AWS Cost Optimization: Reserved Instances vs Savings Plans

Introduction

AWS cost optimization is one of the most underutilized ways to reduce infrastructure expenses. For companies running on AWS, computing costs typically represent 40-60% of the total cloud bill. By understanding the differences between Reserved Instances (RIs), Savings Plans, and On-Demand pricing, you can reduce costs by 30-70% without changing your architecture.

This guide provides practical strategies for optimizing AWS costs with real calculations and implementation examples.

Why AWS Cost Optimization Matters

The Cost Problem

Most organizations start with On-Demand pricing:

On-Demand EC2: $0.0965/hour for t3.medium in us-east-1
Annual cost for 1 instance: $0.0965 × 24 × 365 = **$845/year**
For 100 instances: $84,500/year

The Optimization Solution

With Reserved Instances or Savings Plans:

RI price: $0.0483/hour (50% savings)
Annual cost for 100 instances: $42,250/year
Savings: $42,250/year (50% reduction)

This scales rapidly as your infrastructure grows.

On-Demand Pricing: Baseline

On-Demand pricing means you pay for every hour of compute time without any commitment.

Pros

Maximum flexibility
Pay only for what you use
No commitment required
Easy to scale up/down

Cons

Highest per-unit cost
Unpredictable bills
Poor ROI for steady-state workloads

Use Case

On-Demand is appropriate for:

Development/testing environments
Unpredictable workloads
Temporary infrastructure
Short-term testing

Reserved Instances (RIs): Deep Discount Commitment

A Reserved Instance is a commitment to use a specific instance type in a specific region for 1 or 3 years.

How RIs Work

On-Demand Price:     $0.0965/hour
1-Year RI Discount:  -40% = $0.0579/hour
3-Year RI Discount:  -50% = $0.0483/hour

Monthly Savings (100 instances, 1-year RI):
= (0.0965 - 0.0579) × 24 × 30 × 100
= $0.0386 × 720 × 100
= $2,779/month
= $33,350/year

RI Payment Options

All Upfront
- Largest discount (up to 50%)
- Pay entire 1-3 year cost upfront
- Best for predictable, long-term workloads
- Example: $420 upfront for t3.medium (1-year, all upfront)
Partial Upfront
- Medium discount (25-40%)
- Pay ~30% upfront, rest monthly
- Good balance of savings and flexibility
- Example: $130 upfront + $30/month
No Upfront
- Smallest discount (10-20%)
- Monthly payments, no upfront cost
- Better cash flow management
- Example: $35/month with no upfront

RI Flexibility

RIs are locked to:

Instance family (e.g., t3, m5, c5)
Instance size (e.g., medium, large, xlarge)
Availability Zone (optional)
Region (e.g., us-east-1)

Limitations:

Cannot change region
Changing instance size within family has fees
Non-transferable (generally)

Real-World RI Example

Scenario: Running 10 t3.medium instances in us-east-1 for 3 years

On-Demand Cost:

= $0.0965/hour × 24 hours × 365 days × 3 years × 10 instances
= $0.0965 × 262,800 × 10
= $253,452

3-Year All-Upfront RI Cost:

= $1,051 upfront × 10
= $10,510
= Total 3-year cost: $10,510
= Hourly equivalent: $0.0483/hour

Savings = $253,452 - $10,510 = $242,942 (96% savings!)

Savings Plans: Modern Alternative

AWS Savings Plans (introduced 2019) offer more flexibility than RIs while maintaining significant discounts.

How Savings Plans Work

You commit to spending a dollar amount (not instance count) for 1 or 3 years.

On-Demand: $0.0965/hour × 730 hours = $70.45/month
Savings Plan (1-year): $42.10/month (40% discount)
Savings Plan (3-year): $35.14/month (50% discount)

Types of Savings Plans

1. Compute Savings Plans (Most Flexible)

Apply to any instance type, region, or size
Can switch between EC2, Fargate, and Lambda
Best for dynamic workloads
30-40% discount (1-year), 40-50% discount (3-year)

2. EC2 Instance Savings Plans (Medium Flexibility)

Apply to specific instance family (e.g., t3, m5)
Work across sizes and regions within family
35-45% discount (1-year), 45-55% discount (3-year)

