Design Cost-Optimized Architectures is the smallest of the four SAA-C03 domains at 20% of the scored content, and the most commonly misread. Candidates expect arithmetic — which option costs less — and the exam almost never asks that. It asks which option costs less for this workload, and the distance between those two questions is the entire domain.
Cost optimization is constraint-matching, not penny-pinching
Here is the pattern that makes SAA-C03 cost questions feel slippery: several of the four options genuinely save money somewhere. Spot is cheaper than On-Demand. Glacier is cheaper than Standard. A Savings Plan beats list price. The wrong answers are rarely expensive options — they are discounts whose conditions this particular workload cannot meet.
Every discount AWS offers is payment for something you give up: flexibility, immediacy, or certainty. So every stem contains one deciding constraint — can the workload survive an interruption, how quickly must the data come back, will usage hold steady long enough to commit. Find that constraint first, before you look at the options, and most questions in this domain collapse to a single surviving answer.
The compute purchasing ladder
The diagram above shows the ladder the exam keeps climbing. On-Demand is the baseline: no commitment, full list price, and the correct answer only when the stem describes spiky, unpredictable, or short-lived usage that nothing else fits. Savings Plans trade a one- or three-year hourly-spend commitment for discounts of up to roughly 66% (Compute Savings Plans) while staying flexible across instance families, sizes, and even compute services. Reserved Instances commit harder — to instance attributes, not just spend — and reach up to roughly 72%. Spot sits at the deep end, up to roughly 90% off, because you agree that AWS can reclaim the capacity with two minutes’ notice.
The question that unlocks Spot is always the same: can this workload die mid-run and carry on? Checkpointing, stateless workers, fault-tolerant batch processing, containerised jobs that reschedule — these phrases are the exam telling you interruption is survivable. A steady 24/7 baseline points to a commitment instead, and the Savings Plan versus Reserved Instance choice then turns on how much flexibility the future needs: if the stem hints that instance types or Regions might change, the flexible commitment wins.
The S3 storage ladder and the retrieval question
Storage classes run the same trade in a different currency. From S3 Standard down through Standard-IA, Glacier Instant Retrieval, Glacier Flexible Retrieval, and Glacier Deep Archive, the per-GB storage price falls — and two other numbers rise: the cost of retrieving an object and the time a retrieval takes.
The exam’s decider is nearly always the retrieval-time requirement, stated plainly in the stem. Milliseconds keeps you on the instant side of the ladder — the only question left is how infrequent access is, which separates Standard, Standard-IA, and Glacier Instant Retrieval. Minutes to hours opens Glacier Flexible Retrieval. Twelve hours of patience unlocks Deep Archive, the cheapest storage AWS sells.
Two automation tools sit on top of the ladder, and the exam distinguishes them sharply. Lifecycle policies are for access patterns you know: data that is hot for 30 days, cool for 60, and archival for seven years descends the ladder on a schedule. S3 Intelligent-Tiering is for patterns you do not know: it watches each object, moves it between tiers as usage changes, and charges no retrieval fees. “Access patterns are unknown” in a stem is as close to an answer key as this exam gets.
Three cost-management services show up as each other’s distractors, and each answers a different verb. Cost Explorer analyses: interactive charts, filters, forecasts — the tool for understanding spend that has happened. AWS Budgets alerts: you define a cost or usage threshold and it notifies you — or triggers actions — when actual or forecast spend crosses it. The Cost and Usage Report is the raw detail: hourly, resource-level line items with tag columns, delivered to S3 for Athena or a data warehouse to consume. Match the stem’s verb — understand, warn, export — and the trio stops being confusable.
A fourth tool answers the verb recommend: AWS Compute Optimizer applies machine learning to CloudWatch utilisation history and names the instance type a workload should be running on. When a stem asks for specific right-sizing recommendations rather than a flag on waste, Compute Optimizer is the answer the exam wants.
The hidden cost topic: data transfer
The quietest recurring cost question involves no purchasing options at all. Instances in private subnets reaching S3 through a NAT gateway pay a per-GB data-processing charge on every byte — a cost many architectures carry without noticing. The exam’s recurring right answer is the gateway VPC endpoint: for S3 and DynamoDB, it routes the traffic over the AWS network with no hourly fee and no per-GB charge, and the NAT gateway’s processing fee on that traffic disappears entirely. It is worth internalising as a pattern, not a fact — one of the few exam answers that removes a cost rather than discounting it, which is why the question keeps coming back.
How to use the questions below
Read each stem and name the deciding constraint — interruption tolerance, retrieval time, commitment horizon, exposure requirement — before you look at the options. Then answer, and read the explanation past the correct letter: each wrong option is wrong for a specific, nameable reason, and several would save money in a different scenario. Learning which scenario each one belongs to is the skill this domain scores — and the one that quietly earns marks across the rest of the exam too.
Updated for SAA-C03 (July 2026). The sample questions above are our own work
in the style of the exam — not real exam items. The exam itself is set and marked
by the certification body.