AWS Pricing Models and Cost Tools

Domain 4 (Billing, Pricing, and Support) carries 12% of the CLF-C02 score, and two of its three task statements live inside this lesson: Task 4.1 "Compare AWS pricing models": On-Demand, Reserved Instances, Spot Instances, Savings Plans, Dedicated Hosts, Dedicated Instances, Capacity Reservations, storage tiers and inbound versus outbound data transfer, and Task 4.2 "Billing, budget, cost management": Budgets, Cost Explorer, the Pricing Calculator, AWS Organizations consolidated billing, cost allocation tags and the Cost and Usage Report. Roughly six scored items come from Domain 4, and most of them are arithmetic-and-judgement questions rather than definitions. Students who fail this area rarely fail because they never heard of Savings Plans; they fail because they cannot say which tool answers which question, or because they quote a discount percentage without knowing what it commits you to.

In this lesson you will:

• state the four general pricing principles in AWS's own wording and name the three cost drivers;
• explain why the same instance type costs different amounts in different Regions and work the arithmetic;
• read the compute purchasing comparison table: commitment terms, flexibility, interruptibility and when the exam picks each option;
• walk the purchase-option decision tree from free lab to committed fleet;
• separate Standard vs Convertible RIs and Compute vs EC2 Instance Savings Plans by discount ceiling and flexibility;
• explain exactly what Spot costs you in interruption risk, and what a Dedicated Host buys over a Dedicated Instance;
• name the three Free Tier flavors with sourced examples, plus the post-July-2025 credit structure;
• list the billing dimensions of S3, EBS, RDS, Lambda, data transfer, Route 53 and CloudFront;
• price requests, egress and hosted zones in worked numeric examples;
• choose the right cost tool: Pricing Calculator, TCO/Migration Evaluator, Billing console, Cost Explorer, Budgets;
• map four real AWS customer cost stories (Box, Canva, FarEye, NASA JPL) to the lever that produced each saving, and read the sourced 2026 updates box;
• practise with 12 exam-style questions plus four interactive checks.

---

1. The four general pricing principles

1.1 The principle ladder

AWS's How AWS Pricing Works whitepaper (published 24 February 2023, and now labelled by AWS as a historical reference) compresses the whole model into one sentence: "You pay as you go, pay for what you use, pay less as you use more, and pay even less when you reserve capacity." The live pricing page adds the contract half of the deal: pay-as-you-go "without requiring long-term contracts."

Two supporting ideas complete the model. First, economies of scale: AWS says customers "Benefit from massive economies of scale", which is what produces "lower pay-as-you-go prices": the aggregated usage of all AWS customers is why the list price is already low before any commitment. Second, the three cost drivers: AWS names compute, storage and outbound data transfer as "three fundamental drivers", explicitly noting there is no charge for inbound data transfer.

1.2 Pay less by using more: tiering and what is free

"Pay less by using more" is not a slogan about discounts you must negotiate; it is tiering inside the list price. AWS's pricing page states that S3 and data transfer out are tiered, while data transfer IN is always free of charge. The exam turns this into two questions:

1. Which direction is charged? Outbound. Inbound from the internet is free, and traffic between services in the same Availability Zone is free.
2. Is the allowance per service or global? Global: see the "Did you know?" box below.

AWS also documents a fourth, quieter cost: idle. The pricing whitepaper states that shutting down unused instances can save "70 percent or more"; which is why turn it off is always a plausible correct answer in a cost-optimisation stem, and why idle-capacity cleanup comes before any commitment decision.

• 📚 Did you know? The free data-transfer allowance is 100 GB of free data transfer out per month, aggregated across all AWS services and all Regions (excluding the China Regions and AWS GovCloud); it is not 100 GB per service and not 100 GB per Region. Amazon CloudFront adds a separate 1 TB of free data transfer out every month. AWS expanded this allowance from the older per-Region figure, so old study material quoting "1 GB per Region" is out of date (AWS Free Tier data-transfer expansion page, as of Oct 2026).

