Cut Compute Costs Up To 90% With Azure Spot Instances

When cloud costs spike, compute is often the culprit. Using Azure Spot Instances could cut your compute costs by up to 90%. But Spot VMs come with trade-offs, including unpredictable evictions and capacity constraints. And that makes them tricky to use without the right strategy and visibility.

In this guide, we will share how to make them work for you. We’ll cover how Azure spot pricing actually works, the best use cases (and when to avoid them), and how to turn Spot VM savings into long-term gains.

Stick around, and we’ll also show you how innovative teams at companies like Moody’s, New Relic, and Skyscanner view, understand, and optimize their spot VMs without sacrificing reliability.

What Are Azure Spot Instances?

Spot isn’t a special type of virtual machine. It’s one of several pricing options you apply to standard Azure VMs. The other two main ones are: 

See also: Azure Savings Plans 101: Tips, Tactics, And Best Practices To Apply Now

Microsoft Azure continuously monitors unused compute capacity across its global network of data centers. 

Image: Azure Availability Zones 

When spare capacity is available, the cloud provider offers it at a discounted price through Spot VMs.

These instances are identical in performance to regular VMs. They use the same underlying infrastructure and come in many of the same sizes and types as standard Azure virtual machines.

However, Spot pricing offers the biggest discounts, but without any SLA or availability guarantees.

So, Azure Spot Instances refer to a pricing option that lets you run standard Azure Virtual Machines (VMs) using unused Azure compute capacity — at up to 90% less than the standard (pay-as-you-go) rate.

Yet, there are gotchas and gems you need to know here.

How Azure Spot Pricing Works

Pricing for Azure Spot instances is dynamic. The price you pay can change based on supply and demand across Azure’s global infrastructure. When demand is low, prices drop. When demand rises, prices increase, and so does the risk of eviction.

Here’s a breakdown of how that really works so you can take full advantage of Spot VMs on Azure:

1. You can set a max price (or let Azure decide)

When deploying a Spot VM, you have two pricing options:

Setting a max price gives you control, but keep in mind that even if the current rate is low, you might get outbid when capacity tightens.

2. Evictions can happen at any time

Azure can evict your Spot VM at any point if:

You’ll typically get a 30-second eviction notice before Azure reclaims your Spot VM. Depending on your configuration, the instance will either be deallocated (stopped and preserved) or deleted entirely.

If you’re familiar with AWS Spot Instances, the concept is similar, though AWS provides a longer 120-second notice.

However, you can choose between two eviction policies:

Both approaches require you to design your workloads to tolerate interruptions.

3. Availability and pricing vary by region and VM size

Spot capacity isn’t consistent across the board. Some Azure regions, availability zones, and VM sizes offer more spot capacity than others. It pays to experiment with and monitor where your workloads can run most cost-effectively.

Pros And Cons Of Azure Spot Instances (And What It Means For You)

From Azure’s perspective, it’s a smart way to monetize spare capacity that would otherwise sit idle. 

For cost-conscious engineering teams, it’s an opportunity to run dev/test environments, batch jobs, container workloads, and AI/ML training jobs at a fraction of the normal cost.

Let’s say you’re running a machine learning training job using a standard D4s_v3 VM. On pay-as-you-go, it might cost $0.40/hour. As a Spot VM, the price could drop to $0.08/hour or lower. That’s 80% off. 

But, if Azure suddenly needs that capacity back, your job will be interrupted unless you’ve built in failover, checkpointing, or some other mitigation. You only get 30 seconds to act each time, after all.

Spot VMs are standard Azure VMs running on the same infrastructure. You’re not sacrificing CPU, memory, or networking performance. You’re just paying less for temporary capacity.

Jobs that can be resumed from checkpoints or tolerate interruptions, such as CI/CD pipelines, containerized microservices, or dev/test environments, are perfect candidates for Spot.

Better yet, you can deploy Spot VMs at scale using Azure Virtual Machine Scale Sets (VMSS), Kubernetes node pools in AKS, or automation tools like Terraform and Bicep. This lets you scale up compute power when it’s cheap, and fall back gracefully when capacity dries up.

So, what are some ideal use cases for Azure Spot instances (and when should you avoid them)? 

When To Use Azure Spot VMs (Without Regret)

Spot is a strong fit when speed is optional, but savings are critical. The best scenarios often share a few traits: stateless, interruptible, and distributed. Consider these:

There are also times when the risk outweighs the reward.

When To Avoid Spot VMs

Spot instances are not ideal for anything stateful, sensitive, or latency-dependent, including:

In short, the right workloads for Spot are the ones you’d happily let fail, and then rerun without consequence.

By now, you’ve picked up several actionable best practices for running Azure Spot VMs. You’ve seen the trade-offs and the opportunities. Up next, as promised, we bring it all together. 

