Definition: Spare AWS compute capacity available at discounts of up to 90% compared to on-demand pricing, with the possibility of interruption.
— Source: NERVICO, Product Development Consultancy
What are Spot Instances
Spot Instances are spare AWS compute capacity offered at discounts of up to 90% compared to on-demand pricing. The model works like a marketplace: AWS has EC2 capacity that is not being used at a given moment and offers it at a reduced price. The tradeoff is that AWS can interrupt a Spot Instance with a 2-minute notice when it needs to reclaim that capacity for on-demand or reserved instances. Not all workloads are suitable for Spot, but for those that tolerate interruptions, the savings are significant.
How It Works
When launching a Spot Instance, you select the instance type and availability zone. AWS allocates the instance if capacity is available at the current Spot price, which fluctuates based on supply and demand but is typically 60-90% below on-demand pricing. When AWS needs to reclaim capacity, it sends an interruption notification 2 minutes in advance through instance metadata and Amazon EventBridge. The instance can be terminated, stopped, or hibernated depending on configuration. To maximize availability, it is recommended to use Spot Fleet or EC2 Fleet, which distribute instances across multiple types and zones, reducing the probability of simultaneous interruption.
Why It Matters
For workloads that tolerate interruptions, such as batch processing, rendering, machine learning model training, testing, CI/CD pipelines, or data analytics, Spot Instances offer savings that can transform a project’s economic viability. A processing cluster that would cost $10,000 monthly at on-demand pricing can run for $1,500 with Spot Instances. Combined with Auto Scaling and an instance type diversification strategy, interruptions become infrequent and manageable.
Practical Example
A data engineering team runs a daily ETL pipeline processing 2 TB of data on a cluster of 50 c5.2xlarge instances. At on-demand pricing, the cost would be $12,240 monthly. The team configures a Spot Fleet with 6 alternative instance types distributed across 3 availability zones. The pipeline implements checkpointing every 15 minutes so that if an instance is interrupted, the work resumes from the last checkpoint on another instance. The average interruption rate is 5%, and the monthly cost drops to $1,836, an 85% savings.