Cloud computing gives teams more flexibility than ever, resources can be scaled up or down as needed, and new workloads can be launched in seconds. But that flexibility comes with a cost. As environments grow more complex, keeping cloud costs in check has become harder, especially when resources are overprovisioned or left running longer than needed.
According to the FinOps Foundation’s 2025 State of FinOps Report, over 50% of practitioners ranked waste reduction as their top priority, underscoring just how widespread and persistent cloud cost inefficiency has become.
To address this, teams are increasingly looking beyond traditional discount models and turning to more dynamic cost-saving levers.
One such option for compute-heavy, fault-tolerant workloads is Azure Spot Virtual Machines. They offer a way to significantly reduce compute costs without sacrificing scalability when used in the right scenarios.
In this guide, we’ll break down where Spot VMs fit into your Azure cost strategy, how they’re priced, ideal use cases, and tips to get the most out of them.
What Are Azure Spot Virtual Machines (VMs)?
Azure Spot VMs (Virtual Machines), also called Azure spot instances, are deeply discounted compute resources that let you take advantage of Azure’s unused capacity. Offered at up to 90% less compared to pay-as-you-go prices, Spot VMs are a cost-effective option for non-critical, interruptible workloads.
The trade-off: these VMs can be evicted at any time based on Azure’s capacity needs. As overall demand increases, Azure may reclaim that capacity with minimal notice of 30 seconds, which means Spot VMs are not suitable for workloads that require guaranteed availability.
They’re best used for fault-tolerant tasks like batch jobs, testing, stateless applications, or machine learning training, where occasional disruption won’t derail performance or outcomes.
Benefits of Using Azure Spot VMs
Azure Spot VMs offer many practical benefits for businesses as they plan their cloud deployments, including:
Cost savings
For businesses focused on cost efficiency, Azure Spot VMs are an effective way to greatly reduce cloud costs. Spot VMs can save cloud users up to 90% when compared to Pay-As-You-Go (PAYG) pricing — one of the highest discounted rates available on the Azure platform.
For flexible and interruptible workloads, Azure Spot VMs can be an incredible Azure cost optimization strategy to get started with.
Scalable for variable workloads
Spot VMs are a good fit for workloads with fluctuating demand. Instead of committing to fixed capacity through VM Reservations, teams can scale Spot VMs up or down as needed, paying only for the time used. This flexibility is especially useful during traffic spikes, simulations, or short-term projects where cost control matters as much as performance.
Flexible pricing
Since Spot VM pricing changes based on real-time capacity and demand, Azure allows you to set a maximum price to prevent cost overruns. If the current rate exceeds your set threshold, the instance is evicted, giving you control over how much you’re willing to spend, while still taking advantage of deep discounts when available.
How Do Azure Spot VMs Work?
Azure Spot VMs are built on the idea of using Azure’s surplus compute capacity. Microsoft Azure continuously monitors the usage of cloud services across its global infrastructure and over time, host machines on specific servers will have idle resources that go unused. When this excess becomes available, Azure temporarily offers them at significantly reduced prices (up to 90%) less than standard Pay-As-You-Go (PAYG) rates.
However, since these are excess resources, Azure can reclaim them at any time. That’s why Spot VMs come with no SLA and are suitable only for workloads that can tolerate interruptions.
Here’s how the provisioning and eviction process works:
Requesting a Spot VM
To acquire a Spot VM, users initiate a request using the Azure portal, CLI, or API. During setup, you define the virtual machine’s size, region, and the maximum price you’re willing to pay. Azure compares your max price with the current spot market price. If your price is equal to or greater than the current rate and capacity is available, Azure will deploy the VM at the discounted rate. If your price is lower, the VM won’t be deployed, and you’ll receive an error.
Max price behavior
The maximum price setting determines how Azure provisions and maintains your Spot VM. If you set your max price to -1, Azure interprets this as a willingness to pay up to the PAYG rate, but never more. This ensures price protection without unexpected charges.
If at any point the current spot price exceeds your specified max, Azure will evict the VM with a short warning (usually 30 seconds). Evicted VMs aren’t restarted automatically, you’ll need to redeploy them manually at the current spot rate.
Source: Azure Documentation
Eviction policy
Azure may evict Spot VMs with 30 seconds’ notice when it needs capacity for PAYG workloads or when the current spot price exceeds your max price. When creating a Spot VM, you can define an eviction policy: either Deallocate (default) or Delete. The Deallocate option stops the VM and allows you to redeploy it later, but with no guarantee of success. You’ll continue to be charged for disk storage, and the deallocated VM will still count against your quota.
The Delete policy, on the other hand, removes the VM and its underlying disks entirely, stopping all associated charges. You can also enable scheduled event notifications, which attempt to notify you up to 30 seconds before eviction.
Restarting and updating VMs
If a Spot VM is stopped or deallocated, it can only be restarted if your max price still meets or exceeds the current market price and there’s available capacity. Otherwise, Azure will block the restart attempt. If you need to change the max price of an existing Spot VM, you must first deallocate the VM, update the price setting, and then start the instance again.
For a full breakdown of Spot VM pricing logic, check out Azure’s official documentation.
