Cost Optimization

AWS SAA SAA-C03 · Design Cost-Optimized Architectures

By Abu Anns · Last updated 20 November 2026Exam version: SAA-C03 (current)

Cost optimization is an architecture discipline

The cost-optimized architectures domain is 20% of the SAA-C03 exam, but cost thinking runs through every other domain: the exam constantly asks for the solution that meets the requirements at the lowest cost, or the MOST cost-effective option among several that would all work. The underlying principle is simple — pay only for what you use, and pay less for what you commit to — but applying it requires knowing the pricing model of each service and the tools AWS provides to see and control spending.

Cost optimization works at three levels. At the architecture level, you choose services and designs that avoid waste: serverless for spiky workloads, the right storage class for the access pattern, caching to reduce database load, CloudFront to reduce origin requests. At the purchasing level, you commit to steady-state usage through Reserved Instances and Savings Plans, and use Spot for interruptible work. At the operations level, you find and eliminate waste: idle instances, unattached EBS volumes, old snapshots, oversized instances, and forgotten resources.

The cost tools: seeing where the money goes

AWS Cost Explorer visualizes historical spending by service, Region, account, or tag, and forecasts future costs. It is the first stop for understanding a bill. AWS Budgets sets alerts (and can trigger actions) when spending, usage, or reservations cross a threshold — for example, emailing the team when the monthly bill reaches 80% of budget. AWS Cost and Usage Report (CUR) provides the most detailed billing data, delivered to S3 for analysis with Athena or QuickSight. AWS Pricing Calculator estimates the cost of an architecture before you build it. AWS Compute Optimizer uses machine learning on CloudWatch metrics to recommend right-sized instance types — it finds instances that are over-provisioned (paying for capacity they never use) or under-provisioned (at risk of poor performance). AWS Trusted Advisor flags specific waste: idle RDS instances, unassociated Elastic IPs, underutilized EBS volumes, and idle load balancers. Tagging resources (by team, project, or environment) with cost allocation tags makes it possible to attribute spending accurately.

Compute cost: the big lever

Compute is usually the largest part of an AWS bill, so it gets the most exam attention. The decision framework: On-Demand for variable, short-lived, or unpredictable workloads; Reserved Instances or Savings Plans for steady-state workloads (up to ~72% discount for a 1- or 3-year commitment); Spot Instances for fault-tolerant, interruptible work (up to ~90% discount). Compute Savings Plans apply to EC2, Fargate, and Lambda usage across Regions and instance families, offering more flexibility than Reserved Instances at a slightly smaller maximum discount. EC2 Instance Savings Plans apply to a specific instance family in a specific Region at the deepest discount. For databases, RDS Reserved Instances work the same way. For serverless, Lambda's pay-per-use model is cheapest when usage is low or bursty, but a high-volume, always-on workload can be cheaper on reserved EC2 or containers.

Right-sizing is the ongoing discipline of matching instance sizes to actual usage. Compute Optimizer recommendations, CloudWatch metrics, and Trusted Advisor checks all feed into it. The exam also rewards stopping what is not needed: development and test instances stopped overnight, non-production environments torn down when idle. And Auto Scaling is a cost tool as much as a resilience tool: scaling in when demand drops is the difference between paying for peak capacity all day and paying for it only during the peak.

Storage, data transfer, and the hidden costs

Storage costs are driven by the class you choose and the lifecycle of the data. S3 lifecycle policies that move aging data to cheaper classes (and expire it when retention ends) are the standard exam answer for storage cost. EBS snapshots should be cleaned up when no longer needed, and old AMIs deleted. DynamoDB's provisioned versus on-demand choice is a cost decision: provisioned with auto scaling for predictable traffic, on-demand for unpredictable traffic.

