AWS Cloud Architect & Developer · AWS Cloud Architecture
Billing, Pricing and Cost Optimization
AWS pricing models, cost management tools and strategies to optimize spend.
Seven concepts on how AWS charges and how to control it — purchase models first, then the tools that watch spend, then the arithmetic that turns hours and rates into a bill.
- AWS Cloud Architect & Developer
- Easy level
- 7 concepts
- 5 practice questions
1EC2 purchase options
EC2 is priced five ways, and the exam question is always which option fits the workload. On-Demand bills per second with no commitment — the default for unknown or short-lived use. Reserved Instances and Savings Plans trade a one- or three-year commitment for up to about 72% off On-Demand on steady baseline compute. Spot offers up to about 90% off for workloads that tolerate a two-minute reclaim warning. Dedicated Hosts reserve physical servers for compliance or BYOL licensing.
Figure. Three bars on the same baseline, heights set to relative On-Demand cost: full height for On-Demand, about one-tenth for Spot at ~90% off, about 28% for a three-year all-upfront Reserved Instance at ~72% off. The spread is the whole point — commitment depth buys depth of discount.
How to choose
- Steady baseline24/7 predictable load → Reserved Instances or a Compute Savings Plan on that baseline.
- Interruptible burstBatch, rendering, analytics that can checkpoint → Spot for the deepest discount.
- Unknown or briefSpikes, experiments, overflow → On-Demand; never commit capacity you may not use.
| Model | Commitment | Discount vs On-Demand | Best for |
|---|---|---|---|
| On-Demand | None | 0% (baseline) | Short-term, unpredictable |
| Reserved Instance | 1 or 3 years, instance-specific | Up to ~72% (3 yr, all upfront) | Steady, known instance type |
| Savings Plan | 1 or 3 years, $/hr committed | Similar to RI, more flexible | Steady spend across EC2/Fargate/Lambda |
| Spot | None; reclaimable | Up to ~90% | Fault-tolerant, flexible batch |
| Dedicated Host | Physical server | Varies | Compliance, per-socket licensing |
A nightly ETL job can restart from its last checkpoint if interrupted. Which purchase model minimizes cost?
- On-Demand Instances
- Spot Instances
- Three-year Reserved Instances
Checkpointing makes the job fault-tolerant, which is exactly what Spot requires — up to ~90% off in exchange for a possible two-minute reclaim. Reserved Instances fit a steady 24/7 baseline, not a nightly batch. On-Demand is safe but leaves the largest discount on the table.
2Savings Plans flexibility
A Compute Savings Plan commits you to a consistent dollars-per-hour of usage for one or three years and applies the discount across EC2, AWS Fargate, and Lambda — regardless of instance family, size, operating system, or Region. That flexibility is the trade: Reserved Instances lock to a specific instance type and Region, often saving a bit more on that exact shape but wasting money if the fleet changes.
Figure. Compute Savings Plans flex across instance family/region/OS more than EC2 Instance Plans.
How it works
- Measure baselineUse Cost Explorer to find the steady $/hr of compute you actually run month after month.
- Commit that rateBuy a Savings Plan covering that hourly spend — not peak, not wishful growth.
- Apply automaticallyAny matching On-Demand usage in EC2, Fargate, or Lambda draws from the plan until the commitment is exhausted.
| Dimension | Compute Savings Plan | Standard Reserved Instance |
|---|---|---|
| Commitment unit | $/hr of compute | Specific instance type + Region |
| Flexibility | Any family, size, OS, Region | Locked to what you bought |
| Services covered | EC2, Fargate, Lambda | EC2 (or RDS with a separate plan) |
| Risk if fleet changes | Low — discount follows usage | High — unused RI is wasted spend |
3Cost visibility tools
Three services answer three different questions about money. AWS Cost Explorer visualizes historical spend, forecasts trends, and recommends Savings Plans or Reserved Instances — it tells you where dollars already went. AWS Budgets sets cost or usage thresholds and sends alerts when actual or forecasted spend crosses them — it warns before or as you overspend. The Cost and Usage Report exports line-item billing detail to S3 for accounting, chargeback, or analysis in Athena.
Three named jobs: Cost Explorer for trends, Budgets for thresholds and alerts, Cost and Usage Report for line-item export. The three-row tools table is the lookup; boxes labelled with service names would only repeat those rows.
Which tool when
- Analyze past spendCost Explorer — filter by service, tag, or time range and drill into the drivers.
- Alert on a limitAWS Budgets — set a monthly cap and notify at 80% actual or when forecast exceeds the cap.
- Export every line itemCost and Usage Report to S3 — the audit trail finance or Athena needs.
| Tool | Question it answers | Typical action |
|---|---|---|
| Cost Explorer | Where did we spend? | Find top services, forecast, buy RIs/SPs |
| AWS Budgets | Are we about to exceed a limit? | Email/SNS alert at threshold |
| Cost and Usage Report | What is every billable line? | Export to S3, query in Athena |
A team must receive email when projected monthly spend will exceed $5,000 before the month ends. Which service should they configure?
