Visual overview
Select a model from interruption tolerance, commitment, and capacity requirements.
Technical reference
Purchase models change economics and guarantees, not the EC2 compute API.
Flexible usage for uncertain duration or shape.
Discounted eligible compute usage in exchange for commitment.
Discounted spare capacity for fault-tolerant workloads.
Reserves matching EC2 capacity; assurance and billing discount are separate concerns.
Start by separating price from capacity
EC2 offers several purchasing and capacity mechanisms, but they do not all solve the same problem. Some options primarily change the price you pay for eligible usage. Others help ensure that capacity is available in a particular location. Dedicated options change the tenancy or hardware-isolation model. Treating all of them as interchangeable 'reservations' leads to bad architecture decisions.
On-Demand is the flexible baseline: you use eligible EC2 capacity without making a long-term compute commitment. It is useful for new, variable or short-lived workloads where preserving flexibility is more valuable than committing to a predictable baseline. Once usage becomes steady and measurable, commitment-based pricing can reduce the effective rate for eligible compute.
Savings Plans and Reserved Instances
Savings Plans are commitment-based pricing models. Compute Savings Plans are designed to apply discounted rates across eligible EC2, AWS Fargate and AWS Lambda usage while offering broad flexibility; EC2 Instance Savings Plans trade some flexibility for a commitment around an EC2 family in a Region. The key concept is a usage commitment that AWS applies automatically to matching eligible usage—not ownership of a particular virtual machine.
EC2 Reserved Instances are an EC2 billing construct with Standard and Convertible options and defined terms. A Regional RI provides a billing benefit when matching usage is present. A zonal Reserved Instance can also provide a capacity reservation in the selected Availability Zone. AWS currently recommends Savings Plans over Reserved Instances for many discount-focused scenarios, but Reserved Instances remain relevant when their EC2-specific attributes or capacity behavior match the requirement.
Spot Instances
Spot Instances use spare EC2 capacity at a steep discount relative to On-Demand prices, but the capacity can be interrupted when EC2 needs it back. That trade-off makes Spot powerful for workloads that can tolerate interruption, distribute work, retry, checkpoint progress or replace workers easily. Examples include batch processing, some CI workers, rendering, stateless services with diversified capacity and other fault-tolerant compute.
Spot is not simply 'cheap EC2.' The architecture must explicitly handle interruptions and capacity variability. A workload that requires one particular instance to stay running continuously is a poor match. Auto Scaling mixed-instance policies and AWS Batch can help applications use diversified Spot capacity without building all of the fleet-management logic manually.
Capacity Reservations and dedicated hardware
On-Demand Capacity Reservations reserve EC2 compute capacity for you in a specific Availability Zone and are useful when launch certainty matters, such as planned events, disaster-recovery capacity or other business-critical requirements. They address availability of capacity rather than being a long-term discount mechanism. Capacity Blocks address advance reservation for supported accelerated ML capacity and are a more specialized tool.
Dedicated Instances and Dedicated Hosts address hardware isolation. Dedicated Instances run on hardware dedicated to a single customer account, while Dedicated Hosts give additional visibility and control over a physical server and can support some licensing requirements. Because these options affect cost and operational choices, select them for an actual isolation, compliance or licensing requirement—not simply because 'dedicated' sounds more secure.
Key takeaways
- 01
On-Demand preserves flexibility; commitment models exchange flexibility for lower eligible compute rates.
- 02
Compute Savings Plans can apply across eligible EC2, Fargate and Lambda usage; Reserved Instances are EC2-specific.
- 03
Spot requires interruption-tolerant architecture.
- 04
Capacity Reservations solve launch-capacity assurance; dedicated options solve tenancy and hardware-isolation requirements.
Official AWS sources
Use these primary AWS resources for the source material behind this article and for deeper reference.