examhub .cc The most efficient path to the most valuable certifications.
Vol. I
In this note ≈ 21 min

AWS Pricing Models (On-Demand, Reserved, Spot, Savings Plans)

4,120 words · ≈ 21 min read

AWS pricing models let you match every workload to the cheapest compute contract that still meets its availability and flexibility requirements. For the CLF-C02 exam, Task Statement 4.1 asks you to compare AWS pricing models — which means knowing when On-Demand, Reserved Instances, Spot Instances, Savings Plans, Dedicated Hosts, Capacity Reservations, and the AWS Free Tier each win. These AWS pricing models are the single highest-frequency scenario pattern in Domain 4, and Savings Plans vs Reserved Instances is the most cited trap in post-exam retros.

This study guide walks through every AWS pricing model the exam asks about, uses three very different analogies in the 白話文解釋 section to lock the mental model in, dissects the Reserved Instances vs Savings Plans vs Spot Instances trap in depth, and closes with must-memorize numbers and FAQ. By the end you should be able to read any CLF-C02 scenario — "steady 24/7 workload", "fault-tolerant batch job", "predictable spend but mixed instance families" — and immediately name the correct AWS pricing model with no second-guessing.

What are AWS Pricing Models?

AWS pricing models are the menu of commercial contracts you can choose from when you consume AWS compute, database, or analytics capacity. They all sit on top of the same underlying "pay-as-you-go" principle — you rent infrastructure by the second or hour rather than buying hardware — but each AWS pricing model trades off flexibility, commitment, and discount depth differently.

The five AWS pricing models tested on CLF-C02

  1. On-Demand — pay per second or per hour, no commitment, walk away any time. Highest price per unit of compute.
  2. Reserved Instances (RI) — commit for 1 or 3 years to a specific instance configuration in exchange for up to 72% off On-Demand. Available for EC2, RDS, ElastiCache, Redshift, OpenSearch, and DynamoDB reserved capacity.
  3. Savings Plans — commit to a dollar-per-hour compute spend for 1 or 3 years in exchange for up to 72% off. More flexible than Reserved Instances; Compute Savings Plans even cover AWS Lambda and AWS Fargate.
  4. Spot Instances — bid on spare EC2 capacity for up to 90% off On-Demand, but AWS can reclaim the instance with a two-minute warning. Only for interruption-tolerant workloads.
  5. Dedicated Hosts / Dedicated Instances — physical server reserved for your account only; used for Bring-Your-Own-License (BYOL) and regulatory compliance.

Supporting constructs you also need to recognize:

  • Capacity Reservations — reserve EC2 capacity in a specific AZ without committing to a price discount.
  • AWS Free Tier — three tiers (12-month introductory, Always Free, Short-term trials) that let you explore AWS at no cost, per account.

An AWS pricing model is the commercial contract you select when consuming AWS capacity. It defines how long you commit, how flexibly AWS must deliver capacity, and therefore what discount you earn against the On-Demand list price. All AWS pricing models share the pay-as-you-go foundation; they differ only on commitment, flexibility, and interruption risk. Source ↗

Why AWS pricing models matter for the exam

Community retros consistently report that EC2 pricing scenario questions are one of the top-3 exam traps on CLF-C02. The difficulty comes from two angles: the qualifier keyword in the question ("MOST cost-effective", "LEAST flexible", "tolerates interruption") flips the correct AWS pricing model answer, and Reserved Instances vs Savings Plans look nearly identical until you drill into their flexibility scope.

Core Operating Principles — Pay-as-you-go, Commitment Discounts, and Spot

Every one of the AWS pricing models we just listed is built on three commercial primitives: pay-as-you-go, commitment discounts, and spot-market discounts.

Pay-as-you-go as the default

Pay-as-you-go means you rent infrastructure by the second (for EC2 Linux, EKS, Fargate, Lambda) or by the hour (for EC2 Windows, RDS) with zero upfront commitment and zero termination penalty. On-Demand is the reference price that every other AWS pricing model discounts from. If you see a question describing an unpredictable workload or a short-lived experiment, On-Demand is almost always the answer.

Commitment discounts — RI and Savings Plans

Reserved Instances and Savings Plans both offer commitment discounts. The trade is simple: you commit to consumption over 1 or 3 years, and AWS trades you a deep discount (up to 72% off) for that predictability. The two AWS pricing models differ only on what exactly you're committing to — a specific instance configuration (RI) or a dollar-per-hour spend (Savings Plans).

