The AWS Spot Instance Ecosystem - Mature Interruption Management Behind Up to 90% Discounts
AWS Spot Instances offer up to 90% discounts and, with interruption management that combines the 2-minute notice, Spot Placement Score, and Capacity Rebalancing, are adopted even for production workloads. We compare the differences with Azure Spot VMs and GCP Spot VMs from the perspectives of interruption notice lead time, fleet management, and allocation strategies.
Spot Instance Basics and Discount Structure
AWS Spot Instances are a purchase option that lets you use EC2's surplus capacity at up to 90% off on-demand pricing. When first offered in 2009, pricing used an auction format with fluctuating prices, but the pricing model was revised in 2017 to a stable, supply-and-demand-based pricing system. This change dramatically improved price predictability and accelerated adoption for production workloads. The actual discount is determined per capacity pool, that is, per combination of instance type, region, and Availability Zone, and the only figure AWS states officially is the ceiling of up to 90%. Because the level differs widely from pool to pool, the practical approach is not to assume a specific discount at design time but to check the current price of the target pools in the Spot price history. AWS reclaims instances with a 2-minute notice when it needs the capacity. Interruption frequency also varies greatly by instance type and region, so stable operation comes from choosing pools starting from the interruption frequency bands published by the Spot Instance Advisor and combining them with interruption-aware design such as checkpointing and automatic evacuation.
The Depth of Interruption Management Tools
AWS provides a comprehensive toolset for handling Spot Instance interruptions. The EC2 metadata service interruption notice issues a notification 2 minutes before an instance is reclaimed, giving applications time for graceful shutdown. Integration with EventBridge allows Lambda functions to be triggered by interruption notices, automating job evacuation and checkpoint saving. Spot Placement Score pre-evaluates how available a specific instance configuration is in a region or Availability Zone on a scale of 1 to 10, enabling planned selection of placements with lower interruption risk. Capacity Rebalancing proactively detects instances with elevated interruption risk and automatically initiates migration to new Spot Instances. These tools working together in an integrated fashion transform Spot interruptions from operational risks into manageable events.
Fleet Management and Diversification Strategies
EC2 Fleet and Spot Fleet are features for managing fleets that combine multiple instance types, Availability Zones, and purchase options. For Spot, the official documentation (as of August 2026) provides five allocation strategies: price-capacity-optimized, capacity-optimized, capacity-optimized-prioritized, diversified, and lowest-price. AWS recommends price-capacity-optimized, which first narrows the candidates down to pools with ample spare capacity and then picks the cheapest among them, making it the default choice for many workloads such as stateless container applications, microservices, data analytics, and batch processing. For workloads where interruptions are costly, capacity-optimized prioritizes capacity over price, and capacity-optimized-prioritized additionally honors your instance type priority order on a best-effort basis. For large, long-running fleets, diversified spreads instances evenly across all pools to dilute the impact of price swings or capacity exhaustion in any single pool. By contrast, lowest-price looks only at price and ignores capacity, so it carries the highest interruption risk and AWS advises against it. Because lowest-price is still the default in the AWS CLI, the standard practice is to explicitly override it with price-capacity-optimized. Integration with Auto Scaling groups enables automatic launch of replacement instances when Spot Instances are interrupted, maintaining fleet capacity. Mixed instance policies allow specifying the ratio of On-Demand to Spot, securing baseline capacity while scaling out with Spot.
Comparison with Azure Spot VMs
Azure Spot VMs reached general availability in 2020. While matching AWS in offering up to 90% discounts, they differ in how you prepare for interruptions and in the granularity of fleet management. Azure Spot VM eviction policies offer two choices, stop/deallocate or delete, and the eviction notice arrives 30 seconds in advance via Scheduled Events. Since this lead time is shorter than AWS's 2-minute notice, you need to design so that safely draining containers or writing out large checkpoints fits within 30 seconds. For fleet management, Azure Compute Fleet is available in addition to the traditional Virtual Machine Scale Sets (VMSS). According to the official documentation (as of August 2026), Compute Fleet can launch up to 10,000 VMs (or 100,000 vCPUs) per fleet across up to 3 regions through a single API, mixing Spot and pay-as-you-go VMs, and lets you choose an allocation strategy for both Spot and Standard from lowest-price-focused, capacity-focused, or a combination of the two. It also supports automatic replacement of evicted Spot VMs, attribute-based VM selection, and bundling of purchase options including Reserved Instances and Savings Plans, so configurations close to AWS's EC2 Fleet can now be built. Where the two differ is the granularity of strategies and the means of pre-evaluation: on the AWS side you can go as far as priority-aware strategies such as capacity-optimized-prioritized, and Spot Placement Score lets you estimate in advance how easy each instance configuration is to obtain as a score from 1 to 10. Compute Fleet's strength is handling mixed purchase options within a single fleet, so which is more advantageous depends on whether the workload needs accurate interruption prediction or the bundling of purchase options.
Comparison with GCP Spot VMs
GCP rebranded its former Preemptible VMs as Spot VMs in 2022, removing the 24-hour maximum runtime limitation. GCP Spot VM interruption notice is 30 seconds, shorter than AWS's 2-minute notice for response time. While Managed Instance Groups (MIG) can manage Spot, there is no mechanism for choosing among allocation strategies the way EC2 Fleet offers. As for Sustained Use Discounts (SUD), which are often cited on the cost side, note that they are not a discount that serves as a fallback for Spot. According to the official documentation (as of August 2026), SUD applies to the vCPUs and memory of the N1, N2, N2D, C2, M1, and M2 machine types, to sole-tenant nodes, and to GPUs attached to N1 machines, with discounts of up to 30% (up to 20% for N2, N2D, and C2). Spot VMs are not among the eligible resources, so there is no relationship in which SUD automatically fills in the hours when Spot capacity cannot be obtained. It is more accurate to estimate SUD separately from Spot, as a discount that applies automatically to long-running on-demand VMs. Comparing Spot-specific management mechanisms alone, AWS lets you combine the 2-minute interruption notice, pre-evaluation with Spot Placement Score, proactive migration with Capacity Rebalancing, and five allocation strategies. On the GCP side, the 30-second interruption notice and recreation by MIG are the core, and pre-evaluation of obtainability and fine-grained control over pool selection are left to the user's own design.
Summary
Since their launch in 2009, AWS Spot Instances have progressively built up interruption management tools, fleet management, and allocation strategies. The combination of the 2-minute interruption notice, Spot Placement Score expressing obtainability as a score from 1 to 10, Capacity Rebalancing that detects rising interruption risk and migrates in advance, and five allocation strategies including price-capacity-optimized is distinctive in that it breaks the mechanisms for running Spot in production workloads down into individual features. Azure centers on Compute Fleet with mixed purchase options and multi-region deployment, while GCP centers on Spot VMs with no runtime limit and recreation by MIG, so even with the same headline discount of up to 90%, the assumptions to settle at design time differ. The key to safely leveraging the discount is understanding the differences among these tools, checking the current price and interruption frequency of the target pools, and designing appropriate diversification strategies.
References (Official AWS Resources)
The primary sources for this page are the official AWS website and documentation. Check the official pages below for the latest specifications and pricing.
If this page and the official documentation disagree, treat the official documentation as authoritative.