Usage-Based Pricing

Usage-based pricing connects a cloud bill to measurable quantities of resources or activity. A workload might incur charges for compute seconds, gigabyte-months of storage, API requests, database operations, and outbound data transfer. A useful starting estimate is billable quantity multiplied by the applicable rate, summed across all chargeable components. Actual calculations can also include pricing tiers, minimum charges, discounts, commitments, and free allowances.

The important distinction is that measured usage does not always mean useful application work. A running virtual machine typically incurs compute charges even when its CPU is idle. A provisioned disk usually incurs storage charges even when it contains little data. Billing granularity and minimum durations vary by service and configuration, so engineers should verify the relevant pricing rules rather than assume everything is billed per second.

Compute purchasing options change the rate, commitment, availability, and interruption characteristics of the same workload. AWS On-Demand Instances generally require no long-term commitment and suit short-lived or unpredictable workloads. Reserved Instances provide billing discounts for matching usage in exchange for a one- or three-year commitment, with payment options that can include no upfront, partial upfront, or all upfront payment. AWS Savings Plans instead commit to a consistent amount of eligible compute spending per hour for one or three years. Both approaches can reduce steady-state costs, but unused commitments still cost money.

Discounts and capacity guarantees are separate considerations. Regional EC2 Reserved Instances do not reserve capacity; zonal Reserved Instances include a capacity reservation in the selected Availability Zone. Savings Plans do not reserve capacity. EC2 On-Demand Capacity Reservations hold matching capacity in a particular Availability Zone, and unused reserved capacity is generally charged at the applicable On-Demand rate. Eligible discounts can apply to matching reservations, but a Capacity Reservation itself is not a discount mechanism.

Spot Instances use spare compute capacity at discounted rates and can be interrupted when the provider needs that capacity back. They suit fault-tolerant batch processing, distributed workers, and other workloads designed to recover from interruptions. Dedicated Instances run on hardware dedicated to one customer, while Dedicated Hosts allocate a physical server and provide host-level placement and licensing visibility. Dedicated tenancy is an isolation or licensing choice, not inherently a cheaper pricing option; host allocation and dedicated-tenancy charges must be evaluated separately.

Storage pricing requires comparing the entire access pattern, not just the advertised storage rate. Object storage classes can charge separately for stored bytes, requests, retrievals, lifecycle transitions, and data transfer. In Amazon S3, infrequent-access and archive classes can offer lower storage rates but introduce retrieval charges, minimum billable object sizes, minimum storage durations, or restore delays, depending on the class. Frequently accessed application assets and rarely restored compliance records therefore need different storage choices.

Block storage has different billing dimensions. Amazon EBS commonly charges for provisioned volume capacity, while some volume types also charge for provisioned performance beyond included levels. Snapshots introduce additional storage charges. Stopping an EC2 instance generally stops its instance compute charges, but retained EBS volumes and other associated resources can continue billing. Similarly, serverless execution can reduce idle compute spending without eliminating charges for requests, execution duration, logs, storage, or networking.

For example, an application could cover its predictable compute baseline with a commitment, serve temporary demand with On-Demand capacity, and run restartable overnight processing on Spot. Its recent files might remain in a frequently accessed storage class while older files transition to archival storage. This arrangement only saves money if commitments are used, interruptions are acceptable, and retrieval and transition costs do not outweigh storage savings.

Use official service pricing pages to check regional rates, billing units, operating system or licensing differences, and transfer charges. AWS Pricing Calculator estimates costs from workload assumptions; AWS Price List APIs provide programmatic pricing information for supported services. AWS Cost Explorer helps analyze actual spending and usage, while AWS Budgets can notify teams when configured thresholds are crossed. Estimates are not guaranteed bills, and budget alerts are not automatic spending caps. Reconcile estimates with observed usage and adjust resource sizes, schedules, storage classes, and commitments accordingly.