1.3 Commitment is the only lever that changes the unit price downward

Tiering lowers the price of the next unit; commitment lowers the price of every unit. AWS's pricing page groups this under "Save when you commit", and Savings Plans are documented as a commitment over a one- or three-year period. Reserved Instances use the same 1-or-3-year terms with three payment options (All Upfront, Partial Upfront, No Upfront). Everything in §2 is a variation on that one sentence; which is why the exam loves asking "which option gives the deepest discount with the least flexibility?"

1.4 Prices are set per service and per Region

The exam guide's own cost-optimisation guidance (COST07-BP02) says it plainly: "Resource pricing may be different in each Region" because of land, fiber, electricity and taxes. Local Zones are priced differently from (higher than) their parent Region. The same instance type, same OS, same hour can therefore cost materially different amounts depending on where you launch.

Worked example E1: Regional spread on one instance type (as of Oct 2026). AWS's own On-Demand price feed for t3.medium on Linux (no security-group selection), retrieved 7 October 2026:

AWS bills On-Demand instances by the hour (or by the second with a 60-second minimum), and the Pricing Calculator standardises a month at 730 hours. So one t3.medium running the whole month:

$$\text{N. Virginia} = 730 \times 0.0416 = \$30.37 \qquad \text{São Paulo} = 730 \times 0.0672 = \$49.06$$

Difference: $49.06 −$30.37 = $18.69 per month, about +61%, for identical software. Exam lesson: Region is a pricing decision, but only after latency, data-residency and feature requirements are satisfied: AWS never publishes a "cheapest Region" ranking, so treat cost as a tie-breaker, not the first filter.

---

2. Compute purchasing options: the comparison the exam is built on

Domain 4 Task 4.1 names the options explicitly: On-Demand Instances, Reserved Instances, Spot Instances, AWS Savings Plans, Dedicated Hosts, Dedicated Instances, Capacity Reservations, plus storage tiers and data transfer. Below is the whole skill in one table, with every percentage dated as of Oct 2026.

2.1 The master comparison table

Every official ceiling in one reference (all "up to", as of Oct 2026):

2.2 Worked example E2: the discount ladder

Take a hypothetical On-Demand list cost of 100 units per month for the same steady workload and apply AWS's published ceilings (illustrative arithmetic, ceilings as of Oct 2026):

Two lessons ride on those numbers. First, Savings Plans and Reserved Instances are mapped 1:1 by ceiling: Compute SP ≈ Convertible RI at 66%, EC2 Instance SP ≈ Standard RI at 72%, AWS's Savings Plans FAQ says exactly that ("66% (just like Convertible RIs)" and "72% (just like Standard RIs)"). The difference between the pairs is therefore flexibility, not savings. Second, the real bill lands above these figures, because a ceiling is the best AWS will ever do for a perfectly matched workload, actual savings depend on how completely your usage matches the commitment.

2.3 Reserved Instances: the fine print the exam reuses

⚠️ A Reserved Instance is a billing discount, not a reserved server. AWS's wording is "a capacity reservation when used in a specific Availability Zone", the discount and the reservation are two separate things travelling together. And the trap family works in both directions: a Savings Plan reserves nothing, while an On-Demand Capacity Reservation reserves capacity at full price with no discount at all. If a stem says "guaranteed capacity in AZ X at a discount", the answer is an AZ-scoped Standard RI, nothing else in the table does both jobs.

2.4 Savings Plans: the two types in one view

AWS prices a Savings Plan year as 365 days and three years as 1,095 days, and offers the same All / Partial / No Upfront payment choices. AWS's own Savings Plans types page frames the choice as "flexibility to change your usage": commit to a dollar amount per hour (Compute) or to a specific family in a Region (EC2 Instance).