Spot market — Spot Instances

Spot Instances are AWS's way of reselling unused capacity at up to 90% off. You bid (or just accept the current Spot price) and AWS runs your EC2 instance until someone willing to pay On-Demand needs that capacity back. When that happens, you get a two-minute interruption notice and your instance is reclaimed.

Isolation pricing — Dedicated Hosts and Dedicated Instances

Dedicated Hosts and Dedicated Instances both give you physical isolation on AWS hardware. The differences are about licensing (Dedicated Hosts let you see the physical sockets/cores needed for per-socket license models like Windows Server or Oracle) and billing model. These AWS pricing models are niche on CLF-C02 — you need to recognize the use case, not memorize prices.

Memorize this one-sentence decision tree for AWS pricing models scenarios: "Can it be interrupted? → Spot. Is it steady and predictable? → Savings Plans (or RI if you're locked to one instance type). Is it unpredictable or short-term? → On-Demand. Do you need hardware isolation or BYOL? → Dedicated Hosts." Most CLF-C02 pricing scenarios are solvable by this one tree. Source ↗

On-Demand Instances — Full Flexibility, Highest Per-Hour Rate

On-Demand is the default AWS pricing model. You pay a fixed per-second or per-hour rate with no upfront cost, no commitment, and no discount. You can start and stop instances whenever you want.

When to use On-Demand

  • Unpredictable workloads — spiky traffic, ad hoc dev/test, one-off migrations where you can't forecast usage.
  • Short-term workloads under 1 year — no commitment-discount AWS pricing model is worth it if the workload won't outlast the commitment period.
  • Development and experimentation — you want to be able to tear everything down without financial penalty.
  • First workload of a new application — you need real production data before you can right-size and commit to Savings Plans.

Per-second vs per-hour billing

  • Per-second billing (60-second minimum) applies to EC2 Linux, EBS volumes, Amazon EMR, and AWS Batch on Linux.
  • Per-hour billing applies to Windows and most third-party AMIs, RDS, and a few other managed services.

This distinction rarely shows up on CLF-C02 directly, but recognize that "per-second" is an On-Demand Linux concept.

Reserved Instances — 1-Year vs 3-Year Commitment, Standard vs Convertible

Reserved Instances (RI) are one of the most tested AWS pricing models on CLF-C02. You commit to a specific EC2, RDS, ElastiCache, Redshift, or OpenSearch configuration for 1 or 3 years and receive up to 72% off On-Demand. RIs are a billing discount layered on top of running instances — they don't provision capacity separately.

Standard RI vs Convertible RI

  • Standard Reserved Instances — deepest discount (up to 72% off for 3-year all-upfront). You can change AZ, network type, and instance size within the same family, but you cannot change the instance family, OS, or tenancy.
  • Convertible Reserved Instances — slightly smaller discount (up to ~66% off for 3-year all-upfront) but you can exchange them for RIs of a different instance family, OS, or tenancy as long as the new RI is of equal or higher value.

Regional vs Zonal RIs

  • Regional RI — discount applies across all AZs in the Region, but does not reserve capacity in any specific AZ.
  • Zonal RI — discount plus capacity reservation in a specific AZ. Use zonal when you absolutely need guaranteed capacity in one AZ.

Payment options

All three payment options exist for both Standard and Convertible RIs:

  • All Upfront — pay the full 1-year or 3-year cost up front; deepest discount.
  • Partial Upfront — pay some up front, pay the rest monthly; middle discount.
  • No Upfront — pay $0 up front, full monthly payments for the commitment term; shallowest of the three RI discounts but still much cheaper than On-Demand.

Services that support Reserved Instances

Reserved Instances are available for EC2, Amazon RDS, Amazon ElastiCache, Amazon Redshift, Amazon OpenSearch Service, and Amazon DynamoDB reserved capacity. They are not available for AWS Lambda or AWS Fargate — that is a classic CLF-C02 trap and the key reason Savings Plans exist.

Reserved Instances: 1-year or 3-year commitment. Up to 72% off On-Demand. Standard (deepest discount, limited change scope) vs Convertible (slightly less discount, exchange to different family/OS). Regional (no capacity reservation) vs Zonal (capacity reservation in one AZ). Available for EC2, RDS, ElastiCache, Redshift, OpenSearch, and DynamoDB — NOT for Lambda or Fargate. Three payment options: All Upfront, Partial Upfront, No Upfront. Source ↗

Savings Plans — Compute vs EC2 Instance vs SageMaker

Savings Plans are a newer AWS pricing model (launched late 2019) designed to give you Reserved-Instance-like discounts with far more flexibility. Instead of committing to a specific instance configuration, you commit to a dollar amount of compute spend per hour (for example, $10/hour) for 1 or 3 years.