3. DynamoDB & SageMaker Savings Plans

Service-specific
Simpler to calculate

Savings Plans vs RIs: Key Differences

Feature	RI	Savings Plan
Flexibility	Low	High
Discount Range	30-50%	30-50%
Can switch instance size	Limited	Yes (compute plans)
Can switch region	No	Yes (compute plans)
Supports different services	No	Compute plans do
Purchase complexity	Moderate	Lower
Pricing	Per instance	Per $ committed

Head-to-Head Comparison: RI vs Savings Plans

Scenario: Mixed workload with changing instance needs

Workload:

Start with 5x t3.medium instances ($352.50/month on-demand)
After 6 months, need to scale to 3x t3.large instances
Need flexibility across regions

With Reserved Instances:

Purchase 5x t3.medium 1-year RIs: $1,740
After 6 months, can’t easily convert to t3.large
If you want to switch, you lose savings or buy new RIs
Total cost: $1,740 + new RI purchases

With Compute Savings Plans:

Commit to $200/month compute spending (1-year plan)
Upfront: $2,000 (less than RIs for same period)
Can switch to 3x t3.large after 6 months without penalty
Savings apply automatically
Total cost: $2,000 (fixed)

Winner: Savings Plans (more flexibility, similar cost)

Spot Instances: Ultra-Low Cost Option

Spot Instances are unused AWS capacity at 70-90% discounts. The tradeoff: AWS can interrupt them with 2-minute notice.

When to Use Spot

Batch jobs (data processing, analytics)
Fault-tolerant workloads
CI/CD pipelines
Development/testing
Time-insensitive tasks

When NOT to Use Spot

Production databases
Long-running services
Real-time applications

Cost Example

t3.medium pricing:
- On-Demand: $0.0965/hour
- 1-Year RI: $0.0579/hour
- Spot: $0.0289/hour (70% discount)

For 100 instances running 24/7:
- On-Demand: $84,500/year
- 1-Year RI: $50,700/year
- Spot: $25,300/year

Implementation Strategy

Step 1: Analyze Current Usage (Week 1)

Use AWS Cost Explorer:

Go to AWS Cost Management → Cost Explorer
Filter by service (EC2)
View last 3 months of compute usage
Identify sustained, predictable workloads

What to look for:

Instances running 24/7
Consistent instance types
Production vs development

Step 2: Calculate Savings Potential (Week 1)

For each instance type:

Monthly On-Demand Cost = $X
Estimated RI Savings = $X × 50%
Annual Savings = (Monthly Savings × 12)
Payback Period = RI Cost / Monthly Savings

Step 3: Purchase RIs or Savings Plans (Week 2)

For predictable, steady-state workloads: Reserve Instances For flexible/changing workloads: Savings Plans

Go to EC2 → Reserved Instances
Click “Purchase Reserved Instances”
Select instance type, term, payment option
Review hourly cost and commit

Step 4: Implement Spot for Dev/Test (Week 2)

Identify non-critical workloads
Launch Spot instances instead of On-Demand
Set max price based on On-Demand price
Use Spot Fleet for multi-instance deployments

Step 5: Monitor and Optimize (Ongoing)

Track RI utilization:

Cost Explorer → Reserved Instance Utilization
Aim for >80% utilization
Adjust future purchases based on patterns

Real-World Case Studies

Case Study 1: SaaS Company (100 EC2 instances)

Before:

100x t3.large instances (on-demand)
Monthly cost: $7,104
Annual cost: $85,248

Optimization:

Purchase 100x t3.large 3-year RIs (all upfront)
Cost: $35,700 upfront
New hourly rate: $0.049 (vs $0.0965)

Results:

Year 1 savings: $49,548
Year 2 savings: $49,548
Year 3 savings: $49,548
3-year total savings: $113,880 (58% reduction)

Case Study 2: Data Pipeline Company (Mixed workload)

Before:

20x m5.xlarge for production (on-demand)
50x t3.medium for development (on-demand)
Monthly cost: $4,500
Annual cost: $54,000

Optimization:

Purchase 20x m5.xlarge 1-year RIs: $7,500
Implement Spot Fleet for dev (20 instances): $200/month
Use Savings Plans for burst capacity: $1,500/year

Results:

