Design Cost-Optimized Architectures
Cost optimization, about 20 percent of SAA-C03, means delivering the required business outcome at the lowest possible price, matching the Cost Optimization pillar. This domain asks you to choose the right pricing model for each workload, tier and lifecycle storage by access pattern, right-size and eliminate idle resources, minimize data-transfer charges, and use AWS cost-management tools to track and control spend. The recurring skill is finding the cheapest option that still meets the performance, availability, and durability requirements, never sacrificing a stated requirement just to save money. This chapter covers EC2 purchasing models, S3 storage classes and lifecycle policies, right-sizing and elasticity, data-transfer economics, and the cost-visibility toolset.
EC2 and Compute Purchasing Models
Choosing the right compute pricing model is often the single largest cost lever, and the exam expects you to match each model to a usage pattern. On-Demand pricing charges by the second or hour with no commitment and is best for short-term, unpredictable, or spiky workloads and for development and testing where you cannot forecast usage. Savings Plans and Reserved Instances reward committed baseline usage with discounts up to around seventy percent in exchange for a one- or three-year commitment: Compute Savings Plans offer the most flexibility, applying the discount across instance families, sizes, Regions, and even Fargate and Lambda, while EC2 Instance Savings Plans and Standard Reserved Instances give the deepest discount but lock you to a family or Region. Choose these for steady-state, always-on workloads such as a production database or a baseline web fleet. Spot Instances offer the deepest savings, up to about ninety percent off On-Demand, by using spare capacity that AWS can reclaim with a two-minute warning, so they suit fault-tolerant, interruptible, stateless workloads such as batch processing, big-data analysis, CI/CD, and containerized jobs that can be restarted, but never a stateful workload that cannot tolerate interruption. A cost-effective pattern combines models: cover the steady baseline with a Savings Plan, absorb variable demand with On-Demand, and run interruptible batch work on Spot. Dedicated Hosts and Dedicated Instances cost more and are chosen only for compliance or licensing needs, not to save money. For serverless, Lambda and Fargate charge only for actual execution, eliminating idle cost. On the exam, read the workload's predictability and interruption tolerance: predictable and steady points to Savings Plans or Reserved Instances, interruptible and fault-tolerant points to Spot, and short-lived or unpredictable points to On-Demand or serverless.
S3 Storage Classes and Lifecycle Policies
Storing objects in the class that matches their access pattern is a major cost lever, since S3 charges by storage class, and moving cold data to cheaper tiers can cut bills dramatically. S3 Standard suits frequently accessed hot data with no retrieval fees. S3 Standard-Infrequent Access and One Zone-Infrequent Access lower storage cost for data accessed less often, with a per-GB retrieval charge; One Zone-IA is cheaper because it stores data in a single AZ, appropriate only for re-creatable or non-critical data. For archives, the Glacier classes offer the lowest storage prices with retrieval trade-offs: Glacier Instant Retrieval for rarely accessed data that still needs millisecond access, Glacier Flexible Retrieval for archives you can wait minutes to hours to retrieve, and Glacier Deep Archive for long-term compliance data retrieved in hours at the very lowest cost. When access patterns are unknown, unpredictable, or changing, S3 Intelligent-Tiering automatically moves each object between frequent, infrequent, and archive tiers based on actual usage for a small monitoring fee and no retrieval charges, making it the safe default when you cannot predict access and want to avoid retrieval surprises. Lifecycle policies automate the economics: define rules that transition objects to cheaper classes after a set age, for example Standard to Standard-IA after thirty days and to Glacier after ninety, and expire or delete objects that are no longer needed, including old noncurrent versions when versioning is enabled. Combine lifecycle transitions with the right initial class rather than paying Standard rates for cold data indefinitely. On the exam, map the retrieval requirement to the class: predictable hot data to Standard, predictable cold data to IA or Glacier by required retrieval speed, unknown or shifting patterns to Intelligent-Tiering, and always automate movement with lifecycle rules.
