Many organizations rely on On-Demand instances for their flexibility, but that flexibility comes at a premium. To reduce costs, cloud providers offer long-term discount instruments like Reserved Instances, Savings Plans, and committed use discounts. They reward steady usage but require commitment, something not every workload can guarantee.
That’s where Spot Instances come in. They offer deep discounts without long-term lock-in, but the tradeoff is unpredictability. The risk of sudden interruptions makes them difficult to trust for production workloads.
Teams are often left guessing which workloads can safely run on Spot, how to manage interruptions, and whether different clouds treat Spot behavior the same way. Without a clear understanding of how Spot Instances work across AWS, Google Cloud, and Azure, it’s easy to either underutilize their value or overcommit and run into stability issues.
This guide breaks down how Spot Instances function in each major cloud provider, what tradeoffs you need to consider, and how to use them effectively without adding operational risk.
What Are Spot Instances?
Spot Instances are deeply discounted compute resources that help businesses reduce cloud costs for workloads that can tolerate interruptions. Offered by all major cloud providers, they provide the same performance as regular compute instances but at a significantly lower price, because they can be reclaimed at any time with limited notice.
This tradeoff makes Spot Instances ideal for stateless, fault-tolerant, and flexible workloads such as batch processing, CI/CD pipelines, container orchestration, and data analysis. They are not recommended for long-running or critical applications where uptime and stability are essential.
Here’s how Spot Instances work across AWS, Google Cloud, and Azure:
AWS Spot Instances
AWS EC2 Spot Instances provide users with the ability to save up to 90% on cloud resources when compared to AWS On-Demand pricing. EC2 Spot Instances come with the same performance capabilities as traditional compute instances and can be launched via the AWS Console, EC2 Fleet, or Auto Scaling groups.
AWS may reclaim capacity with just a two-minute warning, after which the instance is stopped, hibernated, or terminated based on user settings.
Google Cloud Preemptible VMs and Spot VMs
Google Cloud offers its users the opportunity to use Preemptible VMs and Spot VMs to lower their cloud resource costs by 60-91%. Both of these provisions offer discounted spare Compute Engine capacity in exchange for the possibility of interruption with very little notice.
Preemptible VMs are the legacy predecessor to Spot VMs and have a maximum runtime of 24 hours with a cancellation notice of only 30 seconds. Spot VMs, on the other hand, have the same cancellation window but without the fixed runtime limit.
Azure Spot Virtual Machines
Azure Spot Virtual Machines function similarly to AWS and Google Cloud Spot Instances. Depending on factors like region or VM size, Azure users can use Spot VMs to save up to 90% on cloud resources compared to standard On-Demand rates.
To purchase, users set a maximum bid price. The instance provisions automatically if capacity is available and the current Spot price is at or below the user’s bid.
Azure provides 30-second eviction notices if capacity becomes unavailable or Spot pricing exceeds the maximum bid. Similar to AWS, users can configure an eviction policy to deallocate or delete the VM upon eviction.
Benefits of Spot Instances and VMs
While Spot Instances come with some reliability tradeoffs, they offer several cost and flexibility advantages worth considering:
Reduce cloud costs
Spot Instances can cut compute spend by as much as 90 percent compared to On-Demand pricing, helping lower the overall cost of running cloud infrastructure. Businesses can then combine these savings with other cost optimization initiatives to help them create a more agile and cost-efficient cloud infrastructure.
Accelerate experimentation and testing
Spot Instances and VMs offer businesses a cost-effective way to create temporary testing environments for cloud applications. Instead of committing to long-term, full-cost provisioning, businesses can leverage Spot Instances to accelerate development cycles and test new software features without heavily impacting project budgets.
Avoid long-term commitments
Spot Instances and VMs are available to businesses flexibly and without requiring any multi-year commitment contracts. This gives businesses the flexibility to leverage deep discounts on their cloud consumption when it is most advantageous to them, avoiding lock-in risks and the need to plan out long-term capacity needs.
How Do Spot Instances Work?
Spot Instances operate on surplus compute capacity that is temporarily available across a provider’s infrastructure. Cloud platforms continuously monitor usage and allocate this idle capacity to Spot users at discounted rates. But because this availability fluctuates, Spot Instances are inherently interruptible.
When a request is made, the cloud providers check if capacity is available and if the current spot price aligns with the user’s conditions (like maximum bid price in Azure). If so, the instance is provisioned at a discount.
Once running, the instance behaves like any standard virtual machine, until the provider needs to reclaim resources. At that point, the system issues an eviction notice, often just 30 seconds to 2 minutes in advance, after which the instance is forcefully stopped or deleted based on the user’s chosen policy.
This temporary nature requires careful orchestration. Spot workloads need to be interruption-tolerant and typically rely on automated tools to manage retries, redistribution, and shutdowns.
What Are the Use Cases of Spot Instances?
Although Spot Instances can provide businesses with considerable cloud compute savings, their unreliable and temporary nature means they’re not suited for all cloud activities.
Below are some of the most common ways businesses leverage these types of instances:
- Batch processing jobs: Spot Instances are well-suited for running interruptible batch processing tasks, such as financial modeling, scientific simulations, or media encoding.
- Big data and analytics: Distributed data frameworks like Hadoop and Apache Spark can handle independent node failures, making Spot Instances an ideal solution for processing more extensive datasets and running analytical workloads.
- Containerized workloads: When running container orchestration platforms like Kubernetes and Amazon ECS, Spot Instances enable businesses to scale applications more cost-effectively, as the orchestrator manages potential interruptions caused by unplanned evictions and reschedules containers accordingly.
- Machine learning model training: For less time-sensitive machine learning tasks, Spot Instances can be combined with checkpointing to save progress and resume training in the event of an interruption.
