size. If a D2s v3 costs £62/month, a D4s v3 (double the vCPUs and RAM) costs about £124/month. This makes it straightforward to estimate costs when scaling up or down.

The key is picking the right size for your workload. Over-provisioning is one of the most common causes of cloud overspend. A VM running at 20% CPU utilisation is wasting 80% of your compute budget. We’ll cover how to right-size effectively later in this guide.

Some families include size variants optimised for specific characteristics. For example, within the E-series you’ll find:

• E4s v3: Standard memory optimised (4 vCPUs, 32GB RAM)

• E4ds v4: Same but with local temp SSD storage

• E4as v4: AMD processor variant (often slightly cheaper)

The AMD variants (“a” in the name) typically cost 5-10% less than Intel equivalents while offering comparable performance for most workloads.

Azure pricing varies by region, and the differences can be meaningful. UK South and UK West don’t always have identical pricing, and regions further afield can vary more substantially. For UK-based organisations, UK South is typically the primary choice for production workloads due to its broader service availability. UK West often serves as a secondary region for disaster recovery purposes.

As a rough guide, European regions tend to price similarly, while US regions are sometimes slightly cheaper. However – data residency requirements, latency, and compliance considerations usually outweigh small pricing differences for UK businesses. Running workloads in the US to save a few percent rarely makes sense when you factor in GDPR implications and network latency.

That said, for non-production workloads without data residency requirements, it might be worth checking whether a different region offers better value, particularly for dev/test environments.

Storage costs catch people out more often than you might expect. Every VM needs disks, and the pricing varies dramatically based on what you choose.

Managed disk tiers

Azure offers several managed disk tiers, each with different performance characteristics and price points:

• Standard HDD is the cheapest option, suitable for dev/test, backups, and infrequently accessed data. Performance is limited (up to 500 IOPS), so these aren’t suitable for production workloads with any meaningful I/O requirements.

• Standard SSD offers better performance than HDD at a modest premium, with lower latency. These work well for web servers and lightly used applications.

• Premium SSD is the go-to for production workloads. A P10 (128GB) costs around £18/month and delivers up to 500 IOPS. Larger Premium SSDs offer higher IOPS: a P30 (1TB) provides 5,000 IOPS for around £124/month. Most production VMs should use Premium SSDs for the OS disk at minimum.

• Premium SSD v2 lets you provision IOPS and throughput independently of capacity, which can be more cost-effective for workloads with specific performance requirements. You pay separately for capacity, IOPS, and throughput.

• Ultra Disks are for the most demanding workloads, offering up to 160,000 IOPS and 4,000 MB/s throughput. Pricing is based on provisioned capacity, IOPS, and throughput. These suit SAP HANA, top-tier databases, and transaction-heavy workloads.

OS disks vs data disks

When you create a VM, an OS disk is created automatically. For Windows VMs, the default is typically a 127GB Premium SSD. For Linux, it’s often smaller.

Data disks are additional disks you attach for application data. You’re charged for these based on the provisioned size and tier, regardless of how much data you actually store on them.

A common mistake is provisioning a 1TB Premium SSD “just in case” when you only need 100GB of storage. You’ll pay for the full 1TB. It’s better to start smaller and resize when needed (which you can do without downtime for most disk types).

Snapshots and images

Disk snapshots are billed based on the actual data stored (not the provisioned disk size) at roughly £0.04/GB/month for Standard storage. If you’re taking regular snapshots for backup purposes, these costs can accumulate. Make sure your snapshot retention policy matches your actual recovery requirements.

Operating system licensing is a big component in VM costs, and the difference between Windows and Linux can be substantial.

Linux VMs don’t include an OS licence cost, so they’re cheaper than equivalent Windows VMs. The Windows Server licence typically adds 30-40% to the base compute cost, depending on the VM size.

For organisations with existing Windows Server licences with active Software Assurance, Azure Hybrid Benefit lets you bring those licences to Azure. This effectively gives you Windows VMs at Linux prices, saving up to 40% on compute costs.

