Spot & Preemptible Instance Pricing Explained (2026)
Compare spare capacity pricing across AWS Spot, Azure Spot VMs, and GCP Spot/Preemptible VMs. Up to 90% discount with interruption risk.
TL;DR
- • AWS, Azure, and GCP offer spot/preemptible VMs at 60-91% off, but instances can be reclaimed with 30 seconds to 2 minutes notice.
- • DigitalOcean, Hetzner, Linode, and Vultr have no spot pricing — but their on-demand rates ($0.007-0.071/hr) are often cheaper than Big 3 spot prices.
- • Use spot for batch jobs, CI/CD, and stateless containers. For steady workloads, compare Big 3 spot prices against smaller providers' fixed rates.
Spot instances (called Spot VMs on GCP/Azure) let you use spare cloud capacity at discounts of up to 91%. The tradeoff: the cloud provider can reclaim these instances with minimal notice. This guide compares how all 8 providers handle spare-capacity and low-cost compute, including the smaller providers where fixed pricing is often more competitive than Big 3 spot rates.
Side-by-Side Comparison
| Feature | AWS | Azure | GCP | DO | Hetzner | Linode | OCI | Vultr |
|---|---|---|---|---|---|---|---|---|
| Service name | Spot Instances | Spot VMs | Spot VMs (formerly Preemptible) | N/A | N/A | N/A | Preemptible Instances | N/A |
| Max discount | Up to 90% | Up to 90% | Up to 91% | — | — | — | Up to 50% | — |
| Interruption notice | 2 minutes | 30 seconds | 30 seconds | — | — | — | 30 seconds | — |
| Max runtime | No limit | No limit (eviction-based) | No limit (Spot) / 24hr (Preemptible) | — | — | — | No limit | — |
| Pricing model | Dynamic (market-based) | Dynamic (capped at on-demand) | Fixed discount from on-demand | — | — | — | Fixed 50% off on-demand | — |
| Price guarantee | None (varies by supply/demand) | Set max price or use default | Fixed at 60-91% off on-demand | — | — | — | Fixed 50% off on-demand | — |
| Container support | Yes (ECS, EKS, Fargate Spot) | Yes (AKS) | Yes (GKE) | — | — | — | Yes (OKE) | — |
| Comparable on-demand (4 vCPU) | — | — | — | $0.071/hr fixed | $0.021/hr fixed | $0.072/hr fixed | — | $0.071/hr fixed |
How AWS Spot Instances Work
AWS Spot Instances use a dynamic pricing model based on supply and demand. Prices fluctuate but are typically 60-90% below on-demand. When AWS needs the capacity back, you receive a 2-minute interruption notice via instance metadata and CloudWatch Events. You can set a maximum price you're willing to pay. Spot Instances can be used with EC2, ECS (Fargate Spot), EKS, EMR, and SageMaker.
Example: A c5.xlarge (4 vCPU, 8 GB) costs $0.170/hr on-demand in us-east-1. The spot price typically hovers around $0.034-0.051/hr — a 70-80% saving. However, this price fluctuates and the instance can be interrupted. AWS Spot Fleet manages collections of Spot and On-Demand instances to maintain target capacity automatically.
How Azure Spot VMs Work
Azure Spot VMs offer up to 90% discount over pay-as-you-go prices. You set a maximum price (or accept the current price), and Azure evicts the VM when capacity is needed or the market price exceeds your maximum. You get a 30-second notice before eviction. Unlike AWS, Azure provides two eviction policies: deallocate (stop and keep the disk) or delete.
Example: A D4s_v5 (4 vCPU, 16 GB) costs $0.192/hr on-demand in East US. Spot pricing typically runs $0.019-0.058/hr — a 70-90% saving. Spot VMs are available in Azure Virtual Machine Scale Sets and AKS node pools. For ML training workloads, Azure also offers low-priority VMs in Batch pools.
How GCP Spot VMs Work
GCP Spot VMs (which replaced Preemptible VMs) offer a fixed discount of 60-91% off on-demand prices. Unlike AWS and Azure, GCP's Spot pricing is predictable — the price doesn't fluctuate based on demand. However, GCP can still reclaim instances when capacity is needed, with a 30-second warning. The key advantage of GCP Spot: you know the exact price in advance.
Example: An n2-standard-4 (4 vCPU, 16 GB) costs $0.194/hr on-demand in us-central1. The Spot price is fixed at $0.048/hr — a 75% saving. Unlike AWS, this price doesn't change based on demand. GCP Spot VMs work with Compute Engine, GKE, and Dataproc. They have no 24-hour lifetime limit (unlike the older Preemptible VMs).
