AWS Pricing Models Explained: On-Demand, Savings Plans, Reserved, Spot, and the Free Tier

AWS offers several pricing models that trade flexibility for discounts. On-Demand charges full price with no commitment, Reserved Instances and Savings Plans exchange a one-year or three-year commitment for significant savings, Spot Instances sell spare capacity at steep discounts that AWS can reclaim at any time, and the AWS Free Tier lets you try many services at no charge. Choosing the right model for each workload is the core cost-optimization skill the CLF-C02 exam measures.

Pay-as-you-go: the foundation

Every AWS pricing model builds on the same principle: pay for what you use, when you use it, with no long-term contract required. There is no upfront hardware purchase and no termination fee. Stop the resource and the charges stop. This consumption model replaces capital expense (buying servers) with variable operating expense, which is one of the headline economic benefits of cloud computing. Two related ideas show up in exam questions: prices generally fall as AWS gains economies of scale, and you can pay less per unit as you use more (tiered pricing, such as S3 storage getting cheaper per GB at higher volumes).

On-Demand: maximum flexibility, highest rate

On-Demand pricing bills compute by the second or hour (depending on the service and operating system) at the published rate, with zero commitment. It is the right choice for unpredictable workloads, short-term projects, development and testing, and anything you are running for the first time and cannot yet forecast. The trade-off is simple: you pay the most per hour in exchange for the freedom to start and stop whenever you want.

Commitment discounts: Reserved Instances and Savings Plans

Both of these models reward a commitment of one or three years, with deeper discounts for three years and for paying more upfront (payment options are All Upfront, Partial Upfront, and No Upfront).

Reserved Instances (RIs) apply to specific services such as Amazon EC2, Amazon RDS, Amazon Redshift, and Amazon OpenSearch Service. A Standard RI locks in attributes like instance family and Region for the largest discount. A Convertible RI allows you to exchange it for different instance configurations during the term, at a somewhat smaller discount. An RI is a billing discount applied to matching usage, not a separate physical machine.

Savings Plans are the newer, more flexible commitment model. Instead of committing to an instance type, you commit to spending a certain amount per hour (for example, a fixed dollar rate per hour) for the term. Compute Savings Plans are the most flexible: the discount follows your usage across EC2 instance families, sizes, Regions, and even across AWS Fargate and AWS Lambda. EC2 Instance Savings Plans commit to a specific instance family in a Region for a larger discount. For steady, predictable compute usage, Savings Plans are usually the modern recommendation because the discount survives architecture changes.

Model	Commitment	Typical discount depth	Flexibility	Best for
On-Demand	None	None (baseline price)	Total	Unpredictable or short-lived workloads
Standard RI	1 or 3 years, specific instance attributes	Deepest of the RI family	Low	Stable workloads that will not change shape
Convertible RI	1 or 3 years, exchangeable	Slightly less than Standard	Medium	Long commitments with room to change instance types
Compute Savings Plan	1 or 3 years, $/hour spend	Comparable to Convertible RI	High (EC2, Fargate, Lambda)	Steady compute spend across changing architectures
Spot	None	Steepest available	AWS can reclaim capacity	Fault-tolerant, interruptible jobs

Spot Instances: spare capacity at a deep discount

Spot Instances let you bid on unused EC2 capacity at discounts that can reach the large majority of the On-Demand price. The catch defines the use case: AWS can interrupt a Spot Instance with a two-minute warning when it needs the capacity back. Spot is therefore only appropriate for workloads that tolerate interruption: batch processing, big data analysis, containerized jobs that can restart, rendering, and CI/CD workers. It is never the answer for a production database or anything that must not stop mid-task.

Dedicated Hosts and Dedicated Instances

Some organizations need physical isolation. A Dedicated Host gives you an entire physical server, with visibility into sockets and cores, which matters for bring-your-own-license (BYOL) software that is licensed per socket or per core, and for strict compliance requirements. Dedicated Instances run on hardware dedicated to your account but without that host-level visibility or placement control. Both cost considerably more than shared tenancy, so the exam expects you to pick them only when licensing or compliance demands it.

The AWS Free Tier

The AWS Free Tier historically comes in three flavors, and the exam tests the distinction:

• Always free offers never expire: examples include a monthly allowance of AWS Lambda requests and Amazon DynamoDB storage.
• 12 months free offers apply for the first year after account creation: the classic example is a monthly allotment of hours on a small EC2 instance type.
• Trials are short-term offers that start when you activate a particular service, such as a fixed number of days of Amazon Inspector.

Be aware that AWS revamped the free tier for accounts created from mid-2025 onward: new accounts choose a free plan that grants promotional credits usable across services for a limited period, rather than the older 12-month per-service allowances. Always-free offers continue to exist. For the exam, focus on the concept that the Free Tier lets you explore services at no cost within defined limits, and that exceeding those limits produces normal charges.

To estimate what a workload will cost before you build it, use the AWS Pricing Calculator. To analyze and control spending after you deploy, reach for the billing toolset covered in AWS Cost Explorer vs Budgets. Purchasing options for compute specifically are covered in more depth in EC2 instance types and purchasing options.

How the CLF-C02 exam tests this

• A scenario describes a workload with a steady, predictable baseline running around the clock for years, and asks for the most cost-effective option. The answer is a Savings Plan or Reserved Instances, not On-Demand.
• A scenario describes a batch or data-processing job that can be stopped and restarted without harm, and asks for the cheapest compute. That is Spot Instances; the interruption tolerance in the stem is your signal.
• A question mentions server-bound software licenses (per-socket or per-core) or a regulatory need for a physical server not shared with other customers. That points to Dedicated Hosts.
• A question asks which tool estimates costs for a planned architecture before deployment. That is the AWS Pricing Calculator, not Cost Explorer (which analyzes existing spend).
• A distinction question asks the difference between Compute Savings Plans and EC2 Instance Savings Plans, or between Standard and Convertible RIs: in both pairs, one option is more flexible and the other discounts more deeply.

Quick reference

• Pay-as-you-go is the baseline: no contracts, no termination fees, stop paying when you stop using.
• On-Demand: no commitment, highest per-hour rate, best for unpredictable workloads.
• Reserved Instances: 1 or 3 year commitment to specific services; Standard = deepest discount, Convertible = exchangeable.
• Savings Plans: commit to $/hour of spend; Compute plans cover EC2, Fargate, and Lambda across Regions.
• More upfront payment and longer terms mean bigger discounts (All, Partial, No Upfront).
• Spot: deepest discount, reclaimable by AWS with a two-minute warning, interruptible workloads only.
• Dedicated Hosts: physical server for BYOL licensing and compliance; Dedicated Instances isolate hardware without host visibility.
• Free Tier: always-free, 12-months-free, and trial offers (newer accounts use a credit-based free plan); Pricing Calculator estimates costs before deployment.