Cloud Cost Optimization Checklist: 15 Ways to Cut Your Bill Today

Cloud waste hit $44.5 billion in infrastructure spending for 2025, according to Harness's FinOps in Focus report. A checklist works because cloud waste is repetitive. The same mistakes — idle instances, unattached storage, missing discounts — show up in nearly every account.

One-time cleanups don't stick. We've found that teams who audit their cloud bill once save 15-20% initially, then slowly drift back within three months. A monthly checklist creates a feedback loop. It catches new waste before it compounds and keeps savings locked in. Think of it like brushing your teeth — the habit matters more than any single session.

The 15 items below are ordered by effort. Five-minute quick wins come first because they deliver immediate results. Right-sizing takes 30 minutes per resource but saves more. Commitment discounts require planning but deliver the deepest cuts. Architecture changes take the longest but compound over time.

For a deeper look at where cloud waste comes from and why it keeps growing, see our guide to cloud waste and overprovisioning.

Idle resources account for the bulk of preventable cloud waste. 84% of organizations report struggling to manage cloud spend (Flexera 2025). These five items cost nothing to fix and often save 5-15% of your monthly bill within the first hour.

1. Delete unused resources

Look for load balancers with no targets, Elastic IPs not attached to running instances, and empty resource groups. AWS charges $0.005 per hour for unattached Elastic IPs — roughly $3.60 per month each. Sounds small, but five forgotten IPs across two accounts adds up fast.

How to check: In AWS, go to EC2 > Elastic IPs and look for any without an associated instance. In Azure, check the resource group for items with no activity. In GCP, review "Unattached disks" under Compute Engine.

Expected savings: 1-3% of total bill.

2. Stop idle instances

An idle instance is running but doing no useful work. Dev and staging environments left running overnight and on weekends are the usual suspects. A t3.medium on AWS costs about $30 per month. Four idle instances? That's $120 per month for nothing. Scheduling non-production instances to stop outside business hours can cut their cost by 65%.

How to check: Sort instances by CPU utilization over the last 14 days. Anything averaging below 5% is a candidate for termination or scheduling.

Expected savings: 3-8% of total bill.

If your AWS bill specifically is out of control, our guide to diagnosing high AWS bills walks through a 15-minute audit process.

3. Remove unattached volumes

When you terminate a VM, its storage volumes often survive. These orphaned disks sit there quietly charging $0.08-$0.10 per GB per month on AWS EBS. Azure Managed Disks and GCP Persistent Disks have similar behavior. A 100 GB orphaned volume costs roughly $10 per month for data nobody accesses.

How to check: In AWS, go to EC2 > Volumes and filter by "State = available." In Azure, filter disks by "Unattached." In GCP, look under Compute Engine > Disks for disks not attached to any instance.

Expected savings: 1-5% of total bill.

4. Clean old snapshots

Snapshots are point-in-time backups of your storage volumes. They cost $0.05 per GB per month on AWS. Over time, teams accumulate dozens of snapshots from instances that were deleted months ago. We've seen small accounts carrying 30-50 orphaned snapshots, costing $20-$50 per month for backups of things that no longer exist.

How to check: Go to EC2 > Snapshots, sort by creation date. Delete anything older than 90 days that isn't required for compliance or disaster recovery.

Expected savings: 1-3% of total bill.

5. Review NAT gateways

NAT Gateways are one of the sneakiest cost items in cloud billing. On AWS, they charge $0.045 per hour ($32/month) just for existing, plus $0.045 per GB processed. Many small teams deploy a NAT Gateway during initial setup and never revisit it. If your private subnet resources only need to reach S3 or DynamoDB, VPC Gateway Endpoints eliminate the NAT fee entirely — and they're free.

How to check: In Cost Explorer, filter by usage type "NatGateway." For a full breakdown, see our AWS NAT Gateway pricing guide.

Expected savings: 2-10% of total bill (varies widely by architecture).

Right-sizing is the single highest-impact optimization most teams skip. Kubernetes clusters average just 10% CPU utilization, according to the CNCF 2024 FinOps survey. The same pattern holds for standalone VMs — teams overprovision "just in case" and never revisit the decision.

6. Downsize over-provisioned instances

If your VM averages below 20% CPU and uses less than half its memory over 30 days, it's oversized. Drop one instance size. On AWS, moving from an m6i.xlarge ($140/month) to an m6i.large ($70/month) saves 50% with no configuration changes beyond a restart. Azure and GCP follow similar pricing tiers.

How to check: Use CloudWatch (AWS), Azure Monitor, or GCP Ops Agent to review 30-day CPU and memory trends. AWS also provides right-sizing recommendations in Cost Explorer under "Recommendations."

Expected savings: 20-50% per instance.

7. Match storage tiers to access patterns

Cloud providers offer multiple storage classes at different price points. AWS S3 Standard costs $0.023 per GB, but S3 Infrequent Access costs $0.0125 per GB — nearly half. If you're storing logs, backups, or data accessed less than once a month, you're paying double what you should. S3 Intelligent-Tiering automates this for $0.0025 per 1,000 objects monitored.

How to check: Enable S3 Storage Lens or review access patterns in CloudWatch. Azure Blob has Hot/Cool/Archive tiers. GCP Cloud Storage has Standard, Nearline, Coldline, and Archive.

Expected savings: 30-70% on storage costs.

8. Optimize database instance sizes

Managed databases are often the second-largest line item on a cloud bill. A db.r6g.xlarge on AWS RDS with Multi-AZ costs about $350 per month. If your database CPU stays below 30% and freeable memory exceeds 50%, dropping to a db.r6g.large at $175 per month saves $2,100 per year. That's real money for a small business.

How to check: Open RDS Performance Insights or the CloudWatch dashboard for your database. Look at 30-day averages, not peak moments. Azure SQL and Cloud SQL on GCP have equivalent monitoring dashboards.

Expected savings: 25-50% on database costs.

Commitment-based pricing is where the biggest percentage savings live. Reserved Instances and Savings Plans offer 20-72% off on-demand pricing (AWS Savings Plans Pricing). Yet 37% of organizations underutilize their existing commitments (Flexera 2025). These discounts require planning, but they're the most impactful line on this checklist.

9. Evaluate Reserved Instances

If you've been running the same instance types for three or more months, Reserved Instances (RIs) are worth evaluating. A 1-year all-upfront RI on AWS saves roughly 40% compared to on-demand. A 3-year commitment saves up to 60%. The catch? You're locked into a specific instance family and region.

How to check: AWS Cost Explorer has an "RI Recommendations" tab that analyzes your usage patterns and suggests reservations. Azure Advisor and GCP recommender offer similar suggestions.

Expected savings: 30-60% on committed compute.

10. Adopt Savings Plans

Savings Plans are more flexible than Reserved Instances. Instead of committing to a specific instance, you commit to a dollar-per-hour spend level. AWS Compute Savings Plans cover EC2, Fargate, and Lambda. They offer 20-40% savings with 1-year terms. For teams whose workloads shift between instance types, Savings Plans are usually the better choice.

How to check: Go to the AWS Savings Plans recommender in Cost Explorer. It shows your recommended hourly commitment based on historical usage. Azure also offers Savings Plans for compute.

Expected savings: 20-40% on compute.

11. Use committed use discounts on GCP

Google Cloud's Committed Use Discounts (CUDs) work similarly to Reserved Instances. A 1-year commitment saves up to 37%. A 3-year commitment saves up to 55%. GCP also offers automatic sustained-use discounts of up to 30% for instances running more than 25% of the month — no commitment required. These stack with other discounts.

How to check: The GCP recommender in the Cloud Console suggests CUDs based on your usage patterns. Review under Billing > Committed use discounts.

Expected savings: 30-55% on committed compute.

12. Use spot instances for batch workloads

Spot instances (AWS), Spot VMs (Azure), and Preemptible VMs (GCP) run on spare cloud capacity at 60-90% off on-demand prices. The trade-off? They can be interrupted with short notice. That makes them ideal for batch processing, CI/CD pipelines, data analytics, and any workload that can handle restarts.

How to check: Identify workloads that don't need guaranteed uptime. Start with CI/CD runners or batch jobs. For a detailed comparison, see our reserved vs spot vs on-demand guide.

Expected savings: 60-90% on eligible workloads.

Architecture changes require the most effort, but they compound. Data egress alone costs $0.08-$0.12 per GB across major providers and represents 5-15% of typical bills (Cloudflare bandwidth analysis). The three changes below pay for themselves within weeks and keep saving as traffic grows.

13. Use a CDN to reduce egress costs

Every byte that leaves your cloud provider costs money. CloudFront, Azure CDN, or Cloud CDN caches your static content at edge locations, reducing the amount of data your origin server needs to send. For content-heavy apps, a CDN can cut egress costs by 40-70%. CloudFront's first 1 TB per month is free.

How to check: Look at your data transfer charges in Cost Explorer. If they exceed 10% of your total bill, a CDN will pay for itself. See our cloud egress costs guide for a full breakdown.

Expected savings: 40-70% on egress costs.

14. Consolidate regions

Every additional region adds cost through duplicated resources, cross-region data transfer ($0.01-$0.02/GB), and operational complexity. If you're a small business serving one geographic market, running in a single region eliminates cross-region fees entirely. Multi-region only makes sense when latency requirements or compliance demand it.

How to check: List all active regions in your cloud console. If any region holds resources that could run in your primary region without meaningful latency impact, consolidate.

Expected savings: 5-15% on data transfer and duplicated resources.

15. Implement auto-scaling

Auto-scaling adjusts your compute capacity based on actual demand. Without it, you're paying for peak capacity 24/7 even though most apps see 2-4x traffic variation throughout the day. AWS Auto Scaling, Azure Virtual Machine Scale Sets, and GCP Managed Instance Groups all offer this. The setup takes 30-60 minutes, and it keeps working forever.

How to check: Review your instance CPU patterns over a week. If utilization swings below 30% during off-peak hours, auto-scaling will save money. A service that needs 4 instances at peak but 1 at off-peak saves 50-60% on compute with proper auto-scaling configured.

Expected savings: 30-60% on compute costs.

Not every optimization is equal. Some take five minutes and save 5%. Others take a weekend and save 50%. The Flexera 2025 report shows that organizations with structured FinOps reduce waste from 27% to under 15%. This matrix helps you pick the right order.

Start at the top. The first five items take less than 30 minutes combined and typically recover 8-20% of your bill. Then move to right-sizing — it's where the highest per-item savings are. Commitment discounts come last because they work best after you've eliminated waste and right-sized. There's no point locking in a reservation for an instance you'll downsize next month. For a real-world example of working through this process on a live account, see how one team cut their cloud bill 40% in one afternoon. Before starting, it also helps to know what you're looking at — our guide to reading your AWS bill explains every line item in plain language.

Want to estimate how much you could save? Try our cloud cost calculator to model your current spend and see where the biggest gaps are.