AWS pricing model questions on the SAA-C03 exam require you to match a workload description to the most cost-effective purchasing option. The key is identifying two things: how predictable is the workload, and can it tolerate interruption?
The Four EC2 Purchasing Options
On-Demand
Pay by the second with no commitments. The most flexible option.
- Cost: Highest per hour
- Commitment: None
- Best for: Unpredictable workloads, short-term projects, testing, development
Reserved Instances (RIs)
Commit to a specific instance type in a specific region for 1 or 3 years in exchange for a discount (up to 72% vs On-Demand).
RI types:
- Standard RI: highest discount, cannot change instance type or family mid-term
- Convertible RI: lower discount, can change instance type/OS/tenancy during term
- Scheduled RI: reserved for specific time windows (deprecated in newer regions)
Payment options: All Upfront (largest discount), Partial Upfront, No Upfront.
Best for: steady-state, predictable workloads running 24/7 (databases, application servers with consistent load).
Savings Plans
Commit to a consistent amount of compute spend per hour (measured in $/hour) for 1 or 3 years. More flexible than RIs.
Types:
- Compute Savings Plans: applies across EC2, Fargate, and Lambda; any instance family, region, or OS
- EC2 Instance Savings Plans: specific instance family in a region; higher discount than Compute Savings Plans
Best for: predictable spending but with flexibility to change instance types, regions, or compute services.
Exam distinction: Savings Plans are more flexible than RIs (can span instance families and services), but specific Standard RIs for a fixed instance type in one region give the maximum discount.
Spot Instances
Use AWS's spare EC2 capacity at up to 90% discount. AWS can reclaim Spot instances with 2 minutes' notice when it needs the capacity back.
- Cost: Lowest (up to 90% off On-Demand)
- Interruption: possible with 2-minute warning
- Best for: fault-tolerant, stateless, interruptible workloads: batch processing, data analysis, CI/CD pipelines, machine learning training, video rendering
Not suitable for: databases, web servers serving real-time traffic, any workload that cannot tolerate interruption.
Choosing the Right Option
| Workload Characteristics | Purchasing Option |
|---|---|
| Predictable, runs 24/7, same instance type for 1+ year | Standard Reserved Instance |
| Predictable spending, but may change instance types | Savings Plans |
| Unpredictable, short-term, or variable load | On-Demand |
| Interruptible batch jobs, fault-tolerant processing | Spot |
| Mix of steady-state + spikes | RI or Savings Plan for baseline + On-Demand or Spot for spikes |
Common SAA-C03 Exam Scenarios
"A company runs a MySQL database that needs to be available 24/7 for the next 3 years."
Reserved Instance (3-year term, Standard) for the database — predictable, steady-state, same instance type.
"A genomics company needs to process large datasets over the weekend. Jobs can be restarted if interrupted."
Spot Instances — interruptible batch processing with significant cost savings.
"A company has a mix of workloads and wants to reduce EC2 costs without committing to specific instance types."
Compute Savings Plans — commitment to spend level, flexibility across instance types.
"A startup is building a new application and cannot predict traffic patterns."
On-Demand — no commitment, pay for what you use, can switch to RIs or Savings Plans once patterns are established.
The Spot Fleet and Auto Scaling Combination
For maximum cost savings with fault tolerance, the SAA-C03 exam commonly recommends combining Spot Instances in a Spot Fleet (multiple instance types to reduce interruption probability) with Auto Scaling Groups that maintain desired capacity.
Practice SAA-C03 cost optimization questions to build automatic recognition of which pricing model each scenario calls for.
Reserved Instance Types — Standard vs Convertible vs Scheduled
Reserved Instances come in three flavors and the exam distinguishes their flexibility.
Standard Reserved Instances provide the highest discount — up to 72% off On-Demand pricing. The tradeoff is inflexibility. You commit to a specific instance family, size, region, and platform (Linux/Windows). You cannot change these after purchase. You can modify certain attributes (like instance size within the same family) or sell unused Standard RIs on the Reserved Instance Marketplace.
Convertible Reserved Instances provide a smaller discount — up to 54% off On-Demand. The benefit is flexibility: you can exchange Convertible RIs for other Convertible RIs of equal or greater value. This lets you change instance family, operating system, or tenancy during the commitment period. Use Convertible RIs when the workload is likely to change — if you're planning to scale up instance types or change from Linux to Windows.
Scheduled Reserved Instances are for workloads that run on a predictable schedule — for example, a batch job that runs every weekday 6 PM to midnight. You commit to capacity in a specific time window and pay only for that window. Scheduled RIs are less commonly tested but appear in scenarios where On-Demand would be used for a regular but non-continuous workload.
Exam decision logic: if the exam says "flexible workloads that may require different instance types," Convertible is the answer. If it says "stable, predictable workload with no expected changes," Standard gives better discount. If it says "workload runs only certain hours," Scheduled fits.
Savings Plans vs Reserved Instances — The New Preferred Option
AWS launched Savings Plans as a more flexible alternative to Reserved Instances, and the exam tests when each is appropriate.
Compute Savings Plans provide up to 66% discount and apply automatically to any EC2 usage regardless of instance family, size, AZ, region, OS, or tenancy. They also apply to Fargate and Lambda. This is the most flexible option — if you commit $10/hour of compute usage, that commitment applies anywhere in your account.
EC2 Instance Savings Plans provide up to 72% discount (same as Standard RIs) but are locked to a specific instance family in a specific region. Within that family, you can change size and OS freely. For example, committing to m5 in us-east-1 covers m5.large, m5.xlarge, m5.2xlarge — all OS variants — automatically.
