The Complete Guide to Cloud Cost Optimization: Optimize AWS, Azure, and GCP Infrastructure for Maximum Efficiency in 90 Days

Why Cloud Costs Spiral Out of Control

Cloud computing promised financial efficiency through pay-as-you-go pricing and elastic scaling. Yet most organizations discover their cloud bills increasing 20-50% year-over-year, often outpacing business growth. After analyzing cloud infrastructure patterns across 150+ companies, Sahi Technologies has identified the root causes of resource inefficiency—and more importantly, proven strategies to optimize infrastructure investment and delivery.

The reality: most companies operate inefficiently on cloud infrastructure, wasting 30-50% of their infrastructure investment. This isn't due to cloud provider issues. It's the result of five systemic issues that plague engineering organizations:

The good news? These problems are solvable. Sahi Technologies has helped organizations optimize cloud infrastructure by 40% on average within 90 days using the strategies outlined in this guide. Unlike one-time audits that deliver temporary improvements, these are sustainable practices that compound over time.

5 Quick Wins for Immediate Value

Before diving into comprehensive optimization strategies, start with these five quick wins that deliver immediate results with minimal effort. Sahi Technologies implements these first in every cost optimization engagement because they typically save 15-25% within the first week.

1. Delete Unattached EBS Volumes and Elastic IPs

When EC2 instances are terminated, their attached EBS volumes often remain—continuing to accumulate charges. Similarly, Elastic IPs cost money when not associated with running instances. Sahi Technologies typically finds $500-$2,000 monthly waste from orphaned resources.

# AWS CLI: Find unattached volumes
aws ec2 describe-volumes --filters Name=status,Values=available \
  --query 'Volumes[*].[VolumeId,Size,CreateTime]' --output table

# Find unattached Elastic IPs
aws ec2 describe-addresses --query 'Addresses[?InstanceId==null]'

Action: Audit your AWS, Azure, or GCP account for orphaned resources weekly. Implement automated cleanup scripts that delete resources unattached for 7+ days.

2. Stop Non-Production Environments During Off-Hours

Development, staging, and QA environments rarely need 24/7 availability. Running them only during business hours (40 hours per week) instead of continuously (168 hours) reduces costs by 76%.

Example value calculation:

• Dev environment: 5x t3.large instances ($0.08/hour) = $300/month if running 24/7
• Same environment running 9am-6pm weekdays: $75/month
• Monthly value: $225 per environment
• 10 non-prod environments: $2,250/month = $27,000/year optimization

3. Enable S3 Intelligent-Tiering

AWS S3 Intelligent-Tiering automatically moves objects between storage tiers based on access patterns. Objects not accessed for 30 days move to Infrequent Access (50% cheaper), and after 90 days to Archive tier (80% cheaper).

Real-world impact: A Sahi Technologies client storing 50TB of application logs reduced S3 costs from $1,150/month to $280/month—a 76% reduction—simply by enabling Intelligent-Tiering with zero application changes.

4. Delete Old Snapshots and AMIs

Snapshots accumulate over time as teams create AMIs for deployments, backups, and disaster recovery. Each snapshot stores only incremental changes, but when the source volume is deleted, snapshots retain full data—and full costs.

Snapshot retention policy recommendation:

• Daily snapshots: Retain 7 days
• Weekly snapshots: Retain 4 weeks
• Monthly snapshots: Retain 12 months
• Delete everything older automatically

5. Right-Size Obvious Offenders

Before comprehensive rightsizing analysis (covered later), tackle the obvious waste. Look for instances with consistent CPU utilization under 10% or massive memory overhead. These are safe, no-regret downsizes.

Combined impact of quick wins: These five actions typically optimize cloud infrastructure by 15-25% within one week, with Sahi Technologies clients achieving $5,000-$15,000 monthly value depending on infrastructure size.

Resource Rightsizing Strategy

Resource rightsizing—matching instance sizes to actual utilization—is the single highest-impact optimization lever. Most organizations overprovision by 2-4x "just in case," wasting 50-75% of compute spend. Sahi Technologies' systematic rightsizing approach delivers 30-40% compute cost reductions while improving performance through better resource matching.

The Rightsizing Process

Step 1: Collect 30 Days of Metrics

Never rightsize based on gut feeling or point-in-time observations. Collect comprehensive metrics over 30 days minimum to capture weekly cycles, monthly patterns, and anomalies:

• CPU utilization: Average, p95, p99, and maximum
• Memory utilization: Average and peak committed memory
• Network throughput: Average and peak bandwidth usage
• Disk IOPS: Read/write operations per second
• Application performance: Response times and error rates

Step 2: Identify Rightsizing Candidates

After collecting 30 days of data, categorize workloads into three buckets:

Safe to downsize (high confidence):

• Avg CPU < 20% AND p95 CPU < 40%
• Avg memory < 30% AND p95 memory < 60%
• Steady utilization with no spikes
• Action: Downsize by 1-2 instance sizes immediately

Potential downsize (requires testing):

• Avg CPU 20-40% with occasional spikes
• Memory utilization inconsistent
• Action: Test smaller size in staging first

Correctly sized or needs upsize:

• Avg CPU > 60% OR p99 CPU > 85%
• Memory pressure indicators present
• Action: Leave as-is or upsize if performance issues exist

Instance Family Optimization

Beyond sizing, evaluate whether workloads are in the optimal instance family. Sahi Technologies frequently discovers massive savings by switching families:

• Compute-optimized (C-series): Use for CPU-intensive workloads (API servers, batch processing). 30-40% cheaper than general-purpose for same vCPU count
• Memory-optimized (R-series): Use for RAM-intensive workloads (databases, caching, in-memory analytics). 50%+ cheaper per GB RAM than general-purpose
• Burstable (T-series): Use for workloads with low baseline CPU but occasional spikes (dev environments, small databases). 60-70% cheaper than fixed-performance instances
• ARM-based (Graviton): AWS Graviton instances offer 20-40% better price-performance than x86. Compatible with most Linux workloads

Reserved Instances and Savings Plans

Reserved Instances (RIs) and Savings Plans offer 30-70% discounts compared to on-demand pricing in exchange for 1 or 3-year commitments. For predictable workloads, this is free money—yet Sahi Technologies consistently finds organizations underutilizing these mechanisms, leaving 20-30% savings on the table.

Reserved Instances vs. Savings Plans: Which to Choose?

Standard Reserved Instances (Up to 72% discount):

• Best for: Static workloads with consistent instance types
• Flexibility: Locked to instance family, size, region, and OS
• Discount: 60-72% for 3-year, all upfront payment
• When to use: Production databases, always-on application servers

EC2 Instance Savings Plans (Up to 72% discount):

• Best for: Flexible workloads that change instance types
• Flexibility: Apply to any instance family/size within region
• Discount: 65-72% for 3-year commitment
• When to use: Application fleets that evolve over time

Compute Savings Plans (Up to 66% discount):

• Best for: Multi-service compute spend (EC2, Fargate, Lambda)
• Flexibility: Apply across services, regions, and instance families
• Discount: 60-66% for 3-year commitment
• When to use: Modern architectures using containers and serverless

How to Buy Reserved Instances Without Overcommitting

The fear of overcommitting keeps many organizations from purchasing RIs, but the strategy is straightforward:

1. Analyze 6 months of historical usage to identify baseline always-on capacity
2. Start conservative: Purchase RIs to cover only 50-60% of current baseline
3. Monitor utilization for 30 days—target 95%+ RI utilization
4. Increase coverage incrementally every quarter based on growth patterns
5. Use Convertible RIs for production workloads likely to change (allows instance type exchanges)

Common RI purchasing mistakes to avoid:

• ❌ Buying RIs for dev/test environments (use schedulers instead)
• ❌ Committing to 3 years on new workloads (start with 1 year)
• ❌ Ignoring utilization reports (unused RIs waste money)
• ❌ Buying Standard RIs for applications that frequently change instance types

Spot Instances for Non-Critical Workloads

AWS Spot Instances, Azure Spot VMs, and GCP Preemptible VMs offer 60-90% discounts on compute in exchange for accepting interruptions with 2 minutes notice. For fault-tolerant workloads, Spot instances are a game-changer. Sahi Technologies uses them extensively for batch processing, data analytics, CI/CD pipelines, and containerized applications.

Perfect Spot Instance Use Cases

• Batch processing and data pipelines: Checkpointable jobs that can resume after interruption
• CI/CD build agents: Failed builds simply retry on new instances
• Big data and analytics: Spark clusters, EMR jobs, data transformations
• Kubernetes worker nodes: Combined with Cluster Autoscaler for resilience
• Machine learning training: Save checkpoints frequently to resume training
• Rendering and video transcoding: Easily parallelizable workloads

Spot Instance Best Practices

1. Diversify instance types: Request 5-10 different instance types in your Spot fleet to reduce interruption risk
2. Use Spot Placement Score: AWS provides interruption likelihood data—choose instance types with low interruption rates
3. Implement graceful shutdown: Handle SIGTERM signals to checkpoint work and clean up properly
4. Never run stateful workloads on Spot alone: Databases, file servers, and stateful apps need on-demand or reserved capacity
5. Monitor Spot interruptions: Track how often your Spot instances are reclaimed and optimize instance type selection accordingly

Expected savings: Sahi Technologies clients using Spot instances for appropriate workloads see 60-70% cost reductions on those specific workload categories, translating to 15-25% overall compute savings when combined with other optimizations.