RI savings: $2,400/year
Spot savings: $14,400/year
Savings Plan optimization: $1,500/year
Total annual savings: $18,300 (34% reduction)

Common Mistakes to Avoid

Mistake 1: Over-Purchasing RIs

Problem: Buy RIs for capacity you don’t use

Solution:

Start with conservative estimates
Monitor utilization monthly
Only commit to proven, sustained workloads

Mistake 2: Not Using Savings Plans

Problem: Remaining on-demand because you’re uncomfortable with RIs

Solution:

Savings Plans offer 80% of RI savings with more flexibility
No downside to switching from on-demand

Mistake 3: Ignoring Spot Instances

Problem: Running dev/test on on-demand or RIs

Solution:

Dev/test should be 90% Spot
Save RIs/Savings Plans for production

Mistake 4: Not Considering Different Services

Problem: Buying RIs for EC2 but ignoring RDS, ElastiCache

Solution:

Apply savings plans to all applicable services
RDS Reserved Instances: 30-55% savings
ElastiCache: 35-50% savings

Tools and Automation

AWS Compute Optimizer

Provides RI/Savings Plan recommendations:

Analyze current usage patterns
Recommend optimal instance types
Estimate savings with different options

Trusted Advisor

Identifies under-utilized RIs and opportunity purchases

Third-Party Tools

CloudHealth by VMware: Advanced cost analytics
Cloudability: Detailed cost tracking and recommendations
Kubecost: For Kubernetes workloads on AWS

Decision Matrix

Choose Reserved Instances IF:
✓ Workload is predictable
✓ Instance type won't change for months
✓ Can commit to 1-3 year terms
✓ Want maximum discount
✓ Upfront budget available

Choose Savings Plans IF:
✓ Need instance type flexibility
✓ Workload varies by size/type
✓ Want 80% of RI savings with flexibility
✓ May change regions
✓ Prefer predictable monthly billing

Choose Spot IF:
✓ Workload is fault-tolerant
✓ Can handle interruptions
✓ Development/testing workload
✓ Batch processing jobs
✓ Need 70%+ discount

Choose On-Demand IF:
✓ Unpredictable workload
✓ Short-term infrastructure
✓ Don't know usage patterns yet
✓ Prototyping new systems

Glossary

Reserved Instance (RI): Commitment to specific instance for 1-3 years
Savings Plan: Dollar-amount commitment offering flexibility
Spot Instance: AWS excess capacity at steep discount
On-Demand: Pay-per-hour without commitment
Utilization: Percentage of time instance is used
Upfront Payment: Initial cost paid at purchase

Introduction

Why AWS Cost Optimization Matters

The Cost Problem

The Optimization Solution

On-Demand Pricing: Baseline

Pros

Cons

Use Case

Reserved Instances (RIs): Deep Discount Commitment

How RIs Work

RI Payment Options

RI Flexibility

Real-World RI Example

Savings Plans: Modern Alternative

How Savings Plans Work

Types of Savings Plans

1. Compute Savings Plans (Most Flexible)

2. EC2 Instance Savings Plans (Medium Flexibility)

3. DynamoDB & SageMaker Savings Plans

Savings Plans vs RIs: Key Differences

Head-to-Head Comparison: RI vs Savings Plans

Scenario: Mixed workload with changing instance needs

Spot Instances: Ultra-Low Cost Option

When to Use Spot

When NOT to Use Spot

Cost Example

Implementation Strategy

Step 1: Analyze Current Usage (Week 1)

Step 2: Calculate Savings Potential (Week 1)

Step 3: Purchase RIs or Savings Plans (Week 2)

Step 4: Implement Spot for Dev/Test (Week 2)

Step 5: Monitor and Optimize (Ongoing)

Real-World Case Studies

Case Study 1: SaaS Company (100 EC2 instances)

Case Study 2: Data Pipeline Company (Mixed workload)

Common Mistakes to Avoid

Mistake 1: Over-Purchasing RIs

Mistake 2: Not Using Savings Plans

Mistake 3: Ignoring Spot Instances

Mistake 4: Not Considering Different Services

Tools and Automation

AWS Compute Optimizer

Trusted Advisor

Third-Party Tools

Decision Matrix

Glossary

Resources

Comments