When Savings Plans beat Reserved Instances: Savings Plans are simpler to manage (no RI inventory to track) and more flexible (Compute SP applies to Lambda and Fargate too). AWS's own cost optimization tools now recommend Savings Plans over RIs for most use cases.
When Reserved Instances still make sense: if you need to sell unused capacity on the Reserved Instance Marketplace (only Standard RIs can be listed), or if you need the exact 72% discount with instance family lock-in and the Savings Plan 66% doesn't meet your budget target.
The exam scenario clue for Savings Plans: "The company uses a mix of instance types and wants to commit to compute spending without managing a catalog of Reserved Instances."
Spot Instance Interruption — What Happens and How to Handle It
Spot Instances can be interrupted with 2 minutes' notice when EC2 needs capacity back. The exam tests both the mechanics and the architectural patterns for handling interruption.
When an interruption is incoming, the EC2 instance receives an interruption notice in the instance metadata at http://169.254.169.254/latest/meta-data/spot/interruption-action and via CloudWatch Events (now EventBridge). The 2-minute window is your application's time to finish work, save state, and shut down cleanly.
Interruption behaviors (configured when requesting the Spot Instance):
- Terminate (default): instance is terminated at interruption
- Stop: instance is stopped; volume persists; can be restarted when capacity becomes available
- Hibernate: instance state and memory are written to EBS, then stopped; resumes where it left off
Spot Fleet launches a mix of Spot and On-Demand capacity across multiple instance types and AZs. It replaces interrupted instances automatically and maintains your target capacity. This is the pattern for fault-tolerant Spot workloads.
The right use cases for Spot Instances (exam vocabulary): batch processing, big data analysis, CI/CD test runners, stateless web tier workers, containerized workloads with short tasks. The wrong use cases: stateful workloads, databases, anything that can't tolerate mid-task interruption.
Data Transfer Costs — The Hidden Bill
AWS doesn't always make data transfer pricing obvious, and the exam tests whether you know which transfers are free and which are charged.
| Transfer Type | Cost |
|---|---|
| Inbound to AWS from internet | Free |
| Same AZ, same service (e.g., EC2 to EC2) | Free |
| Cross-AZ (same region) | ~$0.01/GB each direction |
| Cross-region | Varies, typically $0.02–$0.09/GB |
| Internet egress (outbound to internet) | Tiered: first 10 TB/month cheapest |
| EC2 to S3 in same region | Free |
| EC2 to CloudFront | Free (CloudFront to internet is charged) |
The cross-AZ charge is the one that surprises most people. Traffic between an EC2 instance in us-east-1a and an RDS instance in us-east-1b costs $0.01/GB in each direction. This is why Multi-AZ deployments that generate heavy traffic between AZs (like an app server constantly querying an RDS Multi-AZ standby) aren't as free as they seem. The standby doesn't serve reads, but replication traffic does cross AZs.
The exam scenario: "A company has two EC2 instances in different AZs in the same region that transfer 10 TB of data between them monthly. How can they reduce this cost?" Options include: move them to the same AZ (eliminates cross-AZ charge but sacrifices AZ redundancy), or use a regional load balancer pattern. Same-AZ transfer is always free.
Cost Allocation Tags — For Exam Scenarios About Tracking Spend by Team/Project
AWS lets you tag resources with key-value pairs (e.g., Project: DataPlatform, Team: Engineering). For these tags to appear in cost reports, they must be activated as cost allocation tags in the AWS Billing console.
Once activated, tags appear in AWS Cost Explorer with a lag (up to 24 hours after activation). You can then filter and group costs by tag — seeing exactly how much each project or team is spending.
The exam scenario: "A company wants to track AWS spending by department. How do they achieve this?" The answer involves: 1) tag all resources with a Department tag, 2) activate that tag as a cost allocation tag in Billing console, 3) use Cost Explorer to group by that tag.
Untagged resources can't be attributed to any department — they appear in Cost Explorer as "No tag." This is why tagging governance (enforcing tags via Service Control Policies or AWS Config rules) matters in enterprise accounts.
AWS Budgets can also use cost allocation tags to set alerts for a specific project — "alert me when Project: DataPlatform costs exceed $10,000/month" — which is another pattern the exam tests.
The Auto Scaling Cost Pattern
Auto Scaling adjusts capacity based on demand. The exam tests the scenario where a company is consistently over-provisioned — running large instances at low utilization — and asks how to reduce cost.
The classic setup: "A company runs m5.2xlarge instances in an Auto Scaling group. CloudWatch shows average CPU utilization is consistently below 20%. What should they do to reduce costs?"
The correct answer combines two things:
- Right-size the instance type — if CPU is 20%, the instances are too big. Move to m5.large or even m5.xlarge and measure again.
- Purchase Reserved Instances or Savings Plans for the baseline capacity — once you know the minimum capacity that always runs, commit to it.
The wrong answer: "Use Spot Instances for the entire fleet." Spot is great for fault-tolerant workloads, but if the instances in the ASG are serving production web traffic, Spot interruption would cause outages. The right pattern is On-Demand or Reserved for the baseline, Spot for burst capacity.
The exam also tests the pattern where the minimum capacity in the ASG is set too high. If the business only needs 2 instances during off-peak hours but the ASG minimum is 10, the excess 8 instances run 24/7 doing nothing. Reducing the minimum to 2 and trusting scale-out to handle traffic spikes is the cost-optimization answer.
Practice Question Sets
Working through real SAA-C03 questions is the fastest way to lock in how the exam phrases these scenarios. Pick a session that fits your time: