Compute Cost Savings (Tips 1–8)

Compute consistently represents 50% to 65% of total cloud spend across organizations. These eight tips target the most common sources of compute waste and overspend across AWS, Azure, and GCP.

Tip 1: Rightsize instances using percentile-based analysis, not averages

Average CPU utilization is a misleading metric for rightsizing decisions. An instance averaging 15% CPU but spiking to 85% during batch processing windows cannot safely be downsized based on the average alone. Use 95th percentile metrics over a 30-day window to capture peak workload behavior. Only rightsize when the 95th percentile falls below 40% of the current instance capacity.

# AWS: Get 95th percentile CPU over 30 days
aws cloudwatch get-metric-statistics   --namespace AWS/EC2   --metric-name CPUUtilization   --dimensions Name=InstanceId,Value=i-0abc123def456   --start-time 2026-07-01T00:00:00Z   --end-time 2026-07-31T23:59:59Z   --period 86400   --statistics p95   --output table

Estimated savings: 20%–40% of compute spend for the rightsized instances. See the full Cloud Rightsizing guide for a complete methodology.

Tip 2: Implement non-production environment scheduling

Development, staging, QA, and sandbox environments running 24/7 waste 65% of their compute cost when engineers only use them during business hours. Deploy AWS Instance Scheduler or equivalent automation to stop these environments at 8 PM and start them at 8 AM on weekdays.

Estimated savings: 65% of non-production compute cost. For an organization spending $30K/month on dev/staging environments, that is $19,500/month in immediate savings.

Tip 3: Switch to Graviton (ARM) instances on AWS

AWS Graviton3 instances deliver 25% better price-performance than equivalent x86 (Intel/AMD) instances for most Linux workloads. A workload running on m5.xlarge at $0.192/hour can run on m6g.xlarge at $0.154/hour with equivalent or better performance. If your application stack runs on Linux and does not depend on x86-specific binary dependencies, Graviton migration is one of the highest-ROI changes available.

Estimated savings: 20%–25% per instance migrated to ARM architecture.

Tip 4: Use Savings Plans for stable compute baselines

Compute Savings Plans on AWS offer 30%–60% discounts on compute spend in exchange for a 1-year or 3-year hourly spend commitment. Unlike Reserved Instances, Savings Plans apply automatically across instance families, sizes, and even services (EC2, Fargate, Lambda). Cover 60%–70% of your stable baseline compute with Savings Plans and leave the variable portion on-demand or spot.

The Reserved Instances and Savings Plans guide covers the specific purchasing strategy in detail.

Estimated savings: 30%–60% on covered compute depending on term and payment option.

Tip 5: Deploy spot instances for fault-tolerant workloads

AWS Spot Instances, Azure Spot VMs, and GCP Preemptible/Spot VMs offer 60%–90% discounts compared to on-demand pricing. The trade-off is that the provider can reclaim the instance with minimal notice (2 minutes on AWS, 30 seconds on GCP). Workloads that tolerate interruption — batch processing, CI/CD builds, data pipelines, distributed training, containerized microservices with graceful failover — are ideal candidates.

Run spot instances within Auto Scaling Groups or managed instance groups configured with diversified instance type selection across multiple Availability Zones to minimize interruption impact. The Spot Instance Optimization guide provides architecture patterns that maintain availability above 99.9% while capturing spot discounts.

Estimated savings: 60%–90% on eligible workloads.

Tip 6: Delete zombie Lambda functions and unused API Gateway stages

Serverless architectures accumulate unused functions and API stages over time. Lambda functions that have not been invoked in 90 days, API Gateway stages pointing at decommissioned backends, and Step Functions state machines that have not executed in months all contribute to clutter that, while individually inexpensive, adds up across large serverless deployments. CloudWatch Logs ingestion from verbose Lambda functions can cost more than the compute itself.

Estimated savings: Varies — typically $500–$5,000/month in larger serverless environments.

Tip 7: Use auto-scaling with target tracking instead of fixed capacity

Fixed-capacity deployments provision for peak load and waste money during off-peak hours. Target tracking auto-scaling adjusts capacity dynamically based on metrics like average CPU utilization or request count per target. Set a target CPU of 60%–70% and let the scaling policy add or remove instances as demand fluctuates.

# AWS: Create target tracking scaling policy
aws autoscaling put-scaling-policy   --auto-scaling-group-name my-app-asg   --policy-name target-tracking-cpu   --policy-type TargetTrackingScaling   --target-tracking-configuration '{
    "PredefinedMetricSpecification": {
      "PredefinedMetricType": "ASGAverageCPUUtilization"
    },
    "TargetValue": 65.0
  }'

Estimated savings: 30%–50% compared to fixed-capacity provisioning for workloads with variable traffic patterns.

Tip 8: Consolidate underutilized accounts and subscriptions

Organizations with dozens of AWS accounts or Azure subscriptions often run duplicate baseline services (NAT Gateways, VPN connections, monitoring agents) in each account. Consolidating low-utilization accounts reduces these fixed per-account costs. A NAT Gateway costs $32/month minimum per Availability Zone regardless of traffic volume — multiplied across 50 accounts and 3 AZs, that is $4,800/month in NAT Gateway fixed costs alone.

Estimated savings: $2,000–$15,000/month depending on account sprawl.

Storage Cost Savings (Tips 9–14)

Tip 9: Implement S3 Intelligent-Tiering for unpredictable access patterns

S3 Intelligent-Tiering automatically moves objects between frequent and infrequent access tiers based on actual access patterns, with no retrieval fees for automated tier transitions. For data with unpredictable access patterns, it eliminates the guesswork of manually choosing between S3 Standard ($0.023/GB) and S3 Infrequent Access ($0.0125/GB).

Estimated savings: 30%–50% on storage costs for data with variable access patterns.

Tip 10: Enable S3 Lifecycle Policies for aging data

Data accessed frequently during its first 30 days but rarely afterward should transition automatically: Standard → Infrequent Access after 30 days → Glacier Flexible Retrieval after 90 days → Glacier Deep Archive after 180 days. The cost difference between S3 Standard ($0.023/GB/month) and Glacier Deep Archive ($0.00099/GB/month) is a 95% reduction.

{
  "Rules": [{
    "ID": "archive-aging-data",
    "Status": "Enabled",
    "Transitions": [
      {"Days": 30, "StorageClass": "STANDARD_IA"},
      {"Days": 90, "StorageClass": "GLACIER"},
      {"Days": 180, "StorageClass": "DEEP_ARCHIVE"}
    ]
  }]
}

Estimated savings: 50%–95% on long-term storage costs. The Cloud Storage Cost Optimization guide covers tiering strategies across all providers.

Tip 11: Delete old EBS snapshots

EBS snapshots accumulate silently. An automated daily snapshot retention policy that keeps snapshots for 30 days is reasonable. Organizations without cleanup policies often have snapshots stretching back years, each billing at $0.05/GB/month. An audit of EBS snapshots older than 90 days typically reveals thousands of dollars in monthly waste.

Estimated savings: $1,000–$10,000/month depending on snapshot volume.

Tip 12: Compress data before storing

Compressing log files, backups, and archival data before uploading to cloud storage reduces both storage costs and data transfer costs. Gzip compression typically achieves 5:1 to 10:1 ratios on text-based data like logs, reducing a 100 GB daily log volume to 10–20 GB.

Estimated savings: 60%–90% on storage costs for compressible data types.

Tip 13: Right-size EBS volumes and switch to gp3

Many EBS volumes are provisioned at sizes far exceeding actual data stored. A 500 GB gp2 volume storing 50 GB of data wastes $40/month on unused capacity. Additionally, AWS gp3 volumes provide the same baseline performance as gp2 at 20% lower cost ($0.08/GB vs $0.10/GB), with independently configurable IOPS and throughput.

Estimated savings: 20%–60% on EBS costs through right-sizing and gp2 → gp3 migration.

Tip 14: Use Azure Blob Archive tier and GCS Archive storage

Azure Blob Archive tier costs $0.00099/GB/month — 90% cheaper than Hot tier. GCS Archive storage costs $0.0012/GB/month. Data that must be retained for compliance but is accessed less than once per year should move to these archive tiers automatically via lifecycle policies.

Estimated savings: 85%–95% on cold data storage costs across Azure and GCP.

Network Cost Savings (Tips 15–19)

Tip 15: Replace NAT Gateways with VPC Endpoints for AWS service traffic

NAT Gateway data processing charges ($0.045/GB) apply to all traffic flowing through the gateway, including traffic destined for AWS services like S3, DynamoDB, and SQS. Gateway VPC Endpoints for S3 and DynamoDB are free. Interface VPC Endpoints for other services cost $0.01/hour plus $0.01/GB — still significantly cheaper than NAT Gateway processing for high-volume service traffic.

Estimated savings: $3,000–$20,000/month for architectures with heavy AWS service-to-service traffic through NAT Gateways.

Tip 16: Use CloudFront or CDN for frequently accessed S3 content

S3 data transfer out costs $0.09/GB. CloudFront data transfer for the same content starts at $0.085/GB and decreases with volume, plus CloudFront caching reduces origin requests (and S3 request charges). For content served repeatedly to end users, a CDN reduces both transfer costs and latency.

Estimated savings: 20%–50% on data transfer costs for cacheable content.

Tip 17: Avoid unnecessary cross-region data transfer

Cross-region data transfer costs $0.01–$0.02/GB on AWS, Azure, and GCP. Architectures that replicate data across regions for redundancy should evaluate whether all replicas are necessary. A database replicated to three regions for disaster recovery when the business only requires two regions of coverage wastes one-third of the replication transfer cost.

Review networking cost reduction strategies for architecture patterns that minimize cross-region transfer while maintaining required redundancy.

Estimated savings: $1,000–$15,000/month depending on cross-region traffic volume.

Tip 18: Consolidate load balancers

Each Application Load Balancer on AWS costs $16.20/month minimum (fixed hourly charge) plus LCU-based usage charges. Organizations running separate ALBs for services that could share a single ALB with path-based or host-based routing waste the fixed per-ALB cost. Consolidate where security and routing requirements permit.

Estimated savings: $16–$100/month per consolidated ALB — significant at scale with dozens of underutilized load balancers.

Tip 19: Use AWS PrivateLink instead of NAT for third-party service access

If your instances access third-party SaaS services through a NAT Gateway, the data processing charges apply to all that traffic. AWS PrivateLink provides private connectivity to supported services without traversing the NAT Gateway, eliminating the $0.045/GB processing charge.

Estimated savings: $0.035/GB on third-party service traffic volume.

Database Cost Savings (Tips 20–23)

Tip 20: Use Aurora Serverless v2 for variable database workloads

Aurora Serverless v2 scales database capacity up and down based on actual query load, billing in 0.5 ACU increments. For databases with highly variable load — busy during business hours, nearly idle at night — Serverless v2 can cost 40%–60% less than a fixed-size Aurora instance provisioned for peak capacity.

Estimated savings: 40%–60% compared to fixed-provisioned instances for variable workloads.

Tip 21: Right-size RDS instances using Performance Insights

Enable RDS Performance Insights (free tier available) to analyze database CPU, memory, and I/O utilization. Many RDS instances are provisioned at db.r5.2xlarge or larger when actual workload would run comfortably on db.r5.large — a 75% cost reduction per instance. The Database Cost Optimization guide provides detailed rightsizing methodology for RDS, Cloud SQL, and Azure SQL Database.

Estimated savings: 30%–75% per rightsized database instance.

Tip 22: Apply Reserved Instances to production databases

Production databases run 24/7 by definition — they are the ideal candidates for Reserved Instance pricing. A 1-year All Upfront RI for RDS provides approximately 40% savings over on-demand. A 3-year All Upfront RI saves approximately 60%. Since production databases are rarely terminated, the commitment risk is minimal.