2.5 Spot: 90% off, with a catch the exam always tests

Spot Instances run spare EC2 capacity at "less than the On-Demand price" and are marketed as "up to 90% off". Three mechanics matter:

1. The price moves. The Spot price list is updated every 5 minutes and reflects "long-term trends in supply and demand": unlike On-Demand, whose hourly price is static.
2. Capacity can be taken back. AWS runs the instance "whenever capacity is available" and interrupts when it needs the capacity back. AWS's own Instance Advisor reports an average interruption frequency that has historically been under 5% (30-day view, all Regions and instance types, as of Oct 2026): still non-zero, which is why Spot is never the answer for a stateful, non-fault-tolerant or latency-critical workload.
3. Design for interruption. The examinable pair is fault-tolerant / stateless / restartable workloads: batch jobs, CI pipelines, big-data processing, image rendering, queue consumers that can re-read work.

• 📚 Did you know? AWS's Reserved Instances pricing page currently opens with "We recommend Savings Plans": the same page that sells you the RI is steering flexible customers elsewhere. On the exam, "AWS recommends X over Y" stems are usually testing exactly this line: Savings Plans for flexibility, RIs when you specifically want the RI construct (including an AZ-scoped capacity reservation).

2.6 Dedicated Host vs Dedicated Instance vs Capacity Reservation

These three get conflated constantly. The distinguishing axes are what is billed, what you can see, and what licence control you get:

AWS states there is no performance, security or physical difference between Dedicated Hosts and Dedicated Instances: the split is billing granularity, visibility and licensing.

2.7 The decision tree and the workload match

2.8 Reading the stem: the keyword map

Almost every Domain 4 purchasing item can be decided from two or three words in the stem. Use this map before you read the options, then eliminate anything that violates the keyword:

---

3. Free Tier: three flavors plus the 2025 restructure

3.1 The three classic types

AWS's Free Tier documentation defines three categories, and the exam asks you to tell them apart by how the clock behaves:

Two mechanical rules apply to all three: free usage is "calculated each month across all regions" and "does not accumulate", and Free Tier usage is not available in AWS GovCloud, the 100 GB data-transfer-out allowance additionally excludes the China Regions. One monthly allowance sits outside the classic three: Amazon CloudFront lists 1 TB of data transfer out plus 10,000,000 HTTP(S) requests per month, but the page does not label it Always Free or 12-month, do not assert a duration for it without re-checking (verify current before use).

3.2 What changed on 15 July 2025

The classic 12-month model is still the vocabulary of the exam, but the current sign-up experience is credit-based. As of 7 October 2026, aws.amazon.com/free offers:

• $100 in credits immediately, plus up to $100 more earned, for up to $200 over 6 months;
• a Free plan whose account "closes on its own 6 months after you open it", or when the credits run out, covering "over 90 services", with only the Always Free offerings active on a Free plan account;
• a Paid plan with the full catalog, which is also eligible for the credits;
• unchanged categories: Always free (30+ AWS services), Trials, and time-limited offers.

EC2 free-tier instance types differ by account date: accounts before 15 July 2025 see t2.micro / t3.micro on the 12-month track; accounts on or after that date get t3.micro, t3.small, t4g.micro, t4g.small, c7i-flex.large and m7i-flex.large on the 6-months-or-credits-exhausted track. Verify current before use; this is the fastest-moving area in Domain 4.

3.3 Worked example E3: what "750 hours per month" actually buys

The classic EC2 allowance is 750 hours per month of t2.micro/t3.micro (12-month era figure, as of Oct 2026). AWS's Pricing Calculator standardises a month at 730 hours, so:

$$\frac{750 \text{ free hours}}{730 \text{ hours in a month}} = 1.03 \text{ instances running 24/7}$$

So 750 hours means exactly one instance running around the clock for the month, with almost nothing to spare: two instances split across the month will breach the allowance, and so will one instance plus a leftover from a second Region, because the allowance is calculated monthly across all Regions and does not accumulate. The same arithmetic applies to any hourly free allowance: divide the hours by 730 before you promise anyone "free".