Three flavors of Savings Plans

  1. Compute Savings Plans — most flexible. Discount (up to 66%) applies to any EC2 instance family, any size, any region, any OS, any tenancy, AND also covers AWS Fargate and AWS Lambda. This is the Savings Plan most candidates should default to in scenario questions asking for maximum flexibility.
  2. EC2 Instance Savings Plans — less flexible. Discount (up to 72%, matching RIs) applies to a specific EC2 instance family in a specific region (for example, c5 in us-east-1). You can still change size, OS, AZ, and tenancy within that family/region lock. Does NOT cover Fargate or Lambda.
  3. SageMaker Savings Plans — specific to Amazon SageMaker ML instances. Up to 64% off on eligible SageMaker usage.

Savings Plans commitment mechanics

You commit to $X/hour of compute spend. As you run EC2, Fargate, or Lambda (Compute Savings Plan) usage, AWS applies the discounted Savings Plan rate to that spend until you hit $X/hour. Usage above $X/hour is billed at On-Demand. Usage below $X/hour still charges you the full $X/hour — you pay for the commitment whether you use it or not.

Payment options (same as RI)

Savings Plans offer the same three payment options: All Upfront, Partial Upfront, No Upfront. All Upfront maximizes the discount.

AWS Lambda pricing can be discounted with a Compute Savings Plan (up to 17% off). Reserved Instances do not exist for Lambda, and neither does an EC2 Instance Savings Plan. If a CLF-C02 scenario asks "How does a company save on a large AWS Lambda workload?" the correct AWS pricing model is Compute Savings Plan — not Reserved Instances, not any other Savings Plan flavor. Source ↗

Spot Instances — Up to 90% Discount, Interruption-Tolerant Only

Spot Instances are the cheapest of all AWS pricing models. You pay the current Spot price (which AWS sets dynamically based on supply and demand) and can receive up to 90% off the On-Demand price. In exchange, AWS can reclaim your instance with a two-minute interruption notice whenever they need that capacity back.

What "interruption-tolerant" means

A workload is Spot-friendly if it can handle being paused and resumed — think stateless batch jobs, fault-tolerant distributed processing (Amazon EMR, Apache Spark), CI/CD build workers, stateless web tiers behind a load balancer with enough capacity to absorb the loss, and genetic/scientific computing. If losing a node would cause data loss or user-facing errors, Spot is the wrong AWS pricing model.

Spot interruption notice

  • Two-minute warning — AWS publishes an interruption notice to your instance metadata (http://169.254.169.254/latest/meta-data/spot/instance-action) and as an EventBridge event 2 minutes before reclamation.
  • Interruption behaviors — stop (instance stopped, EBS preserved), hibernate (RAM preserved to EBS), or terminate (default; instance removed).

Spot is not the same as preemptible

A classic misconception on CLF-C02 is treating Spot Instances as a synonym for "preemptible" or "low-priority". Spot means interruption risk in exchange for a deep discount. It does not mean lower CPU priority or degraded performance — a running Spot instance performs identically to its On-Demand equivalent. The only difference is AWS's right to reclaim it.

If a CLF-C02 scenario describes a "steady 24/7 production database" or "always-on web server" and lists Spot as an option, Spot is the wrong AWS pricing model — regardless of how cheap it looks. Spot's 90% discount is only valid if the workload can survive a two-minute eviction. For always-on production, the correct AWS pricing model is Reserved Instances or Savings Plans. The trap is to pick Spot because it has the deepest discount. Source ↗

Dedicated Hosts and Dedicated Instances — Compliance and Licensing Use Cases

Dedicated Hosts and Dedicated Instances are AWS pricing models for scenarios where you cannot share physical hardware with other AWS customers.

Dedicated Hosts

  • Full physical server reserved for your account — you see the sockets, cores, and physical host ID.
  • Primary use case: BYOL (Bring Your Own License) for per-socket or per-core software licenses (Microsoft Windows Server, Microsoft SQL Server, Oracle Database, SUSE Linux Enterprise Server).
  • Secondary use case: regulatory compliance that mandates physical isolation.
  • Available as On-Demand or Reserved — a Dedicated Host Reservation gives commitment-style discounts.