- AWS Cost Explorer
- AWS Budgets with a forecasted alert
- Cost and Usage Report
Budgets supports forecasted-to-exceed alerts tied to SNS or email — exactly the early warning the scenario asks for. Cost Explorer can show the projection but does not push alerts on its own. The Cost and Usage Report is a historical export, not a notification channel.
4Data transfer and egress
Data transfer pricing is asymmetric: inbound traffic to AWS is generally free, but outbound traffic to the internet and cross-Region replication are charged per gigabyte. That asymmetry is the trap — a cheap compute bill paired with heavy user downloads or cross-Region sync can dominate spend. Keeping traffic inside one Availability Zone or Region, or fronting static assets with Amazon CloudFront, reduces egress charges.
Figure. Internet egress is the expensive direction; same-region private paths and CloudFront patterns cut transfer cost.
How charges apply
- Inbound from internetData entering AWS from the public internet is typically free.
- Outbound to internetData leaving AWS to end users is billed per GB — often the surprise line item.
- Cross-Region / cross-AZTraffic between Regions (and many cross-AZ paths) is billed per GB — never treat it as free inbound.
A media site serves large video files to global users directly from an S3 bucket in us-east-1. Which cost is most likely to grow fastest as traffic rises?
- S3 storage per GB-month
- Data transfer OUT to the internet
- Inbound upload to the bucket
Every download is outbound egress, priced per GB and scaling linearly with viewers. Storage grows with catalogue size, not view count. Inbound uploads are generally free, so they are not the scaling cost here.
5Free Tier allowances
The AWS Free Tier bundles three kinds of allowance. Twelve-month free tier credits apply for the first year after account creation — for example 750 hours per month of t2/t3.micro EC2 in eligible Regions. Always-free services offer ongoing monthly limits that never expire, such as 25 GB of DynamoDB storage. Short-term trials give limited free use of specific services for a fixed period after first use, independent of the 12-month clock.
Free Tier has three clocks — Always Free, 12-months Free, and Trials — each with different service caps. The category table carries the policy; a calendar graphic would decorate without changing which bucket a stem falls into.
Three allowance types
- 12-month free tierStarts at account creation; common for new EC2, RDS, and S3 usage caps.
- Always freeNever expires — Lambda requests, DynamoDB storage, and similar ongoing caps.
- Short-term trialPer-service trial window from first use — read the service page for duration.
| Type | Clock | Example |
|---|---|---|
| 12-month free tier | First 12 months of account | 750 h/mo t2/t3.micro EC2 |
| Always free | Never expires | 25 GB DynamoDB storage |
| Short-term trial | Fixed days from first use | Service-specific trial period |
An account is 18 months old. Which Free Tier category can still apply to new Lambda invocations within the published monthly limit?
- 12-month free tier
- Always free
- Short-term trial only
Always-free limits do not expire with the account's first year. The 12-month tier ended at month 12. Short-term trials are per-service and time-boxed, not the standing Lambda allowance.
6Consolidated billing
AWS Organizations lets you combine multiple member accounts under one payer account with consolidated billing. All accounts roll up into a single bill, volume discounts apply to the organization's combined usage, and Reserved Instance and Savings Plan benefits automatically share across linked accounts that match the commitment.
Figure. Organizations consolidated billing rolls member usage to the management account for one invoice and volume discounts.
What consolidation buys
- One billThe management account receives one invoice covering every linked member account.
- Volume pricingCombined usage can cross tier thresholds — for example S3 storage tiers — that no single account would reach alone.
- Shared commitmentsAn RI or Savings Plan bought in any linked account discounts matching usage in the others.
Account A bought a Reserved Instance but Account B runs the matching instance type. Under consolidated billing, what happens?
- Account B pays full On-Demand; the RI is wasted
- The RI discount applies to Account B's matching usage
- Account B must be moved into Account A to share the RI
Consolidated billing shares RI and Savings Plan benefits across linked accounts automatically — Account B draws the discount without moving resources. The RI is not stranded on Account A unless the instance type or Region truly does not match.
7Monthly compute cost
On-Demand EC2 cost is instances times hours times the hourly rate. AWS billing months use 730 hours (365 ÷ 12 × 24). Multiply, then layer discounts: a Compute Savings Plan applies its percentage off the On-Demand total for covered usage, not off an arbitrary flat fee.
Figure. Compute bills multiply instance count, hours (or seconds), and the rate tier — units must match.
How to compute
- Hours in the month730 h for a full month of 24/7 run time unless the question states otherwise.
- Per-instance cost730 × hourly rate — one instance running continuously for the month.