• 📚 Did you know? The Free Tier alert threshold is 85 percent, not 100%: AWS states it "notifies you over email when you exceed 85 percent of your Free Tier limit". The notification is per-service, the 100% overrun is your own job to watch, and a Free plan account has only Always Free offerings active: time-limited allowances do not exist on that plan.

---

4. What you are actually billed for: per-service dimensions

Knowing which service to use is half of Domain 4; knowing which unit that service bills in is the other half. Every price below is as of Oct 2026 and must be re-checked before you rely on it.

4.1 The billing-dimension matrix

4.2 Worked examples E4, E6, request, egress and DNS arithmetic

E4: the S3 request bill (us-east-1, as of Oct 2026). A service performs 500,000 writes and 20,000,000 reads per month:

Point: a read-heavy workload is cheap on requests, and a LIST-heavy workload is not: LIST is billed at the PUT rate. Storage GB-month and egress are additional.

E5: data transfer out with the global free allowance (as of Oct 2026). A workload egresses 600 GB to the internet in a month from us-east-1:

$$(600 - 100)_{\text{free global allowance}} \times \$0.09 = 500 \times 0.09 = \$45.00$$

Point: the 100 GB is aggregated across all AWS services and all Regions (excluding China and GovCloud); you do not get a fresh 100 GB per service, and inbound transfer costs nothing, so a chatty download-heavy architecture is an outbound problem.

E6: Route 53 in a small account (as of Oct 2026). Two hosted zones and 500,000 standard queries in a month:

$$2 \times \$0.50 + \left(\frac{500{,}000}{1{,}000{,}000} \times \$0.40\right) = \$1.00 + \$0.20 = \$1.20 \text{/month}$$

Point: DNS is billed per zone and per query, and private hosted zone queries are free; which is the standard reason a private-zone architecture is the cheaper exam answer.

4.3 The "stopped is not free" rule

The single most-tested billing fact in this section: Amazon RDS charges for provisioned storage and backup storage while the instance is stopped: only DB instance hours stop. EC2 is different: stopping an instance stops instance-hour billing, but the attached EBS volumes and snapshots keep billing. So:

Worked example E7 (illustrative). A team stops a database nightly expecting $0: if the instance hour stops but 100 GB of provisioned storage and its backups remain, the nightly stop saves only the compute line. The exam stem is usually "the team stopped the database at night to save costs; why is the bill unchanged?": the answer is storage and backup charges continue.

• 📚 Did you know? Of the three cost drivers AWS names (compute, storage, outbound data transfer), only one is direction-dependent: data transfer in is free from the internet, same-AZ traffic between AWS services is free, and everything interesting is billed on the out side: aggregated as "AWS Data Transfer Out" with a 100 GB/month free allowance across all services and Regions (except China and GovCloud), as of Oct 2026. That single sentence answers more Domain 4 data-transfer items than any table.

4.4 Data transfer: every direction, priced

Task 4.1 explicitly names "incoming and outgoing data transfer costs (cross-Region, intra-Region)". AWS's own pricing guidance treats inbound as free and outbound as the metered direction:

Worked example E8: one month of mixed transfer (us-east-1 rates, as of Oct 2026).

Point: the entire bill came from one direction (outbound) and one hop (cross-Region). Cheapest architectures on this exam move data in for free, keep it inside an Availability Zone where possible, and push it to the internet through CloudFront rather than straight from an origin.

4.5 Storage tiers: where the "storage" cost driver actually lives

Task 4.1 pairs purchasing options with "storage options and tiers". The pricing logic of every storage tier is the same trade: the lower the monthly GB-month rate, the colder the data must be and the more you pay to read it back:

Exam lesson: "we archived the data" is only a saving if the retrieval pattern matches the tier: a cold class that gets read daily bills more than the hot class it replaced, and any data that leaves AWS for the internet bills as data transfer out.