Storage Optimization Techniques

Storage costs grow silently over time, accumulating snapshots, old backups, and forgotten volumes. While individual storage costs seem small ($0.10/GB/month for S3 Standard), they compound quickly—especially with replication, snapshots, and multi-region setups. Sahi Technologies typically finds 30-50% storage waste in unoptimized environments.

S3 Optimization Strategies

1. Lifecycle Policies for Automatic Tiering

Configure S3 lifecycle rules to automatically transition objects through storage tiers based on age:

• 0-30 days: S3 Standard ($0.023/GB)
• 30-90 days: S3 Standard-IA ($0.0125/GB, 46% cheaper)
• 90-180 days: S3 Glacier Instant Retrieval ($0.004/GB, 83% cheaper)
• 180+ days: S3 Glacier Deep Archive ($0.00099/GB, 96% cheaper)

2. Delete Incomplete Multipart Uploads

Failed multipart uploads leave fragments in S3 that accumulate charges indefinitely. Set a lifecycle rule to delete incomplete uploads after 7 days.

3. Use S3 Intelligent-Tiering for Unpredictable Access Patterns

When you can't predict access patterns, enable Intelligent-Tiering to automatically optimize storage costs without lifecycle policies.

EBS Volume Optimization

• Delete unattached volumes: Volumes from terminated instances often remain, costing $0.08-$0.10/GB/month indefinitely
• Snapshot management: Implement automated snapshot deletion after retention period expires
• Rightsize volume types: Use gp3 instead of gp2 for 20% cost savings with same performance
• Reduce over-provisioned IOPS: Most workloads don't need provisioned IOPS—use gp3 baseline performance

Network Cost Reduction

Data transfer costs are the silent budget killer in cloud infrastructure. While compute and storage costs are obvious in billing dashboards, network egress fees (data leaving your cloud) accumulate without visibility. Sahi Technologies has seen network costs represent 10-30% of total cloud spending for data-heavy applications.

Understanding Cloud Network Pricing

Network transfer costs vary dramatically by scenario:

Network Cost Optimization Tactics

1. Minimize Cross-AZ Data Transfer

Every request between availability zones costs $0.01/GB each direction. For high-volume applications, this adds up quickly.

• Deploy application and database in same AZ when possible
• Use read replicas in same AZ for read-heavy workloads
• Consider multi-AZ only for production workloads requiring high availability

2. Implement CloudFront or CDN

Serving static assets directly from S3 or EC2 incurs full internet egress charges ($0.09/GB). Using CloudFront reduces costs to $0.085/GB for first 10TB and creates faster user experiences through edge caching.

Additional benefits: CloudFront reduces origin server load, improves global performance, and provides DDoS protection. For media-heavy applications, Sahi Technologies typically sees 40-60% network cost reductions after implementing CDN.

3. Use VPC Endpoints for AWS Services

Accessing S3, DynamoDB, and other AWS services over the internet incurs NAT Gateway costs ($0.045/GB processed). VPC endpoints route traffic internally for free.

4. Compress Data Before Transfer

Enable gzip compression for APIs, implement image optimization, and compress backup data before transfer to S3. Typical compression ratios of 60-80% translate directly to cost savings.

Cost Monitoring and Alerting

You can't optimize what you don't measure. Effective cost monitoring provides real-time visibility into spending patterns, anomaly detection to catch cost spikes before they snowball, and attribution to hold teams accountable. Sahi Technologies implements comprehensive cost monitoring as the foundation of every FinOps program.

Essential Cost Monitoring Components

1. Daily Cost Dashboards

Create dashboards showing:

• Daily spend trend with 7-day and 30-day moving averages
• Spend by service (EC2, RDS, S3, network, etc.)
• Top 10 most expensive resources
• Budget vs. actual spending
• Month-to-date projection

2. Cost Anomaly Detection

Configure AWS Cost Anomaly Detection, Azure Cost Management alerts, or GCP Budget alerts to notify when daily spend exceeds historical patterns by 20%+. This catches:

• Accidental resource creation (someone launched 100 instances instead of 10)
• Unexpected traffic spikes driving network costs
• Misconfigured autoscaling policies
• Failed cleanup scripts leaving resources running

3. Tagging Strategy and Cost Allocation

Implement mandatory tagging for all resources:

• Environment: production, staging, development
• Team: engineering, data, infrastructure
• Project: feature name or product area
• Owner: individual or team email
• Cost-center: for internal chargebacks

Without tagging, cost attribution is impossible. With proper tagging, you can bill teams internally, incentivizing efficient resource usage.

