EC2 Pricing Comparison: On-Demand, Reserved, Spot, or Savings Plans?

Key Takeaway

Choose Reserved Instance/Savings Plans for predictable long-term workloads (up to 72% discount), On-Demand for flexibility, and Spot for interruptible workloads (up to 90% discount). AWS recommends Savings Plans over Reserved Instances, and Compute Savings Plans is the most flexible option for most cases.

Exam Tip

Exam Essential: On-Demand = flexibility, Reserved/Savings Plans = long-term commitment discount, Spot = maximum discount + interruption risk

EC2 Pricing Options at a Glance

Pricing Option	Max Discount	Commitment	Flexibility	Ideal Workload
On-Demand	None	None	Highest	Unpredictable, short-term
Reserved Instance	72%	1-3 years	Low	Predictable, long-term
Savings Plans	66-72%	1-3 years	High	Predictable, long-term
Spot Instance	90%	None	Medium	Interruptible, batch

Pricing Option Decision Tree:

Can your workload tolerate interruptions?
       │
       ├── YES → Spot Instance (up to 90% savings)
       │
       └── NO → Will you use it for 1-3 years?
                    │
                    ├── YES → Savings Plans / Reserved Instance
                    │              │
                    │              ├── May change instance types → Savings Plans
                    │              └── Same instance continuously → Reserved Instance
                    │
                    └── NO → On-Demand

On-Demand Instance

What is On-Demand?

On-Demand is the default pricing model where you pay only for what you use without long-term commitments. Billed per second (minimum 60 seconds) or per hour.

On-Demand Cost Example (US East Region, m6i.large):

Per Hour: $0.096
Daily (24 hours): $2.30
Monthly (730 hours): $70.08
Annual: $840.96

When On-Demand is Suitable

Use Case	Reason
Dev/Test environments	Unpredictable usage patterns
Short-term projects	Less than 1 year usage
Traffic spike handling	Auto Scaling burst capacity
New workloads	Resource requirements not finalized
Unpredictable patterns	High usage variability

When On-Demand is NOT Suitable

24/7 production servers
Predictable usage workloads
Cost optimization priority

Exam Tip

Exam Point: When you see "flexibility needed", "short-term usage", "unpredictable" keywords → On-Demand

Reserved Instance (RI)

What is Reserved Instance?

Reserved Instance offers up to 72% discount with a 1 or 3-year commitment. You commit to a specific instance type, region, and tenancy.

Reserved Instance Discount Rates (3-year term):

┌─────────────────────────────────────────────────┐
│  Payment Option        │ Discount   │ Upfront   │
├────────────────────────┼────────────┼───────────┤
│  All Upfront           │ Up to 72%  │ Full      │
│  Partial Upfront       │ Up to 66%  │ Some      │
│  No Upfront            │ Up to 60%  │ None      │
└─────────────────────────────────────────────────┘

Reserved Instance Types

1. Standard Reserved Instance

Features:
├── Up to 72% discount
├── Cannot change instance type
├── Cannot change region
└── Can sell on Marketplace

2. Convertible Reserved Instance

Features:
├── Up to 66% discount (lower than Standard)
├── Can change instance family, OS, tenancy
├── Cannot change region
└── Cannot sell on Marketplace

Reserved Instance Scope

Scope	Description	AZ Capacity Reservation
Regional	Applies to all AZs in region	❌ None
Zonal	Applies to specific AZ only	✅ Yes

When Reserved Instance is Suitable

24/7 production servers
Database servers (RDS also supports RI)
Baseline capacity
Instance type is finalized

Exam Tip

Exam Point: Convertible RI can change instance family but has lower discount (66% vs 72%)

Savings Plans

What are Savings Plans?

Savings Plans offer discounts by committing to a certain hourly compute spend for 1-3 years. More flexible than Reserved Instances, and AWS recommends them.

Savings Plans vs Reserved Instance:

Reserved Instance: "I will use m6i.large for 1 year"
Savings Plans: "I will spend $10/hour on compute for 1 year"

→ Savings Plans applies flexibly to instance types, regions, services

Savings Plans Types

1. Compute Savings Plans

Features:
├── Up to 66% discount
├── Applies to EC2, Fargate, Lambda
├── All instance families, sizes, regions
├── All OS, tenancy
└── Most flexible option

Suitable for:
├── Multi-region operations
├── May change instance types
└── Serverless (Lambda, Fargate) usage

2. EC2 Instance Savings Plans

Features:
├── Up to 72% discount (higher than Compute)
├── Commit to specific instance family + region
├── Flexible on instance size, OS, tenancy
└── Applies to EC2 only

Suitable for:
├── Single region operations
├── Instance family fixed (e.g., M6i)
└── May only change size (large → xlarge)

Savings Plans vs Reserved Instance

Comparison	Savings Plans	Reserved Instance
Commitment Unit	Hourly spend	Instance count
Flexibility	High	Low
Max Discount	66-72%	72%
Services	EC2, Fargate, Lambda	EC2, RDS
Management Complexity	Low	High
AWS Recommended	✅	-

When Savings Plans is Suitable

Multi-region workloads
May change instance types
Container/Serverless usage (Fargate, Lambda)
Prefer simple cost management

Exam Tip

Exam Point: AWS recommends Savings Plans. Compute Savings Plans applies to EC2 + Fargate + Lambda, making it most flexible

Spot Instance

What is Spot Instance?