Data transfer is the cost the exam most often hides in a scenario. Inbound data transfer to AWS is free. Outbound data transfer to the internet is charged per gigabyte, and it can dominate the bill for data-heavy applications. The mitigations the exam tests: serve content through CloudFront (data transfer out through CloudFront is cheaper than direct from S3 or EC2, and cached content is not transferred from the origin at all), keep traffic within the same Availability Zone where possible (cross-AZ transfer is charged), use VPC endpoints to keep AWS service traffic off the internet, and use Direct Connect for large, sustained transfers (lower per-GB pricing than internet egress). The classic trap: a design that moves large amounts of data between AZs or Regions unnecessarily, when a small redesign would keep the data local.

Cost governance: tagging, anomaly detection, and Graviton

Controlling cost at scale is as much about governance as about architecture. A tagging strategy is the foundation: every resource carries tags for its owner, project, and environment, and cost allocation tags make those tags appear in Cost Explorer and the Cost and Usage Report so spending can be attributed accurately. AWS Organizations supports tag policies that enforce consistent tag keys and values across accounts. AWS Cost Anomaly Detection uses machine learning to watch spending patterns and alert when something unusual appears — a sudden spike in a service that normally costs little is often the first sign of a misconfiguration, a runaway process, or an unexpected workload. AWS Budgets can go beyond alerting: budget actions can automatically stop EC2 instances or deny IAM permissions when a threshold is crossed, which is useful for sandbox and development accounts.

Two more cost levers deserve attention. AWS Graviton processors — ARM-based chips designed by AWS — power the latest generations of EC2, RDS, ElastiCache, and other services, offering better price performance than comparable x86 instances (AWS claims up to 40% better price performance for many workloads). Migrating to Graviton-based instance types is often a straightforward change with an immediate saving, provided the application runs on ARM. For databases, RDS Performance Insights and Compute Optimizer help identify over-provisioned database instances, and Aurora Serverless v2 scales database capacity automatically with demand, so you pay for what the workload actually uses rather than a fixed instance size. The exam's cost questions often combine several of these levers: the best answer is usually the one that removes waste at the architectural level (right service, right size, right purchasing option) rather than the one that adds a monitoring tool alone.

Recommendations worth acting on

AWS generates concrete savings recommendations in several places: Cost Explorer suggests Savings Plans and Reserved Instance purchases based on your actual usage history, Compute Optimizer recommends smaller instance types, and Trusted Advisor flags idle resources. Reviewing these recommendations monthly and acting on the largest ones first is the simplest cost-optimization habit an organization can build.

Advertisement

Exam traps to watch for

Practice questions

Test yourself on this guide. Answers are checked instantly, with a full explanation for each question.

Q1.A company runs a steady-state application on 50 EC2 instances 24/7. The instances are currently On-Demand. The company wants to reduce compute cost with a 1-year commitment but needs the flexibility to change instance families during the year. Which option BEST meets these requirements?

Report an error

Q2.A company serves large video files directly from S3 to users on the internet. Data transfer costs are the largest item on the AWS bill. Which solution MOST effectively reduces these costs?

Report an error

Q3.A company wants to identify EC2 instances that are over-provisioned for their actual workload so it can downsize them. Which AWS service provides machine-learning-based right-sizing recommendations?

Report an error

Q4.A company runs development and test environments on EC2 instances that are only used during business hours (8 AM to 6 PM, weekdays). The environments currently run 24/7. Which change provides the LARGEST cost saving with the least effort?

Report an error

Q5.A company stores 500 TB of data in S3 Standard. Analysis shows that 80% of the objects have not been accessed in over a year, but the company cannot predict which objects will be accessed in the future. Which solution reduces storage cost with the least operational effort?

Report an error

Q6.A company's finance team wants an alert when the AWS bill for the current month is forecast to exceed $10,000. Which AWS service should the solutions architect use?

Report an error

Q7.A DynamoDB table receives a steady, predictable 1,000 writes per second around the clock. It is currently configured with on-demand capacity. Which change will MOST reduce the table's cost?

Report an error

Q8.An application runs in a VPC and transfers large volumes of data to Amazon S3 every day. The traffic currently goes through a NAT gateway. Which change reduces the data transfer cost of this traffic?

Report an error