Right-Sizing and Eliminating Waste
Paying only for the capacity you actually use is the most direct cost discipline, and much waste comes from overprovisioned or idle resources. Right-sizing means selecting the smallest instance, volume, or database that still meets performance needs; use CloudWatch metrics to observe real CPU, memory, and I/O utilization and AWS Compute Optimizer to receive data-driven recommendations to downsize or change instance types. An instance running at five percent CPU is a candidate to shrink or move to a burstable T-family instance. Elasticity eliminates idle cost automatically: configure Auto Scaling to release capacity when demand drops so you are not paying for peak capacity around the clock, and schedule non-production environments such as development and test fleets to stop outside business hours, since a stopped EC2 instance incurs no compute charge. Serverless services take this further by charging only for actual usage: Lambda bills per invocation and duration, Fargate per running task, Aurora Serverless and DynamoDB on-demand scale capacity to load, so idle time costs nothing, making them cost-effective for variable or intermittent workloads. Delete unattached EBS volumes, release unused Elastic IP addresses, remove old snapshots, and clean up idle load balancers, all of which accrue charges while unused. For storage, right-size EBS volumes and prefer gp3 over the older gp2 for a better price-performance ratio. Continuously review utilization rather than treating sizing as a one-time decision, because workloads change. On the exam, when a scenario describes consistently low utilization, the answer is right-sizing or moving to serverless; when it describes paying for idle capacity overnight or on weekends, the answer is Auto Scaling to scale in or scheduling resources to stop; and when unused resources accrue charges, the answer is to identify and remove them.
Data Transfer and Network Cost Economics
Data transfer is a frequently overlooked expense, and designing to minimize it can materially lower the bill. The core rule is that data transferred out of AWS to the internet is charged, and so is traffic that crosses Availability Zones or Regions, while inbound data from the internet is generally free and traffic within the same AZ using private IP addresses is free. This shapes several design choices. Keep chatty components in the same AZ and use private IP addresses to avoid cross-AZ charges, though balance this against the resilience need for multi-AZ deployment. Serve heavy outbound content through Amazon CloudFront, whose data-transfer-out pricing is lower than direct S3 or EC2 egress and which reduces origin load, so a media or download-heavy site saves by caching at the edge. Use VPC endpoints to reach AWS services such as S3 and DynamoDB privately: a gateway endpoint is free and keeps traffic off the internet and NAT gateway, avoiding both NAT processing charges and internet egress, which is a common cost-optimization answer for private-subnet workloads that call S3. Note that NAT gateways charge both an hourly rate and a per-GB data-processing fee, so routing large S3 traffic through a gateway endpoint instead of a NAT gateway saves money. For large one-time or periodic bulk data migrations, the AWS Snow Family physically ships data on rugged devices, which is far cheaper and faster than saturating a network link for petabyte-scale transfers, and DataSync efficiently moves data online for ongoing transfers. Place resources close to their consumers to reduce cross-Region transfer, and evaluate transfer patterns early rather than discovering egress charges after launch. On the exam, when private-subnet instances access S3, choose a gateway VPC endpoint to cut NAT and egress cost; when serving global downloads, choose CloudFront; and when migrating petabytes once, choose the Snow Family.
Cost Visibility, Governance, and Managed Services
You cannot optimize what you cannot see, so the final piece is the AWS cost-management toolset and the governance practices that keep spend aligned with value. AWS Cost Explorer visualizes and analyzes spending and usage over time, surfaces trends, and provides right-sizing and Savings Plans recommendations, making it the tool for understanding where money goes. AWS Budgets lets you set custom cost and usage budgets and sends alerts, or triggers actions, when spending exceeds or is forecast to exceed a threshold, so choose it when a scenario wants proactive notification before a bill grows too large. AWS Cost Anomaly Detection uses machine learning to flag unusual spend automatically. The AWS Cost and Usage Report provides the most granular line-item data for deep analysis, often queried through Athena or loaded into QuickSight. Cost allocation tags are the backbone of governance: tag resources by project, team, environment, or cost center, activate the tags, and then break down and charge back spending accurately in Cost Explorer, which is the standard answer when a company needs to attribute costs to departments. In multi-account setups, AWS Organizations with consolidated billing aggregates usage so volume discounts and Reserved Instance or Savings Plans benefits apply across all accounts, and service control policies enforce guardrails. Beyond tools, managed and serverless services frequently lower total cost of ownership by removing the operational overhead of patching, scaling, and running infrastructure, so migrating from self-managed instances to Aurora, Fargate, or Lambda can reduce both direct and staffing costs. On the exam, map the need to the tool: analyze and forecast spend with Cost Explorer, alert on thresholds with Budgets, detect surprises with Cost Anomaly Detection, and attribute costs with allocation tags under consolidated billing.
Keep going: the full AWS Solutions Architect Associate (SAA-C03) guide covers every section of the exam. AWS Solutions Architect Associate (SAA-C03) — Complete Study Guide (2026) — PDF + EPUB, $14.99 · 14-day refund →
Practice stays free. The full AWS Solutions Architect Associate (SAA-C03) study guide is the material itself, taught start to finish — a downloadable PDF + EPUB you keep.