Use Cases for Azure Spot VMs
Although Spot VMs are a great way to save on compute costs, you may want to be selective when deciding where and how to use them. Below are some of the optimal use cases where Spot VMs are most beneficial:
Batch processing and high-performance computing
Microsoft Azure Spot VMs are a great choice when handling batch jobs or compute-intensive tasks like financial modeling or scientific simulations. These processes are typically designed with inherent fault tolerances, making intermittent resource availability more manageable.
For example, many scientific simulations run smaller, independent computing tasks. If interruptions occur, businesses can restart or redistribute these tasks without causing significant disruptions.
Likewise, financial models often make use of checkpointing, allowing computations to be periodically saved during longer calculations. In this case, if evicted from a Spot VM, the model can resume from the last saved checkpoint, minimizing lost time.
Machine learning model training
Many businesses are leveraging cloud computing power to build and test new AI-powered applications, and Spot VMs are a cost-effective solution for training new machine learning (ML) models.
ML developers rely on hyperparameter tuning to run independent training experiments. Since each experiment can tolerate interruptions, Spot VMs are a great option.
Distributed training is another common tactic for ML development. With training workloads spread across multiple virtual machines, minor interruptions are easier to manage. Because of this, Spot VMs are an ideal choice for large-scale ML model training.
Containerized applications using Kubernetes (AKS)
Businesses running Azure Kubernetes Services (AKS) can benefit from increased cost optimization when running various web services or background application tasks. Instead of relying on one type of virtual machine, organizations can group their VMs into node pools.
Spot node pools allow businesses to manage all of their Spot VMs as one unified group. They can then specify various eviction policies, including whether to deallocate, delete, or reassign workloads as VMs resources become unavailable.
AKS seamlessly integrates these Spot node pools into existing clusters, providing a flexible way to mix standard and Spot VMs. This ensures that Spot VMs are only reserved for interruption-tolerant workloads, while other critical applications and services run on standard VM node pools.
CI/CD pipeline workloads
The interruptable nature of Spot VMs aligns well with Continuous Integration and Continuous Deployment (CI/CD) pipelines. Not only can they provide a much more cost-effective environment for building, testing, and deploying new software, but they’re also highly flexible when running parallel tasks.
For example, when running automated testing suites, businesses can use Spot VMs to distribute their workloads. If an eviction impacts a workload, CI/CD systems will identify a test failure and simply reassign tests to another Spot VM.
And because Spot VMs are so cost effective, businesses can deploy significantly more VMs at once. This provides additional flexibility and available capacity, letting development teams run quicker tests and create faster feedback loops.
What Are the Restrictions of Azure Spot Virtual Machines?
When setting up Spot VMs, there are certain restrictions and limitations you should be aware of, including:
- Unsupported VM Sizes: Spot VMs don’t allow for B-series VM sizes. In addition, all Dv2, NV, NC and H promo VM sizes won’t be eligible for Spot VM discounts.
- Limited regional availability: Although Azure users have a wide range of region availability choices when provisioning a Spot VM, they aren’t applicable in all areas. For example, 21Vianet, which operates Microsoft Azure in China, does not support Spot VMs.
It’s also important to remember that Spot VMs’ regional availability will change frequently, impacting pricing and increasing the likelihood of evictions.
Azure Spot VMs vs. VM Reservations
Both Spot VMs and VM Reservations are useful for reducing Azure compute costs. However, each applies different discount mechanisms.
Spot VMs offer the highest discounted compute rates, but they trade off these savings with less reliability, anytime eviction and no service guarantees. VM Reservations, on the other hand, offer lower but predictable discounts (up to ~72%) in exchange for a one- or three-year commitment.
Depending on your specific workload requirements, either solution can be a great way to add significant cost savings to your cloud deployments. Here’s how you can differentiate between the two, helping you choose the right fit:
| Feature | Azure Spot VMs | Azure VM Reservations |
|---|---|---|
| Pricing | Variable costs with up to 90% savings versus pay-as-you-go model | Fixed, discounted pricing with up to 72% savings versus pay-as-you-go models |
| Availability | Interruptible, no availability guarantees | Guaranteed reserved capacity and high availability |
| Commitment | No commitment required | One- or three-year term commitments |
| Eviction rate | Evictable at any time, 30 seconds minimum notice | Not Applicable |
| Best use case | Batch processing, ML model training, CI/CD testing | Production workloads, cloud databases, core applications |
Automatically Optimize Azure Costs With ProsperOps
If you’re already using Spot VMs to cover interruption-tolerant workloads, the next logical step is expanding savings across your predictable compute usage. Azure VM Reservations and Savings Plans offer long-term discounts, but unlike your usage, these commitments are static.
To avoid wasted spend from overcommitting or missed savings from undercommitting, these instruments require continuous monitoring and active management.
That’s where ProsperOps can help. We take out the headache of manual processes and help you save money automatically with cloud-savings-as-a-service.
With the Autonomous Discount Management platform, we help you optimize Microsoft Azure’s native discounts to reduce your cloud spend and place you in the 98th percentile of FinOps teams. Our platform setup is quick, and our systems work behind the scenes to optimize your cloud costs. This allows your teams to concentrate on innovation and growth, while we automate cloud cost optimization for you.
To see ProsperOps in action, book a demo today.
