AWS Pricing and Billing

How Amazon Web Services charges, and the account-level machinery for predicting and controlling that charge. The central premise is that capital expenditure on hardware is traded for variable operating expenditure that tracks actual usage.

Key points

• Purchasing options for compute — the same axis appears throughout Amazon EC2:
  • On-Demand — pay by the hour or second with no commitment. Suits intermittent, spiky or unpredictable workloads, and short experimental runs. Cannot be interrupted by AWS.
  • Reserved Instances — a one or three year commitment in exchange for up to about 72% off On-Demand, with a capacity reservation in a specific Availability Zone. Suits a steady, predictable baseline. Payment options are No upfront, Partial upfront and All upfront; the longer the term and the more paid upfront, the larger the discount, so a 3-year All Upfront Standard RI gives the largest savings. Convertible RIs trade some of that discount for the ability to change instance family during the term.
  • Savings Plans — a commitment to a steady dollar-per-hour spend for one or three years, more flexible than Reserved Instances across instance families and services. Four current types: Compute Savings Plans (most flexible, spans EC2, Fargate and Lambda regardless of family/Region/OS, up to 66% off On-Demand), EC2 Instance Savings Plans (tied to one instance family in a Region, up to 72% off), SageMaker AI Savings Plans (up to 64% off), and Database Savings Plans (up to 35% off, covering Aurora, RDS, DynamoDB and others).
  • Spot Instances — bid on spare capacity for up to about 90% off On-Demand. Can be interrupted, so it suits stateless, fault-tolerant, restartable work: testing, batch, massively parallel jobs.
  • Dedicated Instances and Dedicated Hosts — isolated hardware for licensing or compliance reasons, at a premium.
• Free Tier — AWS now structures this as a Free plan (up to $200increditsover6monthsfornewaccounts,limitedservices)ora\ast \ast Paidplan\ast \ast (fullpay-as-you-goaccess,alsoeligibleforthe$200 credit); layered on either plan are Always free services (30+ services, free within monthly limits indefinitely) and short-term trials on selected paid services.
• Consolidated billing through AWS Organizations — one bill for many accounts at no additional cost. Usage aggregates across member accounts, so volume discount tiers are reached sooner, while per-account charges remain individually trackable. The management (payer) account settles the total.
• Cost allocation tags — tags promoted to billing dimensions, so spend can be attributed to a business unit, project or environment. The prerequisite for meaningful chargeback.
• Total cost of ownership (TCO) — the on-premises baseline (hardware, facilities, staff, refresh cycles) that a migration case is measured against; needed to state realised savings credibly.
• The tooling, in the order it is normally used — AWS Pricing Calculator to estimate before provisioning, AWS Budgets to set thresholds and alert on them, AWS Cost Explorer to visualise and forecast actual spend, AWS Cost and Usage Report for the line-item detail, Reserved Instance Reporting for commitment utilisation and coverage, AWS Cost Anomaly Detection for unusual spend, and AWS Compute Optimizer for rightsizing recommendations.
• Common patterns worth remembering — data transfer in to AWS is generally free while data transfer out to the internet is charged; within that, transfer within the same Availability Zone is free but cross-AZ and cross-Region transfer is charged. AWS Identity and Access Management (IAM), AWS CloudFormation and AWS Organizations cost nothing themselves, you pay only for what they provision or manage. Since February 2024, every public IPv4 address — in-use or idle, whether on an EC2 instance, RDS instance, NAT Gateway or elsewhere — carries an hourly charge (currently $0.005/hour), a cost that didn’t previously exist and is easy to overlook when estimating.

Sources

• Savings Plans types
• AWS Free Tier
• New – AWS Public IPv4 Address Charge
• Amazon VPC pricing
• Service quotas are a separate concern from cost but often surface alongside it: see AWS Service Quotas and the service-limit checks in AWS Trusted Advisor.