Azure Hybrid Benefit can also apply to SQL Server licensing if you’re running SQL Server on Azure VMs. The combined savings from Windows Server and SQL Server Hybrid Benefit can be substantial. If you’re migrating from on-premises infrastructure and have qualifying licences, make sure you’re taking advantage of this.

For workloads that don’t specifically require Windows, Linux VMs offer straightforward cost savings. Many web applications, containerised workloads, and development environments run perfectly well on Linux at a lower price point.

Given the significant cost differences, GPU VMs deserve closer attention if you’re considering them.

GPU pricing reflects both the VM compute resources and the GPU hardware. Costs vary widely depending on the GPU type and quantity:

• NC-series (T4 GPUs): Entry point for ML inference and smaller training jobs. Expect to pay several hundred pounds per month for a single T4 GPU configuration.

• ND-series (A100 GPUs): High-end deep learning. The larger configurations with multiple A100s can cost £20,000+/month.

• NV-series (visualisation): Designed for graphics workloads rather than compute. Pricing sits between NC and ND series.

For virtual desktop infrastructure (VDI) with graphics requirements, the NV-series is designed for visualisation rather than compute. These work well for users who need GPU acceleration for CAD applications, video editing, or other graphics-intensive work.

Spot pricing (covered below) can dramatically reduce GPU VM costs for interruptible workloads like batch training jobs, often by 60-80% compared to pay-as-you-go rates. Check Microsoft’s pricing page for current GPU VM pricing, as these tend to fluctuate more than standard compute.

Azure has several ways to pay for VM compute. Choosing the right model can reduce costs by 70% or more compared to default pricing.

Pay-as-you-go

The default pricing model charges by the second (billed hourly) with no commitment. A D2s v3 running Windows at pay-as-you-go rates costs around £115/month if it’s running continuously.

Pay-as-you-go gives you maximum flexibility: spin up a VM when you need it, delete it when you don’t, and pay only for what you use. This suits variable workloads, testing, and situations where you can’t predict usage.

The downside is cost. Pay-as-you-go is the most expensive option for VMs that run continuously.

Reserved Instances

If you know you’ll need a VM for at least a year, Reserved Instances offer some pretty big savings. You commit to a specific VM size in a specific region for 1 or 3 years, and in return you get a discounted rate.

Typical savings:

• 1-year commitment: around 13% discount vs pay-as-you-go

• 3-year commitment: around 25% discount vs pay-as-you-go

A D2s v3 that costs £115/month at pay-as-you-go drops to around £86/month with a 3-year savings plan, saving nearly £350/year on a single VM.

Reserved Instances are ideal for production workloads with predictable, steady-state requirements. They’re less suited to workloads that might need to scale up or down significantly or might be retired before the term ends.

Azure Savings Plans

Savings Plans are an alternative to Reserved Instances that offer more flexibility. Instead of committing to a specific VM size, you commit to a certain amount of compute spend per hour (e.g., £10/hour). That commitment applies across any VM in any region, as long as you’re using eligible compute services.

Based on current UK South pricing, Savings Plans offer around 13% savings for 1-year commitments and around 25% for 3-year commitments. Reserved Instances can offer slightly higher discounts for specific VM sizes, but Savings Plans provide more flexibility if your VM requirements might change.

Azure Spot VM Pricing

Spot VMs let you access unused Azure capacity at steep discounts, often 80-90% below pay-as-you-go rates. The catch is that Azure can evict your VM with 30 seconds’ notice when it needs the capacity back.

A D2s v3 that costs £115/month at pay-as-you-go drops to around £14/month as a Spot VM, an 87% saving. The exact price fluctuates based on demand for that VM size in that region. Spot VMs suit workloads that are:

• Interruptible: Can handle being stopped and restarted

• Stateless: Don’t rely on local state that would be lost on eviction

• Flexible on timing: Don’t need to complete by a specific deadline

Common use cases include batch processing, CI/CD build agents, dev/test environments, and machine learning training jobs with checkpointing. They’re not suitable for production web servers, databases, or anything that needs continuous availability.