Dedicated Instances

  • EC2 instances that run on hardware dedicated to a single customer account, but you don't see the physical server details.
  • Use case: multi-tenant isolation within your own account hierarchy without needing BYOL per-socket licensing.
  • Less flexible for BYOL than Dedicated Hosts.

On CLF-C02, "Dedicated Hosts" is the more commonly tested term. If a question says "BYOL" or "per-socket Windows license", the answer is Dedicated Hosts.

On-Demand Capacity Reservations — Capacity Without Pricing Commitment

Capacity Reservations let you reserve EC2 capacity in a specific Availability Zone without making a pricing commitment. This is a distinct construct from both Reserved Instances and Savings Plans:

  • Capacity Reservations reserve capacity but charge you On-Demand rates (unless combined with a Regional RI or Savings Plan, which then apply their discount on top).
  • Reserved Instances can give both a discount and (if Zonal) a capacity reservation.
  • Savings Plans give only a discount, never a capacity reservation.

Use Capacity Reservations when you must guarantee EC2 capacity in an AZ for disaster recovery, predictable event-driven traffic (Black Friday), or compliance — but don't want to lock into a 1-year or 3-year commitment.

Data Transfer Pricing — Inbound Free, Outbound Charges

Data transfer is not an AWS pricing model in the same sense as On-Demand or Reserved Instances, but it is a line item on every AWS bill and the exam expects you to know the basics.

  • Inbound data transfer (into AWS Regions) — almost always free.
  • Outbound data transfer (out of AWS to the internet) — charged per GB, decreasing with volume.
  • Cross-region data transfer — charged per GB; varies by region pair.
  • Same-region, cross-AZ traffic — $0.01/GB in each direction.
  • Within a single AZ (private IP) — free.

Data transfer savings are a major driver for CloudFront (cached at edge) and VPC design choices (keeping traffic within an AZ).

Storage Pricing Tiers — Cost Ladder

Storage cost tiers are loosely analogous to compute AWS pricing models — you pay less per GB the less frequently you access the data:

  • Amazon S3 Standard — highest per-GB storage cost, lowest retrieval cost, milliseconds access.
  • Amazon S3 Standard-IA (Infrequent Access) — lower per-GB, retrieval fee per GB accessed, milliseconds access.
  • Amazon S3 Glacier Instant Retrieval — even lower storage, instant retrieval, higher retrieval fee.
  • Amazon S3 Glacier Flexible Retrieval — very low storage, minutes-to-hours retrieval.
  • Amazon S3 Glacier Deep Archive — cheapest, 12+ hour retrieval.

Storage tiering is covered more deeply in the storage-services topic; for pricing-models, you just need to know that "cheaper storage class = longer retrieval time = higher retrieval fee per GB".

AWS Free Tier — Three Categories You Must Know

The AWS Free Tier is not one of the five headline AWS pricing models, but the exam guide explicitly calls it out under 4.1. Free Tier is how AWS lets you explore services at zero cost. There are three distinct categories:

12-Month Introductory Free Tier

Available for the first 12 months after account creation. Examples:

  • Amazon EC2 — 750 hours/month of t2.micro or t3.micro instances (Linux or Windows).
  • Amazon S3 — 5 GB of Standard storage, 20,000 GET requests, 2,000 PUT requests.
  • Amazon RDS — 750 hours/month of db.t2.micro, db.t3.micro, or db.t4g.micro.
  • Amazon CloudFront — 1 TB of data transfer out.
  • AWS CloudTrail — 1 copy of management events.

Always Free

Available forever, for all AWS customers, not just new accounts. Examples:

  • AWS Lambda — 1 million requests and 400,000 GB-seconds of compute per month.
  • Amazon DynamoDB — 25 GB of storage and 25 RCU/25 WCU per month.
  • Amazon CloudWatch — 10 custom metrics and 10 alarms.
  • AWS X-Ray — 100,000 traces recorded per month.

Short-term Trials

Specific services offer 30, 60, or 90-day trials from the date of first use. Amazon Redshift, AWS Inspector, and Amazon GuardDuty all have trial periods.

The AWS Free Tier is a per-AWS-account benefit, not a per-IAM-user benefit. If your company creates 100 IAM users in one AWS account, they collectively share one set of Free Tier allowances — not 100 sets. This is a common CLF-C02 trap. The 12-month clock also starts when the AWS account is created, not when you start using the service. Source ↗

Reserved Instances vs Savings Plans vs Spot — The CLF-C02 Canonical Trap

This is the single highest-frequency AWS pricing models trap on the exam. Let's build a clean decision matrix.