---

5. Cost tools: five questions, five answers

Task 4.2 names the toolkit: AWS Budgets, AWS Cost Explorer, the AWS Pricing Calculator, AWS Organizations (consolidated billing), cost allocation tags and the Cost and Usage Report. The exam does not test feature depth; it tests which tool answers which question, plus a handful of free/paid thresholds.

5.1 The tool ladder

5.2 Worked example E9: picking the tool before doing the math

Budget arithmetic worth memorising (as of Oct 2026): monitoring is free; the first two action-enabled budgets cost nothing; the third costs $0.10 per day, so a 30-day month on that budget is $3.00, and five such budgets are $15.00/month. Budgets without actions are free, which is the reason the correct exam answer to "set up a free spend alert" is a $0 cost budget: not a paid tier.

5.3 Inside Cost Explorer and Budgets: the mechanics the exam quotes

Cost Explorer is an analysis layer, not a billing system: AWS states you can look at 13 months of history, forecast the next 18 months, and receive "recommendations for what Reserved Instances to purchase" (user-guide figures, as of Oct 2026; marketing pages quote different windows, so verify current before use). Several mechanical details appear as distractors:

AWS Budgets is the forward-looking guardrail, and it comes in three budget types, cost, usage, and utilization/coverage, with a fixed alert vocabulary:

Organizations and tags complete the picture (Task 4.2): AWS Organizations gives you consolidated billing across member accounts, and cost allocation tags let you split a single bill by team, product or environment. The Cost and Usage Report (CUR) is the line-item evidence behind both. A practical pairing worth remembering: a $0 cost budget guards the Free Tier while AWS's own 85% Free Tier email gives you the service-level warning: two free early-warning systems for the price of none.

• 📚 Did you know? The Pricing Calculator's 730 hours per month assumption (720 hours in a 30-day month + 10) is baked into its monthly cost lines; which is why a calculator estimate and a real bill disagree in 31-day months, February, and any month where a resource ran for only part of the time. Multiply any "$/hour × 730" figure you see in study material by 1 before trusting it in a real forecast.

2026 Updates (as of October 2026)

---

Real-World Case Studies

Domain 4 is scored on judgement, and AWS's own customer stories are where that judgement was formed. Every figure below is as AWS published it on case-study pages accessed October 2026, and every percentage is a customer result, never an AWS guarantee. The exam rarely asks for the company name; it asks which lever produced the saving: architecture, commitment, interruptible capacity, or price-performance. Four levers cover every cost story AWS publishes.

Box: $2.23 million from billing dimensions, not from commitments

AWS's case study, headlined "…Unpacks over $2.23 Million in Savings", follows Box, an enterprise SaaS platform used by more than 120,000 organizations, through a series of AWS Well-Architected Framework reviews with AWS solutions architects. Notice where the money actually came from: not from a 1- or 3-year commitment, but from the §4 billing dimensions:

The mechanics were architectural: S3 and Glacier lifecycle tiering, EBS volume and snapshot hygiene, CloudTrail event filtering (pay for the events you keep), and routing traffic around internet gateways so bytes never became egress. Box's director of FinOps and SRE, speaking for the customer: "Our use of AWS best practices led to savings of over 2 million dollars, setting a new baseline…"

Exam lesson: the single largest line in this story is data transfer out, §4.1's third cost driver, and the fix cost no commitment term and no interruption risk. When a stem says "reduce spend without changing the purchase option", reach for §4's billing dimensions first; architecture comes before commitment in every cost-optimisation sequence AWS teaches.

Canva: the purchase-model mix

Canva, a SaaS design platform that has run on AWS since day one, needed cost-effective scale with different reliability tiers per user plan. AWS's cost-optimisation case study reports that Canva reduced compute costs by 46 percent in less than 2 years by deliberately mixing purchase options instead of picking one:

AWS's own page quotes the ceilings Canva leaned on: RIs at "a discount of up to 72 percent compared to On-Demand", Savings Plans at "up to 72 percent … in exchange for a 1- or 3-year hourly spend commitment", and Spot at "up to 90 percent discount" (all as of Oct 2026).

