EC2 Instance Types and Purchasing Options: On-Demand, Reserved, Spot, and Savings Plans

Amazon Elastic Compute Cloud (EC2) gives you resizable virtual servers, and you pay for them through one of several purchasing options: On-Demand for flexibility, Reserved Instances and Savings Plans for commitment-based discounts, Spot Instances for deep discounts on interruptible capacity, and Dedicated Hosts or Dedicated Instances for single-tenant hardware. The right choice depends on two questions the CLF-C02 exam asks over and over: how predictable is the workload, and can it tolerate interruption?

What EC2 instance types are

An EC2 instance type defines the hardware profile of a virtual server: how much virtual CPU, memory, storage, and network capacity it gets. Instance types are organized into families, each optimized for a category of work:

• General purpose (for example, the M and T families): balanced CPU and memory, good default for web servers and small databases. Burstable T instances accumulate CPU credits during quiet periods and spend them during spikes.
• Compute optimized (C family): high CPU relative to memory, suited to batch processing, media transcoding, and high-performance web tiers.
• Memory optimized (R and X families): large amounts of RAM for in-memory databases, caches, and real-time analytics.
• Storage optimized (I and D families): fast local NVMe or dense HDD storage for workloads that hammer disk, such as data warehousing nodes and distributed file systems.
• Accelerated computing (P, G, and Inf families): GPUs or purpose-built chips for machine learning training, inference, and graphics.

For CLF-C02 you do not need to memorize specific instance sizes. You need to match a workload description (“in-memory cache”, “video encoding”, “ML training”) to the right family category.

The purchasing options and how they work

Choosing an instance type answers “what hardware”. Purchasing options answer “how you pay for it”.

On-Demand Instances. Pay by the second (with a minimum) with no commitment. You can launch and terminate at will. This is the most expensive per-hour option but the most flexible, and it is the right answer for short-term, spiky, or unpredictable workloads and for anything being developed or tested for the first time.

Reserved Instances (RIs). You commit to a 1-year or 3-year term for a specific instance family in exchange for a significant discount versus On-Demand. Payment can be All Upfront (biggest discount), Partial Upfront, or No Upfront. Standard RIs give the deepest discount but limited flexibility; Convertible RIs allow you to exchange for different instance families during the term at a somewhat smaller discount. RIs are a billing discount applied to matching running instances, not a separate kind of server.

Savings Plans. You commit to a consistent amount of compute spend (measured in dollars per hour) for 1 or 3 years. Compute Savings Plans are the most flexible: the discount automatically applies across instance families, sizes, Regions, and even AWS Fargate and AWS Lambda usage. EC2 Instance Savings Plans commit to a specific instance family in a Region for a larger discount. When an exam question emphasizes commitment plus flexibility across services or families, Savings Plans beat RIs.

Spot Instances. You bid on spare EC2 capacity at discounts that can reach roughly 90 percent off On-Demand. The catch: AWS can reclaim the capacity with a two-minute interruption notice whenever it needs it back. Spot is only appropriate for fault-tolerant, stateless, or flexible-timing workloads: batch jobs, big data analytics, CI builds, rendering. Never the answer for a production database or anything that cannot be interrupted.

Dedicated Hosts and Dedicated Instances. Both give you single-tenant hardware, meaning no other AWS customer shares the physical server. A Dedicated Host gives you visibility into and control over the physical server itself (sockets, cores), which supports Bring Your Own License (BYOL) scenarios for software licensed per socket or per core, such as some Windows Server and SQL Server licenses. Dedicated Instances run on hardware dedicated to your account but without that host-level visibility or placement control. Compliance or licensing language in a question points here.

Option	Commitment	Discount depth	Best for
On-Demand	None	None (baseline)	Unpredictable, short-term, first-time workloads
Reserved Instances	1 or 3 years, instance family	High	Steady-state workloads with known instance needs
Savings Plans	1 or 3 years, spend per hour	High	Steady spend needing flexibility across families or services
Spot	None	Deepest	Interruption-tolerant batch and stateless work
Dedicated Hosts	On-Demand or reservation	Varies	BYOL licensing, strict compliance, hardware visibility

Tenancy versus purchasing

Students often mix these up. Tenancy (shared, dedicated instance, dedicated host) describes whose workloads share the physical hardware. Purchasing option (On-Demand, Reserved, Spot) describes the billing model. A Dedicated Host can itself be paid for On-Demand or through a reservation. The exam occasionally tests that you know these are separate axes.

Cost strategy in practice

Real environments blend options: a baseline of Reserved Instances or a Savings Plan covers the always-on floor of usage, On-Demand absorbs unpredictable spikes, and Spot handles opportunistic batch work. Tools covered in AWS Cost Explorer and AWS Budgets generate RI and Savings Plan purchase recommendations from your historical usage. For the broader pricing philosophy behind all of this, see AWS pricing models explained.

How the CLF-C02 exam tests this

• The steady-state discount pattern. A workload runs 24/7 for the next 1 to 3 years and the company wants to reduce cost. The answer is Reserved Instances or a Savings Plan, with All Upfront payment when the question asks for the maximum discount.
• The interruption-tolerance pattern. A scenario describes batch processing, data analysis, or rendering that can stop and resume without harm, and asks for the lowest-cost compute. The answer is Spot Instances. If the same question said the workload cannot be interrupted, Spot becomes the trap answer.
• The licensing pattern. A company must bring existing per-socket or per-core server licenses to AWS, or a regulator requires that no other customer share the hardware. The answer is Dedicated Hosts (host-level visibility for BYOL) rather than Dedicated Instances.
• The flexibility discriminator. Two answer choices offer commitment discounts, but the scenario mentions that the team may change instance families or also runs Fargate or Lambda. Compute Savings Plans win over Standard Reserved Instances because the discount follows the spend, not a specific instance configuration.

Quick reference

• Instance families: general purpose (balanced), compute optimized (CPU-heavy), memory optimized (RAM-heavy), storage optimized (disk-heavy), accelerated (GPU/ML).
• On-Demand: no commitment, highest per-hour price, best for unpredictable or short-term work.
• Reserved Instances: 1 or 3 year commitment to an instance family; Standard is cheapest, Convertible is exchangeable.
• Savings Plans: commit to dollars per hour; Compute Savings Plans also cover Fargate and Lambda.
• Spot: up to about 90 percent off, reclaimable with a two-minute notice, only for interruption-tolerant work.
• Dedicated Hosts: physical server visibility, supports BYOL and strict compliance; Dedicated Instances lack host-level control.
• Tenancy and purchasing option are independent choices.
• Blend options in practice: reservations for the baseline, On-Demand for spikes, Spot for opportunistic batch.