Decision factors across the three commitment AWS pricing models

Factor Reserved Instances Savings Plans (Compute) Savings Plans (EC2 Instance) Spot Instances
Commitment period 1 or 3 years 1 or 3 years 1 or 3 years None
What you commit to Specific instance config $/hour spend across EC2/Fargate/Lambda $/hour spend in one instance family + region Nothing
Max discount 72% 66% 72% 90%
Change instance family? Only with Convertible RI Yes (free) No N/A
Covers Fargate? No Yes No N/A
Covers Lambda? No Yes No No
Capacity reservation? Only Zonal RI No No No
Interruption risk? None None None Yes (2-min notice)

Scenario → correct AWS pricing model mapping

  • "Steady 24/7 web server, same instance type for 3 years" → Standard Reserved Instance (3-year, All Upfront for deepest discount) OR EC2 Instance Savings Plan. RI is usually cheapest when you truly never change instance family.
  • "Predictable $10/hour spend, but we might switch from c5 to m5 next quarter" → Compute Savings Plan.
  • "Large AWS Lambda invocation volume that's been steady for 6 months" → Compute Savings Plan. Lambda cannot use RI.
  • "Batch image-processing job that can retry on interruption" → Spot Instances.
  • "Temporary 2-week load test" → On-Demand. Not worth a 1-year commitment.
  • "BYOL Microsoft SQL Server per-socket license" → Dedicated Hosts.
  • "Guarantee EC2 capacity in us-east-1a for DR failover without a pricing commitment" → On-Demand Capacity Reservation.

A common wrong assumption is that Savings Plans work like Zonal Reserved Instances — both give a discount, so both must reserve capacity. They don't. Savings Plans are a pure billing discount with zero capacity guarantee. If you need EC2 capacity guaranteed in a specific AZ (for example, for disaster recovery), use a Zonal Reserved Instance or an On-Demand Capacity Reservation. Savings Plans + Capacity Reservation is the combined pattern when you want both a discount and guaranteed capacity. Source ↗

Key Numbers and Must-Memorize Facts

These are the numbers the CLF-C02 exam questions reference. If a scenario contains one of these percentages or durations, one of the AWS pricing models is almost certainly the intended answer.

  • On-Demand — 0% discount baseline. Per-second billing (Linux) or per-hour billing (Windows).
  • Reserved Instances — up to 72% off On-Demand. Commitment: 1 year or 3 years.
  • Savings Plans (Compute) — up to 66% off. Commitment: 1 year or 3 years.
  • Savings Plans (EC2 Instance) — up to 72% off. Commitment: 1 year or 3 years.
  • Spot Instances — up to 90% off. No commitment. 2-minute interruption notice.
  • Dedicated Hosts — no standard discount percentage; priced as full physical server rental.
  • Free Tier EC2 — 750 hours/month of t2.micro or t3.micro for 12 months.
  • Free Tier Lambda — 1 million requests + 400,000 GB-seconds per month, forever.
  • Free Tier S3 — 5 GB Standard storage for 12 months.
  • Payment options (RI and Savings Plans) — All Upfront, Partial Upfront, No Upfront.

Five AWS pricing models in ascending discount, descending flexibility: On-Demand (0% off, most flexible) → Compute Savings Plan (66% off, covers EC2 + Fargate + Lambda) → EC2 Instance Savings Plan (72% off, locked to family + region) → Reserved Instance (72% off, locked to exact config) → Spot (90% off, interruptible). Plus Dedicated Hosts for BYOL/compliance and Capacity Reservations for AZ-guaranteed capacity. Source ↗

Pricing Models vs Cloud Economics — Scope Boundary (4.1 vs 1.4)

CLF-C02 Task 4.1 (this topic) covers AWS pricing models — the specific commercial contracts you select. Task 1.4 (cloud-economics) covers the higher-level financial concepts like CapEx vs OpEx, Total Cost of Ownership (TCO), economies of scale, and rightsizing. A useful boundary: if the question is "which pricing model saves the most for this workload?" it's 4.1. If the question is "why does cloud convert CapEx to OpEx?" it's 1.4. Both topics share the AWS Pricing Calculator as a supporting tool.