Exam lesson: production estates rarely buy a single option. "Compare AWS pricing models" (Task 4.1) means applying the §2 decision tree per workload, the floor of steady capacity on a commitment, the elastic overflow on Spot, and Canva's 46 percent is what that mix produced for Canva, not a ceiling AWS promises anyone.

FarEye: three levers pulled at once

FarEye, a SaaS logistics platform fighting thin last-mile margins, needed predictable spend. AWS's case study credits three levers layered together across 500+ Spot and On-Demand instances:

• Compute Savings Plans: the committed floor (§2.4, up to 66% ceiling);
• EC2 Spot: the interruptible overflow (§2.5, up to 90% ceiling);
• Graviton: price-performance, the fourth lever (same bill dimensions, fewer units consumed).

AWS published the outcome as compute costs reduced by 65 percent, about $1 million per year in cloud cost savings, and +30% performance on Graviton (case study accessed Oct 2026).

Exam lesson: commitment and Spot are not rivals; they are stacked: steady baseline on the Savings Plan, elastic and restartable work on Spot. What they never do is reserve capacity; only an AZ-scoped RI or a Capacity Reservation does that (§2.3).

NASA JPL: Spot where interruption is survivable

For the Perseverance mission, NASA's Jet Propulsion Laboratory became the exam's clearest example of a mixed purchasing strategy: EC2 Auto Scaling combining Spot + On-Demand + Capacity Reservations, processing about 4.4 TB of downlinked data per day into up to 70 TB of final data products. AWS quotes the Spot line verbatim: "up to a 90 percent discount compared to Amazon EC2 On-Demand pricing" (case study accessed Oct 2026).

Exam lesson: all three options appear together for a reason: Spot carries the restartable processing at up to 90% off, On-Demand covers the critical path that cannot be interrupted, and the Capacity Reservation guarantees launch room at the full On-Demand price. The discount never comes from the reservation, and the reservation is never free.

More AWS-published results: customer outcomes, not promises

⚠️ Customer results are not AWS guarantees. The single most common Domain 4 writing error is turning "Zendesk reduced costs by up to 42 percent" into "AWS guarantees 42 percent savings". AWS publishes outcomes achieved by named customers on their own estates, never a promised rate. Second error: numbers lose their dates in prep material. NASA's "almost a million dollars" is from 11 June 2012, Shutterfly's cut landed in March 2025, Netflix/Aurora's −28% is from a 27 November 2025 blog, and every percentage here is as of Oct 2026, restate them with the date, say "customer achieved", and verify current before use.

• 📚 Did you know? Amazon S3 launched on 14 March 2006 at $0.15 per GB per month, backed by roughly 1 PB across about 400 storage nodes in 15 racks and 15 Gbps of total bandwidth, with a 5 GB maximum object size. AWS's own twenty-year retrospective (AWS News Blog, 13 March 2026) says S3 now charges "slightly over 2 cents per gigabyte": about an 85% price reduction since launch, and that S3 Intelligent-Tiering has saved customers over $6 billion. That is what AWS's economies of scale (§1.1) produce over twenty years of one service: the pay-as-you-go price itself falls while the meter keeps running: prices as of Oct 2026, verify current before use.

• 📚 Did you know? Box's $2.23 million came from AWS Well-Architected Framework reviews: free best-practice guidance; which is a different animal from Trusted Advisor (checks your account's configuration against AWS's recommendations) and from an AWS Support plan (paid human help). The three are constantly conflated: Well-Architected reviews the design, Trusted Advisor checks the account, Support talks to AWS, and none of them is a purchasing option (as of Oct 2026).

---

Practice Questions