Estimated savings: 40%–60% on RDS, ElastiCache, Redshift, and other RI-eligible database services.

Tip 23: Optimize BigQuery costs with slot reservations and partitioned tables

BigQuery on-demand pricing charges $6.25 per TB of data scanned. Poorly structured queries scanning entire tables when only recent data is needed multiply costs unnecessarily. Partition tables by date and use WHERE _PARTITIONTIME filters to restrict scans. For organizations consistently scanning more than 500 TB/month, BigQuery flat-rate slot reservations provide cost predictability and often reduce spend by 40%+ compared to on-demand.

Estimated savings: 30%–70% through partitioning, clustering, and appropriate pricing model selection.

Organizational Savings (Tips 24–27)

Tip 24: Enforce mandatory tagging to enable cost accountability

Without tags, you cannot attribute costs to teams, projects, or environments. Without attribution, nobody owns the cost. Without ownership, nobody optimizes. Implement mandatory tagging via policy-as-code using AWS SCPs, Azure Policy, or GCP Organization Policies. Block resource creation when required tags (team, environment, cost-center) are missing.

Estimated savings: Indirect but foundational — enables all other team-level optimization.

Tip 25: Set up automated cost anomaly alerts

A cost anomaly caught within hours costs tens of dollars to resolve. The same anomaly discovered during monthly billing review costs thousands or tens of thousands. Enable AWS Cost Anomaly Detection (free), configure Azure Cost Management alerts, and set custom alerting on GCP using Cloud Monitoring budget alerts. Route all alerts to a dedicated Slack channel monitored by the FinOps team.

Estimated savings: $5,000–$50,000+ per prevented anomaly — the ROI on a single caught runaway spend event often exceeds the annual cost of the alerting infrastructure.

Tip 26: Conduct monthly commitment coverage reviews

Commitment coverage — the percentage of eligible compute spend covered by RIs, Savings Plans, or CUDs — should be reviewed monthly. Industry benchmarks target 70%–85% coverage. Each percentage point of coverage gap represents compute running at full on-demand rates when commitment pricing would save 30%–60%. A commitment intelligence platform automates coverage analysis and purchase recommendations.

Estimated savings: 5%–15% of total compute spend from closing commitment coverage gaps.

Tip 27: Publish team-level cost dashboards and celebrate optimization wins

Transparency drives behavior. When engineering teams can see their infrastructure costs — and how they compare to similar teams — cost optimization becomes a natural part of engineering culture rather than an external mandate. Publish weekly team-level cost reports. Celebrate engineers who identify and implement significant savings. Include cost efficiency as a supporting metric in engineering team OKRs as recommended in the FinOps best practices guide.

Estimated savings: 10%–20% reduction in cloud waste from improved cost awareness and cultural accountability.

Prioritizing Your Savings Roadmap

Not all 27 tips apply equally to every organization, and attempting to implement all of them simultaneously is a recipe for incomplete execution across the board. Prioritize based on impact-to-effort ratio:

The organizations that achieve 40%+ total cost reduction apply these tips systematically over 3 to 6 months rather than cherry-picking individual techniques. Each layer of optimization compounds on the previous — rightsizing reduces the base cost that commitment purchases then discount further, which auto-scaling then adjusts dynamically. The result is an infrastructure cost profile that continuously tightens as each optimization layer matures.

For a structured implementation framework that wraps these tactical tips into an organizational practice, the How to Start with FinOps guide provides a 90-day implementation playbook. For the ongoing operational cadence that sustains these savings long-term, the FinOps Best Practices guide covers the habits and processes that prevent optimized spend from reverting.

Key Takeaway

Cloud cost savings of 40% or more are achievable through systematic application of these 27 techniques across compute, storage, networking, databases, and organizational practices. Start with quick wins — idle resource cleanup, environment scheduling, and tagging enforcement — to build momentum and organizational credibility. Progress to commitment purchases and rightsizing for the largest dollar impact. Complete the stack with architecture-level optimizations and cultural practices that prevent waste from reaccumulating between review cycles.