- CI/CD pipelines and testing environments: Spot Instances create a cost-efficient way for DevOps teams to spin up and tear down temporary build, test, and deployment environments.
How Are Spot Instances Charged?
Cloud providers typically charge Spot Instances by the second or minute of usage. The rate fluctuates based on the real-time supply and demand of spare computing capacity, up to the maximum price businesses agree to pay.
The price remains below the On-Demand rate but can fluctuate during the instance’s lifecycle. You’re only charged while the instance is running. Billing stops when you terminate the instance or when the provider evicts it due to rising demand or price thresholds being exceeded.
Spot Instance Comparison Across AWS, Google Cloud, and Azure
Depending on your primary cloud provider, Spot Instances or VMs may have different distinctions in how they’re structured, purchased, and implemented.
Below, we’ll compare AWS, Google Cloud, and Azure while highlighting these key differences:
| Feature | AWS Spot Instances | Google Cloud Spot VMs | Azure Spot Virtual Machines |
| Name | Spot Instances | Spot VMs and Preemptible VMs | Spot Virtual Machines |
| Applies To | EC2 only | Compute Engine only | Virtual Machines only |
| Max Discount Potential | Up to 90% Off | Up to 60-91% Off | Up to 90% Off |
| Availability & Capacity | Subject to AZ capacity with tools for checking availability info | Based on excess Compute Engine capacity; varies by region/zone | Based on Region, VM size, and Demand |
| Interruption/Eviction Notice | 2 minutes | 30 seconds | 30 seconds |
| Eviction Policy Options | Terminate, Stop, or Hibernate | Terminate | Deallocate or Delete |
| Runtime Limit | No limit | No fixed limit for Spot VMs; 24-hour limit existed for Preemptible VMs | No fixed runtime limit |
| Service Integration | EKS, Auto Scaling Groups | GKE, Managed Instance Groups, | AKS and VM Scale Sets (VMSS) |
How To Buy Spot Instances
AWS, Google Cloud, and Azure give their users multiple ways to purchase and manage their Spot Instances:
- AWS
AWS Spot Instances are purchaseable through the AWS Management Console, AWS Command Line Interface (CLI), AWS SDKs, or by using EC2 Auto Scaling Groups or EC2 Spot Fleet. For detailed information, view the AWS Spot Instances documentation.
- Google Cloud
Spot VMs are purchasable using the Google Cloud Console, gcloud CLI, or Compute Engine API. For detailed information, view the Google Cloud Spot VMs documentation.
- Azure
Spot Virtual Machines are purchaseable using the Azure Portal, Azure CLI, Azure PowerShell, or Azure Resource Manager (ARM) templates. For detailed information, view the Azure Spot Virtual Machines documentation.
Best Practices for Purchasing Spot Instances
To maximize the value of Spot Instances in your cloud environment, it’s essential to understand when and how to utilize them. Below are some best practices you can use when purchasing Spot Instances:
Evaluate workload suitability
Before you purchase a Spot Instance, it’s important to assess whether your cloud workloads are the right fit. Spot Instances are regularly interrupted without much notice, so you’ll want to reserve them for tasks that can handle these disruptions.
Focus on using applications designed with fault tolerance, such as batch processing jobs or stateless applications that don’t rely on preserving data. Reserve your mission-critical workloads that require constant uptime for more reliable On-Demand instances.
Set up auto-scaling and workload resilience
To minimize operational disruptions caused by early Spot Instance evictions, implement auto-scaling and redundancy strategies. Leverage tools like Azure VMSS, Managed Instance Groups, or Auto Scaling Groups to distribute your workloads across multiple instances and Availability Zones.
These help to ensure you can quickly replace any terminated Spot Instances without impacting application performance.
Monitor and track Spot Instance availability
Use available cloud provider resources, such as the AWS Spot Instance Advisor or historical zone pricing and eviction pattern data, to continuously track Spot Instance availability trends.
By understanding where instance interruptions or evictions are more likely to occur, you can make more informed decisions about when and where to deploy your Spot Instances for maximum run time.
Establish a backup plan
When leveraging Spot Instances, it is essential to establish a reliable backup strategy. A proven approach is to blend different discount options based on workload needs. For example, use EC2 Instance Savings Plans or Standard RIs for consistent environments, opt for Compute Savings Plans or Convertible RIs when flexibility is needed and Spot Instances for bursty, fault-tolerant tasks like batch processing or containerized jobs.
This blended strategy balances cost and availability, while maintaining resilience.
Automatically Optimize Your Cloud Costs With ProsperOps
Spot Instances are great for lowering compute costs, but they’re inherently unreliable. When evicted with minutes/seconds of sudden notice, your workloads fall back to On-Demand pricing and this shift can cause sudden spikes in your bill.
This is where ProsperOps comes in.
ProsperOps delivers cloud savings-as-a-service, automatically blending discount instruments to maximize your savings while lowering Commitment Lock-In Risk. Using our Autonomous Discount Management platform, we optimize the hyperscaler’s native discount instruments to reduce your cloud spend and place you in the 98th percentile of FinOps teams.
- Weekly commitment purchases that adjust to real-time usage
- Multi-cloud support across AWS, Microsoft Azure and Google Cloud
- Portfolio diversification across Savings Plans, Reserved Instances and CUDs
- Continuous laddering that adapts commitment amounts and timing with no manual intervention
In addition to autonomous rate optimization, ProsperOps now supports usage optimization through its resource scheduling feature, ProsperOps Scheduler. Our customers of Autonomous Discount Management™ (ADM) can now automate resource state changes on weekly schedules to reduce waste and lower cloud spend.
Your teams stay focused on strategic FinOps goals, while ProsperOps automates rate and usage optimization behind the scenes.
Make the most of your cloud spend with ProsperOps. Schedule your free demo today!