Spot Instance lets you use AWS spare EC2 capacity at up to 90% discount. AWS can reclaim instances with 2-minute warning when capacity is needed.

Spot Instance Operation:

1. AWS checks spare capacity
   ↓
2. Spot price determined (supply/demand based)
   ↓
3. Instance starts if below your max price
   ↓
4. When AWS needs capacity or price rises:
   - 2-minute interruption notice
   - Instance terminate/stop/hibernate

Spot Instance Interruption Handling

Option	Description	Data Retention
Terminate	Delete instance	EBS can be preserved
Stop	Stop instance	EBS preserved
Hibernate	Hibernate mode	Memory + EBS preserved

Spot Fleet

Spot Fleet automatically manages Spot Instances across multiple instance types and AZs.

Spot Fleet Strategies:

1. lowestPrice: Start from cheapest pool (cost optimization)
2. diversified: Distribute across pools (availability)
3. capacityOptimized: Choose pool with most capacity (minimize interruptions)
4. priceCapacityOptimized: Price + capacity optimization (recommended)

Suitable Workloads for Spot

Workload	Reason
Batch processing	Can restart after interruption
Big data analytics	Fault tolerance via distribution
CI/CD pipelines	Builds/tests can be re-run
ML training	Can resume from checkpoints
Rendering	Frame-by-frame reprocessing
Web crawling	Stateless

Unsuitable Workloads for Spot

24/7 production servers
Databases
Stateful applications
Risk of data loss on interruption

Exam Tip

Exam Point: Spot appears with "interruptible", "fault-tolerant", "batch", "stateless" keywords

Pricing Option Combination Strategy

Baseline + Burst Pattern

Most production environments combine multiple pricing options:

Traffic Pattern with Pricing Mix:

Traffic
  ↑
  │    ╭───────╮
  │   ╱         ╲      ← On-Demand (peak)
  │  ╱           ╲
  │ ╱             ╲
  │╱               ╲   ← Spot (burst, interruptible)
  │─────────────────── ← Savings Plans/RI (baseline)
  └─────────────────────→ Time

Recommended Combination

Typical Production Environment:

1. Baseline capacity (24/7 operation)
   → Savings Plans or Reserved Instance (70%)

2. Predictable peaks (regular traffic increases)
   → On-Demand + Auto Scaling (20%)

3. Fault-tolerant workloads (batch, ML)
   → Spot Instance (10%)

Scenario-Based Solutions

Scenario 1: 24/7 Production Web Server

Requirement: Stability, predictable cost
Choice: Savings Plans (Compute)
Reason: Flexibility + up to 66% discount, instance change possible

Scenario 2: Dev/Test Environment

Requirement: Minimize cost, occasional use
Choice: On-Demand
Reason: Pay only for usage, no commitment needed

Scenario 3: Nightly Batch Processing

Requirement: Minimize cost, can be interrupted
Choice: Spot Instance
Reason: Up to 90% discount, batch jobs can be rerun

Scenario 4: Global Multi-Region Operations

Requirement: Multiple regions, various instance types
Choice: Compute Savings Plans
Reason: Applies to all regions and instance types

Scenario 5: Single Region Fixed Workload

Requirement: US East only, continue using M6i instances
Choice: EC2 Instance Savings Plans or Reserved Instance
Reason: Up to 72% discount, only instance size changes

Scenario 6: ML Training Cluster

Requirement: Large-scale GPU, minimize cost
Choice: Spot Instance + checkpointing
Reason: Up to 90% discount, training can resume

SAA-C03 Exam Focus Points

✅ On-Demand: Maximum flexibility, no discount, short-term/unpredictable workloads
✅ Reserved Instance: Up to 72% discount, 1-3 year commitment, Standard vs Convertible
✅ Savings Plans: AWS recommended, Compute (flexible) vs EC2 Instance (higher discount)
✅ Spot Instance: Up to 90% discount, 2-minute interruption notice, fault-tolerant workloads
✅ Spot Fleet: Auto-manage across pools, priceCapacityOptimized recommended
✅ Combination Strategy: Baseline (RI/SP) + Peak (OD) + Batch (Spot)

Exam Tip

Sample Exam Question: "A company runs large-scale data analysis at night. Jobs can be interrupted and restarted. What EC2 pricing option minimizes cost?" → Answer: Spot Instance (up to 90% discount + interruptible batch workload)

Frequently Asked Questions

Q: Should I choose Savings Plans or Reserved Instance?

Savings Plans is recommended for most cases. It's more flexible and easier to manage. However, if you're certain the instance will never change and need maximum discount, Standard Reserved Instance is good. AWS console also suggests Savings Plans first.