How To Use Azure Spot Instances Right (And Save Without Losing Track Of Costs)

This is where your Azure Spot instance strategy meets real-world execution, with the tools and visibility to make it work for your team, workloads, and bottom line. Here’s how high-performing teams get Spot right from day one.

1. Use VM Scale Sets to manage Spot capacity like the pros do

Azure Virtual Machine Scale Sets (VMSS) gives you control over how Spot VMs are deployed, evicted, and replaced. Here’s how:

Overall, VMSS gives you an autoscaling mechanism that’s both cost-effective and engineered for graceful degradation.

2. Run cost-tolerant workloads in mixed node pools on AKS

For containerized workloads, the AKS service supports dedicated Spot node pools. So, use taints and tolerations to:

You can use Azure’s Cluster Autoscaler with eviction-aware policies to automatically reschedule pods onto available capacity.

3. Set smart max prices, and don’t “set it and forget it”

Set your maximum price per hour too high, and you risk overpaying. If you set it too low, your VMs may never start. Here’s how to strike the right balance:

For finance, you’ll also want to pair spot pricing trends with unit cost metrics. This will help you calculate when Spot is saving money, and when it’s not worth the risk. Yet, not all cloud cost management platforms offer proper unit cost metrics. 

Check out our guide to Cloud Unit Economics here for immediately actionable tips, including what to track and which platform does it best for your specific goals.

4. Design for eviction (because it will happen)

Evictions aren’t bugs. They are built into the Spot model, and inevitable. So, design your workloads accordingly:

Your engineers can codify these patterns into Helm charts, Terraform modules, or Bicep templates to make Spot a default, not an exception.

5. Track trends across regions and sizes

Smart teams treat Spot availability as a data problem. And you can, too:

One more thing. Running jobs in lower-demand zones can reduce both price and eviction risk (two birds with one Spot!).

6. Monitor spot usage and efficiency continuously

Cost savings are only real if they last. To help you ensure this, track metrics like:

This enables your team to calculate effective savings and understand trade-offs in real-world scenarios.

Overall, when you implement them with engineering foresight and financial accountability, Azure Spot Instances become a high-leverage tool for saving money and scaling intelligently.

Your Next Move: From Spot Usage To Real Business Impact — Connect The Dots With CloudZero

Azure Spot Instances can deliver huge savings. But without visibility, they can also introduce risk. Your engineers may lose track of where Spot is (or isn’t) being used. Finance might struggle to quantify savings or tie compute costs back to the business.

CloudZero turns your Spot usage into real-time, business-aligned intelligence. 

With CloudZero, you can:

And yes, you’ll catch the moment someone ships a new pipeline without Spot enabled well ahead, not weeks later when the bloated invoice shows up.

Don’t just take our word for it. Leading teams at Moody’s, Expedia, and Skyscanner rely on CloudZero to track Spot savings, benchmark usage across engineering teams, and measure real ROI.

You can, too. Risk-free. to get started. We just helped Upstart cut over $20 million from their AWS bill.

FAQs About Azure Spot Instances

Are Azure Spot Instances always cheaper than on-demand VMs?

Spot Instances are typically up to 90% cheaper than pay-as-you-go pricing. However, the exact savings fluctuate based on regional demand and available capacity.

What happens when an Azure Spot VM is evicted?

Your Spot VM is evicted with a 30-second warning when Azure needs the capacity back. You can choose whether the VM is deallocated (stopped but preserved) or deleted entirely.

Can I use Spot Instances in Azure Kubernetes Service (AKS)?

Yes. AKS supports Spot nodes via dedicated node pools. You can assign stateless or flexible workloads to these pools using Kubernetes taints, tolerations, and affinity rules—while keeping critical workloads on standard nodes.

What kinds of workloads are best for Azure Spot VMs?

Spot is ideal for flexible, fault-tolerant workloads such as CI/CD pipelines, containerized microservices, batch jobs, dev/test environments, and machine learning model training, especially when checkpointing is in place.

How can I monitor Azure Spot pricing and availability?

Azure doesn’t expose spot pricing history, but you can check availability guidance for VM sizes and regions in their documentation. For deeper yet easy-to-digest cost monitoring, CloudZero gives you real-time insight into usage, savings, and anomalies.

Is it worth using Spot Instances for production?

Not usually. They do not offer high availability or persistent state. Use them to supplement production infrastructure, not replace it.

Author: Rachel Whitener

Rachel Whitener is CloudZero’s SEO Content Editor and a content strategist with six years of experience writing for SaaS and tech companies. With a background in brand storytelling, product marketing, and user-centered content writing, she brings clarity and cohesion to complex ideas. Rachel collaborates with experts across cloud and FinOps to create insightful, valuable content for the CloudZero community. She’s also a trained pastry chef.