2 AWS Pricing Models

The dimensions behind EC2 choices

EC2 choices balance cost, commitment, flexibility, interruption risk, and hardware control. The right fit depends on how predictable the workload is and what would happen if it stopped.

On-Demand, Savings Plans, Reserved Instances (RIs), and Spot Instances are purchasing or billing choices. Dedicated Instances and Dedicated Hosts concern tenancy and hardware control. These categories address different needs and can be combined where AWS supports it.

Compare billing and purchasing choices

On-Demand charges for compute as it is used and requires no long-term commitment. It can fit short projects, variable or uncertain workloads, or workloads that must remain available. For steady usage, it usually costs more than commitment-based options.

Savings Plans commit to a specified amount of eligible compute usage, measured in dollars per hour, for one or three years. They can suit a steady baseline when a discount is wanted but instance sizes, families, Regions, or compute services may change. The commitment still applies if eligible usage falls below it, so unused commitment can go to waste.

Reserved Instances (RIs) commit to a specified EC2 configuration for one or three years in exchange for a billing discount. They can fit predictable EC2 usage when the RI's matching requirements are suitable. They are less flexible than a Compute Savings Plan, and their scope and attributes should be checked rather than assumed.

Spot Instances use spare EC2 capacity at a variable discount without a long-term commitment. They can suit fault-tolerant, flexible, or restartable work, including batch processing and suitable stateless tasks. AWS can interrupt Spot capacity; interruption notices generally provide two minutes, but the workload must be designed to handle interruption.

Choose commitment scope and capacity

A Compute Savings Plan may suit a steady baseline when the workload is expected to change instance families or Regions. An EC2 Instance Savings Plan is more specific: it commits to an EC2 instance family and Region. Choose the plan type according to the expected stability of eligible usage and configuration.

A Regional RI provides a discount but does not reserve capacity. A Zonal RI also reserves capacity in its specified Availability Zone. If capacity must be available in a particular Availability Zone, evaluate a Capacity Reservation separately from the discount decision. Capacity Reservations and pricing commitments solve different problems and may be combined.

Choose hardware tenancy and control

Dedicated Instances run on single-tenant hardware and are billed per instance. They may fit a dedicated-hardware requirement when physical-host control is not needed. Dedicated tenancy can cost more and does not provide host-level visibility or placement control.

Dedicated Hosts rent a physical server dedicated to the customer and are billed per host. They may fit a need for host-level control or eligible bring-your-own software licenses tied to sockets, cores, or virtual machines. Host billing can make poor utilization costly, so consider utilization and verify the software license terms.

Apply a workload-based decision process

Use the workload's interruption tolerance, predictability, configuration, capacity needs, and hardware requirements to narrow the choices.

1. If interruption could cause unacceptable harm, avoid relying on Spot. Consider On-Demand or a commitment discount for predictable baseline usage.

2. If usage is short-term or hard to predict, start with On-Demand and reassess when reliable usage history is available.

3. If there is a stable amount of compute usage, consider a Savings Plan. A Compute Savings Plan offers more flexibility across eligible compute usage; an EC2 Instance Savings Plan is more specific to an EC2 family and Region.

4. If a particular EC2 configuration is stable, compare an RI with a Savings Plan. Check the RI's attributes and scope, and do not assume that a Regional RI reserves capacity.

5. If capacity must be available in a particular Availability Zone, evaluate a Capacity Reservation separately from the discount choice.

6. If dedicated hardware or license eligibility is required, compare Dedicated Instances with Dedicated Hosts. Choose a Host when host-level visibility, placement control, or applicable host-based licensing matters; otherwise, per-instance billing for a Dedicated Instance may meet the requirement.

A common strategy is to cover predictable baseline demand with a commitment, retain On-Demand for uncertain peaks, and use Spot for work that can be interrupted. Estimate commitments conservatively: a discount does not help if the committed usage is not consumed.

Match the model to common scenarios

For a short experiment expected to last several weeks, with uncertain continuation, On-Demand is a suitable primary choice: a long-term commitment may outlast the need.

For steady compute usage where instance families and Regions are expected to change, a Compute Savings Plan offers a better fit if usage is reliably above the chosen commitment.

For a nightly batch job that can restart from checkpoints and finish whenever spare capacity is available, Spot fits because restartability makes interruption manageable.

For the same EC2 configuration running around the clock, with a discount desired and capacity assured in one specific Availability Zone, consider a Zonal RI or an On-Demand Capacity Reservation plus an appropriate discount plan. Decide separately how to secure capacity and reduce the bill; a Regional RI alone does not reserve capacity.

For a workload requiring a physical server dedicated to the customer and visibility into sockets and cores for eligible existing licenses, a Dedicated Host provides the relevant physical-host visibility and control. Verify the software license terms and Host utilization before choosing it.