Dev/Test pricing

If you have a Visual Studio subscription (or your organisation has one), Azure offers discounted rates for development and testing. Windows VMs in dev/test subscriptions don’t incur the Windows licence cost, effectively giving you Windows at Linux prices without needing Azure Hybrid Benefit.

This applies to dev/test environments only, not production workloads. It’s an easy saving to miss if you’re not aware of it.

Azure’s pricing calculator helps you estimate costs before deploying resources. It’s worth spending time here before committing to infrastructure.

To estimate VM costs:

1. Add “Virtual Machines” from the product list
2. Select your region (UK South for most UK workloads)
3. Choose the operating system (Windows or Linux)
4. Select the VM tier/size you’re considering
5. Choose your pricing model (pay-as-you-go, reserved, etc.)
6. Specify how many hours per month the VM will run

Don’t forget to add:

• Managed disks for OS and data storage
• Any additional storage or backup requirements
• Networking costs if you expect significant data transfer
• Licensing for SQL Server or other software if applicable

The calculator gives you a monthly estimate you can use for budgeting. It’s also useful for comparing options, like seeing the difference between a D4s v3 and an F4s v2 for your workload, or the savings from a 3-year reservation vs pay-as-you-go.

One tip: the calculator defaults to showing pay-as-you-go pricing. Make sure you toggle to reserved pricing if that’s what you’re planning to use, or you’ll overestimate your costs.

Having helped numerous organisations migrate to Azure and optimise their VM spend, we’ve seen certain patterns come up repeatedly:

• Over-provisioning VMs: Picking a larger size “just to be safe” and never revisiting. A D8s v3 at 15% utilisation wastes ~£180/month vs a correctly-sized D2s v3.

• Forgetting disk costs: Provisioning a 1TB Premium SSD you don’t need costs £95/month regardless of actual usage. Also watch for orphaned disks after VM deletion.

• Not using Azure Hybrid Benefit: If you have Windows Server or SQL Server licences with Software Assurance and aren’t applying Hybrid Benefit, you’re paying twice. Savings can exceed 40%.

• Running dev/test VMs 24/7: Dev VMs idle overnight and weekends account for a big portion of running hours. Schedule shutdowns or use DevTest Labs.

• Ignoring Reserved Instance opportunities: Pay-as-you-go is the default, but VMs running continuously for years should be on reserved capacity.

• Not monitoring utilisation: Without usage metrics, you can’t know if VMs are right-sized. Azure Monitor and Advisor provide recommendations, but only if you check them.

Now you’re aware of those common mistakes, here are some proactive steps to keep VM costs under control:

• Right-size using real data: Use Azure Monitor to track utilisation. VMs consistently under 40% are strong candidates for downsizing.

• Auto-shutdown non-production VMs: Schedule dev/test VMs to stop outside business hours.

• Use Spot VMs for interruptible workloads: Batch processing and training jobs can often tolerate eviction in exchange for 60-90% savings.

• Match disk tiers to requirements: Not every disk needs Premium SSD. Dev VMs and infrequently accessed storage can use Standard SSD or HDD.

• Tag resources and track costs: Use tags to categorise VMs by project, team, or environment. Azure Cost Management can break down spend by tag and alert you to unexpected increases.

• Review reserved capacity quarterly: Make sure your reservations still match your needs. Azure allows exchanges within the same VM family if your requirements have shifted.

• Consider VM Scale Sets: For variable workloads, auto-scaling beats provisioning for peak load around the clock.

Azure VM pricing has a lot of moving parts, but most organisations can reduce their spend a lot through right-sizing, reserved capacity, and better management of non-production environments. Build a habit of reviewing costs quarterly, and you’ll catch opportunities for savings before your costs compound.

If you’d like help understanding your Azure VM costs or want a second opinion on your current setup, get in touch with the Synextra team today.