Compare also to billing-budget-cost-management (Task 4.2), which covers AWS Cost Explorer, AWS Budgets, AWS Pricing Calculator, Cost Allocation Tags, and AWS Organizations consolidated billing — the tools you use to monitor what your AWS pricing models are actually costing you.

Practice Question Links — Task 4.1 Mapped Exercises

Scenarios to rehearse before the CLF-C02 exam:

  1. A company runs a steady-state production database 24/7 for at least the next 3 years on a c5.xlarge. Which AWS pricing model minimizes cost? → 3-year Standard Reserved Instance, All Upfront.
  2. A batch video-transcoding job can tolerate interruption and retry. Which AWS pricing model offers the deepest discount? → Spot Instances.
  3. A fintech company runs predictable AWS Lambda workloads around $5,000/month and wants to reduce cost. Which AWS pricing model applies? → Compute Savings Plan.
  4. A SaaS company runs EC2 workloads but expects to change instance families every 6 months. Which AWS pricing model gives commitment discounts with family flexibility? → Compute Savings Plan (or Convertible RI as a secondary answer).
  5. A healthcare customer must run Microsoft SQL Server with a per-socket BYOL license. Which AWS pricing model is required? → Dedicated Hosts.
  6. Which AWS pricing model supports short-term experimentation with no commitment? → On-Demand.
  7. Which AWS pricing model reserves EC2 capacity in a specific AZ without a pricing commitment? → On-Demand Capacity Reservation.
  8. Which AWS Free Tier category offers AWS Lambda's 1M requests per month forever? → Always Free.

FAQ — AWS Pricing Models Top 6 Questions

1. What's the single biggest difference between Reserved Instances and Savings Plans?

Reserved Instances lock you to a specific EC2 instance configuration (family, size, region, OS, tenancy) in exchange for up to 72% off. Savings Plans lock you to a dollar-per-hour spend amount and let the discount apply flexibly across instance types, and (for Compute Savings Plans) even across EC2, Fargate, and Lambda. RI = specific config commitment. Savings Plans = spend commitment with family flexibility. Both are 1-year or 3-year AWS pricing models with identical payment options.

2. Can I use both Reserved Instances and Savings Plans in the same account?

Yes. AWS applies Reserved Instance discounts first (because they're tied to specific instance usage), then applies Savings Plans discounts to remaining eligible usage, then charges On-Demand for anything uncovered. You can mix AWS pricing models freely — in fact, sophisticated FinOps teams often combine a Reserved Instance baseline for rock-solid workloads with Compute Savings Plans for flexibility on top.

3. What happens to a Spot Instance when AWS reclaims capacity?

AWS sends a Spot interruption notice to your EC2 instance metadata and as an EventBridge event, giving you approximately 2 minutes to wrap up work. The default interruption behavior is "terminate" (instance removed, ephemeral storage lost, EBS volumes retained depending on DeleteOnTermination). You can configure "stop" or "hibernate" behaviors for EBS-backed Spot Instances so state is preserved. Spot is one of the AWS pricing models only suited for fault-tolerant workloads that can handle this eviction gracefully.

4. Do Reserved Instances cover AWS Lambda or AWS Fargate?

No. Reserved Instances only apply to EC2, RDS, ElastiCache, Redshift, OpenSearch, and DynamoDB reserved capacity. For AWS Lambda and AWS Fargate, the correct AWS pricing model for commitment discounts is the Compute Savings Plan. This is one of the most common CLF-C02 traps — if the question mentions Lambda or Fargate, eliminate every Reserved Instance option immediately.

5. Is the AWS Free Tier unlimited or does it expire?

The AWS Free Tier has three tiers with different durations: the 12-month introductory tier expires 12 months after account creation; the Always Free tier (Lambda 1M requests/month, DynamoDB 25 GB, and others) never expires; short-term trials run for 30, 60, or 90 days from first use of the specific service. Free Tier allowances are per AWS account, not per IAM user — 100 users in the same account share one Free Tier allowance.

6. If a workload is steady 24/7, why might On-Demand still be the right AWS pricing model?

Even a steady 24/7 workload can justify On-Demand in specific cases: the workload will run less than 12 months (shorter than any RI or Savings Plan commitment term), you're in the early validation phase and haven't confirmed the right instance family yet, or you're testing a new architecture and might refactor to Lambda or Fargate in 6 months. Commitment-based AWS pricing models only pay back if the commitment period is shorter than the workload's remaining lifetime.

Further Reading

Official sources