- Fleet totalMultiply by instance count; then apply any Savings Plan or RI discount to covered usage.
Baseline fleet with a Savings Plan
Four m5.xlarge instances run 24/7 for one month. On-Demand rate is $0.192/h. A Compute Savings Plan covers this usage at 40% off On-Demand. Find the On-Demand bill and the discounted bill.
- 730 h × $0.192/h$140.16 per instance
- 4 instances × $140.16$560.64 On-Demand
- 40% Savings Plan: $560.64 × 0.60$336.38
Pro tip. Match commitment to the steady baseline only — if the fleet later shrinks, unused Savings Plan commitment is still owed. Pair this arithmetic with Cost Explorer coverage reports so the discount matches actual hours.
One t3.medium runs 24/7 for a month at $0.0416/h. The On-Demand cost is closest to
- $30.37
- $249.60
- $30.37 × 730
730 × $0.0416 = $30.368 ≈ $30.37. $249.60 forgets the 730-hour month and multiplies by calendar days only. The third option repeats the correct product as a distractor shape — the question asks for the cost, not the formula.
Notes
- EC2 Purchase Options: On-Demand (pay per second, no commitment), Reserved Instances and Savings Plans (discount for 1/3-year commitment), Spot (deep discount for interruptible workloads), and Dedicated Hosts.
- Savings Plans: Commit to a consistent dollars-per-hour of usage for 1 or 3 years to save across EC2, Fargate, and Lambda more flexibly than Reserved Instances.
- Cost Visibility Tools: Cost Explorer visualizes and forecasts spend, AWS Budgets alerts on thresholds, and the Cost and Usage Report gives line-item detail.
- Free Tier: AWS offers 12-month, always-free, and short-term trial free-tier allowances (e.g., 750 hours/month of t2/t3.micro for a year).
- Consolidated Billing: AWS Organizations aggregates billing across accounts, sharing volume discounts and Reserved Instance/Savings Plan benefits.
Formulas
- Spot savings: up to 90% off On-Demand; instances can be reclaimed with a 2-minute warning.
- Reserved Instances: up to 72% savings vs On-Demand for a 3-year, all-upfront commitment.
- Data transfer: inbound data transfer is generally free; outbound to the internet and cross-Region transfer are charged.
- S3 Intelligent-Tiering: auto-moves objects between access tiers with a small per-object monitoring fee and no retrieval charges.
- Compute Savings Plans: apply across EC2, Fargate, and Lambda regardless of instance family, size, or Region.
Exam traps & shortcuts
- Steady, predictable baseline compute => Reserved Instances/Savings Plans; spiky/interruptible => Spot; short-term/unknown => On-Demand.
- To alert before overspending, use AWS Budgets; to analyze where money went, use Cost Explorer.
- For unknown or changing access patterns on S3, choose Intelligent-Tiering to auto-optimize without lifecycle guesswork.
- Data transfer OUT to the internet is the sneaky cost - keep traffic within a Region/AZ or use CloudFront to reduce egress.
Reference tables
Match workload shape to model before the exam scenario names a service.
| Workload signal | Model |
|---|---|
| Steady 24/7 baseline | Reserved Instance or Savings Plan |
| Interruptible, checkpointed batch | Spot |
| Spiky or unknown duration | On-Demand |
| Physical isolation / BYOL | Dedicated Host |
Beyond picking the right purchase model.
| Problem | Tool or tactic |
|---|---|
| Unpredictable S3 access patterns | S3 Intelligent-Tiering (small monitoring fee, no retrieval charge) |
| Surprise egress bill | CloudFront caching, keep traffic in-Region |
| RI/SP bought but unused | Cost Explorer coverage report; right-size commitment |
| Multi-account spend invisible | Organizations consolidated billing + Cost Explorer by linked account |
Recap
Read this before a Cloud Practitioner or Solutions Architect billing question.
- Models
- Steady baseline → Savings Plan/RI; interruptible → Spot; unknown → On-Demand. Spot saves up to ~90%, three-year RI up to ~72% — both have strings attached.
- Tools
- Cost Explorer shows where money went; Budgets alerts before you blow a cap; CUR exports every line to S3.
- Egress
- Inbound is free; outbound to the internet and cross-Region traffic are not. That asymmetry is the sneaky bill.
- Free Tier
- Three clocks: 12-month from account creation, always-free forever, short trial per service.
- Organizations
- One bill, volume tiers, RI and Savings Plan benefits shared across linked accounts.
- Arithmetic
- Monthly On-Demand = instances × 730 h × $/h; apply commitment discount to that total.
Practise Billing, Pricing and Cost Optimization
Reading is free and needs no account. Practice, mocks and progress live in the app.
- 5 exam-style questions on this topic, with explanations
- A 5-question practice set that ends the chapter
- Timed mocks scored with the real marking scheme
- Readiness tracked per topic, kept on your device