2026-02-25 · CalcBee Team · 8 min read
Cloud Cost Optimization: Strategies to Cut Your AWS, Azure, and GCP Bill
Cloud bills are the new electricity bills — everyone has them, nobody fully understands them, and they almost always cost more than expected. The average company wastes 30–35% of its cloud spend on unused or oversized resources. Here's how to find and eliminate that waste.
The Optimization Framework
Cloud cost optimization has four levels, each building on the previous:
| Level | Strategy | Typical Savings | Effort |
|---|---|---|---|
| 1 | Visibility — know what you're spending | Prerequisite | Low |
| 2 | Right-sizing — match resources to actual usage | 15–25% | Medium |
| 3 | Pricing models — reservations, savings plans, spot | 20–40% | Medium |
| 4 | Architecture — redesign for cost efficiency | 30–60% | High |
Model your potential savings with our Cloud Cost Optimization Calculator.
Level 1: Visibility — Know What You're Spending
You can't optimize what you can't see. Start with these actions:
| Action | Tool/Method |
|---|---|
| Enable cost allocation tags | Tag every resource with team, project, environment |
| Set up billing alerts | Threshold alerts at 50%, 80%, 100% of budget |
| Review monthly cost reports | AWS Cost Explorer, Azure Cost Management, GCP Billing |
| Identify top-5 cost drivers | Usually compute, storage, data transfer, databases |
| Track cost per customer/unit | Links cloud spend to business value |
The typical breakdown for a SaaS company:
| Service | % of Cloud Bill |
|---|---|
| Compute (VMs, containers) | 40–50% |
| Data storage | 15–25% |
| Databases (managed) | 10–20% |
| Data transfer (egress) | 5–15% |
| Other services | 10–20% |
Level 2: Right-Sizing
The single most impactful quick win. Most VMs are oversized for their actual workload.
How to Identify Oversized Resources
| Metric | Oversized If | Action |
|---|---|---|
| CPU utilization | Avg < 20% for 2+ weeks | Downsize instance type |
| Memory utilization | Avg < 30% for 2+ weeks | Downsize or switch type |
| Storage capacity | < 50% used, not growing | Resize volume |
| Database connections | < 20% of max | Downsize DB instance |
Right-Sizing Example
| Current Instance | MonthLy Cost | Avg CPU | Recommended | New Cost | Savings |
|---|---|---|---|---|---|
| m5.2xlarge (8 vCPU) | $280 | 12% | m5.large (2 vCPU) | $70 | $210/mo |
| r5.xlarge (4 vCPU, 32GB) | $182 | 8% CPU, 15% RAM | r5.large (2 vCPU, 16GB) | $91 | $91/mo |
| db.r5.2xlarge | $620 | 5% | db.r5.large | $155 | $465/mo |
Across 50 instances, right-sizing typically saves $3,000–$10,000/month.
Zombie Resources
| Resource Type | How to Find | Typical Waste |
|---|---|---|
| Unattached EBS volumes | Filter: state = "available" | $0.08–$0.10/GB/month |
| Unused Elastic IPs | AWS: unassociated IPs | $3.65/IP/month |
| Old snapshots | Age > 6 months, no restore history | Varies widely |
| Idle load balancers | Zero connections for 2+ weeks | $18–$25/month each |
| Unused NAT gateways | Zero data processed | $32/month + data |
Level 3: Pricing Models
Reserved Instances / Savings Plans
Commit to 1–3 years of usage for significant discounts:
| Commitment | Discount vs. On-Demand |
|---|---|
| No commitment (on-demand) | Baseline |
| 1-year, no upfront | 20–30% |
| 1-year, all upfront | 30–40% |
| 3-year, no upfront | 35–45% |
| 3-year, all upfront | 50–60% |
Best practice: Reserve your steady-state baseline (the minimum capacity you always use), and use on-demand or spot for variable workloads.
Spot/Preemptible Instances
Use spare cloud capacity at 60–90% discount:
| Provider | Name | Discount | Interruption Risk |
|---|---|---|---|
| AWS | Spot Instances | 60–90% | 2-minute warning |
| Azure | Spot VMs | 60–90% | 30-second warning |
| GCP | Preemptible/Spot VMs | 60–91% | 30-second warning |
Ideal for: Batch processing, CI/CD, data analytics, stateless workers, dev/test environments.
Not suitable for: Production databases, single-instance services, anything requiring guaranteed uptime.
Storage Tiering
| Storage Tier | AWS Name | Cost/GB/month | Best For |
|---|---|---|---|
| Hot (frequent access) | S3 Standard | $0.023 | Active data |
| Warm (infrequent) | S3 IA | $0.0125 | Monthly access |
| Cold (rare access) | S3 Glacier Instant | $0.004 | Quarterly access |
| Archive | S3 Glacier Deep | $0.00099 | Compliance, 12-hour retrieval OK |
Moving 10 TB from Standard to Glacier Instant saves $190/month.
Level 4: Architecture Optimization
| Strategy | Savings Potential | Complexity |
|---|---|---|
| Containerize (ECS/EKS/GKE) | 30–50% | High |
| Serverless (Lambda/Functions) | 40–70% for variable workloads | Medium |
| Auto-scaling | 20–40% | Medium |
| Multi-region optimization | 10–20% (cheaper regions) | Medium |
| Caching layers (ElastiCache, CDN) | 20–40% on compute + data transfer | Medium |
Scheduling Non-Production Resources
Dev and staging environments don't need to run 24/7:
| Schedule | Hours/Week | Savings vs. 24/7 |
|---|---|---|
| Business hours only (9–6 M–F) | 45 | 73% |
| Extended hours (7–10 M–F) | 75 | 55% |
| Business hours + Saturday | 54 | 68% |
A dev environment costing $2,000/month drops to $540 with business-hours-only scheduling.
Quick Win Checklist
- Delete unattached storage volumes (5-minute task, immediate savings)
- Release unused static IPs
- Schedule dev/staging to shut down at night
- Enable S3 Intelligent-Tiering for all non-critical buckets
- Right-size your top-10 most expensive instances
- Reserve steady-state compute with 1-year Savings Plan
- Set up billing alerts at key thresholds
Frequently Asked Questions
How much should cloud cost for a typical SaaS startup?
As a rough benchmark, cloud infrastructure should be 15–25% of revenue for early-stage SaaS and 5–15% at scale. If you're above these ranges, optimization will have significant impact.
Should I use reserved instances or savings plans?
Savings Plans (AWS) and Committed Use Discounts (GCP) offer more flexibility than traditional Reserved Instances. They apply automatically across instance families and sizes, making them the better default choice.
How often should I review cloud costs?
Monthly for budget tracking, quarterly for right-sizing and reservation adjustments, annually for architecture-level optimization reviews.
Is multi-cloud cheaper than single-cloud?
Rarely. Multi-cloud adds operational complexity, reduces volume discounts, and requires expertise across platforms. Use multi-cloud for resilience or vendor requirements, not cost savings.
Cloud cost optimization isn't a one-time project — it's an ongoing practice. Start with visibility, capture the quick wins, then systematically work through right-sizing and pricing commitments. The 30% you save goes straight to your bottom line.
Category: Tech
Tags: Cloud costs, AWS, Azure, GCP, Cost optimization, DevOps, Infrastructure, Cloud computing