Oracle Cloud Preemptible Instances
Oracle Cloud offers preemptible instances at a fixed 50% discount from on-demand pricing. These are available for most compute shapes and can be reclaimed when OCI needs the capacity. You receive a 30-second warning before termination. The discount is lower than the Big 3 spot offerings, but OCI's on-demand prices are already lower, so the effective price can be competitive.
Example: An E4.Flex (4 OCPU, 16 GB) costs roughly $0.10/hr on-demand. As preemptible, it drops to $0.05/hr. While the 50% discount is lower than GCP's 75%, the starting price is also lower.
Smaller Providers: No Spot, but Cheap Fixed Pricing
DigitalOcean, Hetzner, Linode, and Vultr do not offer spot or preemptible instances. However, their fixed on-demand pricing is often comparable to or cheaper than Big 3 spot prices. This is an important comparison often overlooked.
Hetzner CX41 (4 vCPU, 16 GB): ~$0.021/hr fixed, no interruption risk. This is cheaper than GCP Spot ($0.048/hr) and comparable to AWS Spot (~$0.034-0.051/hr) — without any risk of interruption. DigitalOcean Premium (4 vCPU, 16 GB): $0.071/hr fixed. Linode Dedicated (4 vCPU, 8 GB): $0.054/hr fixed. Vultr High Performance (4 vCPU, 16 GB): $0.071/hr fixed.
When to use a smaller provider instead of spot: If your workload needs reliable uptime (not fault-tolerant) and you're considering spot just for the price, a Hetzner or DigitalOcean server may be a better option — cheaper and never interrupted.
Best Use Cases for Spot Instances
Batch processing: jobs that can be checkpointed and resumed. CI/CD pipelines: build and test environments that are naturally fault-tolerant. Big data analytics: Spark, Hadoop, and similar frameworks with built-in fault tolerance. Machine learning training: distributed training jobs that can handle node loss. Web applications: stateless services behind a load balancer with auto-scaling. Dev/test environments: non-production workloads where interruptions are acceptable. Containerized workloads: Kubernetes clusters with mixed on-demand and spot node pools.
Spot Instance Best Practices
Diversify across instance types and availability zones to reduce interruption risk. Use auto-scaling groups (AWS) or VM scale sets (Azure) to automatically replace interrupted instances. Implement graceful shutdown handlers that respond to interruption notices. Combine spot with on-demand: run your baseline on on-demand/reserved and burst with spot. Use checkpointing for long-running jobs so work isn't lost.
On AWS, use Spot Fleet or EC2 Fleet for automatic capacity optimization across instance types. On GCP, take advantage of the fixed pricing model for predictable budgeting. On Azure, set a max price slightly above the typical spot rate for better availability. For workloads that can't tolerate any interruption, compare Big 3 spot prices against Hetzner/DO/Linode/Vultr on-demand — you may find fixed pricing is both cheaper and more reliable.
Frequently Asked Questions
What happens when a spot instance is interrupted?
AWS gives 2 minutes notice, Azure gives 30 seconds, GCP gives 30 seconds, and Oracle gives 30 seconds. During this time, your application should save its state and gracefully shut down. The instance is then stopped or terminated depending on your settings. Auto-scaling groups can automatically launch replacement instances.
How much can I save with spot instances?
Savings vary by provider: AWS offers up to 90% off, Azure up to 90%, GCP up to 91%, and Oracle a fixed 50% off. Typical savings range from 60-80%. However, for a 4 vCPU server, Hetzner's fixed $0.021/hr may be cheaper than any Big 3 spot price, with zero interruption risk.
Can I run production workloads on spot instances?
Yes, if your application is designed for fault tolerance. Stateless web services, containerized microservices, and queue-processing workers are good candidates. Use a mix of spot and on-demand instances to ensure minimum capacity. For production workloads that need reliability without fault-tolerance engineering, consider a smaller provider's fixed pricing instead.
Which cloud provider has the most predictable spot pricing?
GCP has the most predictable spot pricing because discounts are fixed (60-91% off on-demand) rather than market-based. Oracle also uses fixed 50% discounts. AWS prices fluctuate based on supply and demand. Azure lets you set a maximum price. For maximum predictability with no interruption risk, smaller providers (Hetzner, DO, Linode, Vultr) offer fixed pricing at comparable rates.
Do DigitalOcean, Hetzner, Linode, or Vultr offer spot instances?
No. None of these providers offer spot or preemptible instances. Their pricing model is simple fixed-rate on-demand pricing. The upside is that your instances are never interrupted. The prices are also competitive with or cheaper than Big 3 spot prices, especially Hetzner which can be 80-90% cheaper than AWS on-demand.