Implementing FinOps Culture

Technology and automation enable cost optimization, but sustainable savings require cultural change. FinOps (Financial Operations) is the practice of bringing financial accountability to cloud spending through collaboration between engineering, finance, and business teams. Sahi Technologies has observed that organizations with strong FinOps culture achieve 2-3x better cost optimization results than those treating it as a one-time project.

Core FinOps Principles

1. Everyone Takes Ownership

Cost optimization isn't the finance team's job or the infrastructure team's job—it's everyone's responsibility. Developers writing code that runs on cloud infrastructure should understand cost implications of their architectural decisions.

2. Centralized Visibility, Distributed Decision-Making

Provide cost dashboards and metrics to all teams, but let engineers make optimization decisions within their domains. Sahi Technologies finds this balance drives better outcomes than centralized mandates.

3. Business Value Drives Spending

Don't optimize for lowest cost—optimize for best cost-to-value ratio. A $5,000/month database supporting $1M in revenue is excellent. The same database supporting an internal tool used by 5 people isn't.

Practical FinOps Implementation Steps

1. Monthly cost review meetings: 30-minute sessions reviewing spending trends, anomalies, and optimization opportunities
2. Cost-aware architecture reviews: Evaluate cost implications before building, not after deployment
3. Team-level cost budgets: Give each team monthly budgets with accountability
4. Celebrate cost wins: Recognize engineers who identify and implement optimizations
5. Continuous optimization, not one-time audits: Build cost review into sprint retrospectives

Essential Cost Optimization Tools

Native Cloud Provider Tools (Free):

• AWS Cost Explorer: Analyze spending patterns, forecast costs, identify rightsizing opportunities
• AWS Compute Optimizer: ML-powered rightsizing recommendations for EC2, Lambda, EBS
• Azure Cost Management: Budget tracking, cost analysis, optimization recommendations
• GCP Recommender: Identifies idle resources, rightsizing opportunities, and commitment recommendations

Third-Party Platforms (Paid but Powerful):

• CloudHealth by VMware: Multi-cloud cost visibility and governance
• Cloudability: Advanced cost allocation and FinOps workflows
• Spot.io: Automated Spot instance management with zero manual intervention
• Datadog Cloud Cost Management: Correlate infrastructure costs with application metrics

Sahi Technologies Recommendation: Start with native cloud tools—they're free and surprisingly capable. Invest in third-party platforms only when managing multi-cloud environments or requiring advanced governance features.

Action Plan: Your 90-Day Roadmap

Implementing these strategies systematically over 90 days delivers compounding savings. Here's the Sahi Technologies proven roadmap:

Week 1-2: Quick Wins (Target: 15-20% Savings)

• ✅ Delete unattached EBS volumes and Elastic IPs
• ✅ Stop non-production environments during off-hours
• ✅ Enable S3 Intelligent-Tiering
• ✅ Delete old snapshots and AMIs
• ✅ Rightsize obvious offenders (CPU < 10% utilization)
• ✅ Implement basic cost monitoring and alerting

Week 3-6: Systematic Optimization (Additional 15-20% Savings)

• ✅ Comprehensive rightsizing analysis across all workloads
• ✅ Purchase Reserved Instances or Savings Plans for baseline capacity
• ✅ Implement Spot instances for batch workloads
• ✅ Configure S3 lifecycle policies
• ✅ Optimize network architecture (VPC endpoints, CloudFront)
• ✅ Implement comprehensive tagging strategy

Week 7-12: Culture and Automation (Sustain Savings Long-Term)

• ✅ Launch monthly FinOps meetings with stakeholders
• ✅ Create team-level cost dashboards
• ✅ Implement automated cost optimization workflows
• ✅ Build cost awareness into architecture review process
• ✅ Establish cost budgets and accountability per team
• ✅ Document cost optimization playbook

Need Help Optimizing Your Cloud Costs?

Cloud cost optimization isn't a one-person job, and it's not a one-time project. It requires expertise across multiple cloud providers, deep understanding of pricing models, and the experience to know which optimizations deliver maximum impact with minimum risk.

Sahi Technologies offers comprehensive cloud cost optimization services:

• Free cloud cost audit: We analyze your infrastructure and identify top 10 opportunities
• 90-day optimization program: Hands-on implementation of all strategies in this guide
• Ongoing FinOps management: Monthly optimization reviews and continuous improvement
• ROI guarantee: Our fees are less than the savings we deliver—guaranteed