Cloud Cost Optimization Guide: 10 Strategies That Work (2026)
Practical strategies to reduce your AWS, Azure, and GCP bills. Covers rightsizing, reservations, spot instances, and architectural optimizations.
TL;DR
- • The 3 biggest levers: rightsize instances (save 20-40%), use commitment discounts on baseline usage (save 30-60%), and delete zombie resources (save 5-15%).
- • For small/medium workloads, choosing Hetzner or DigitalOcean over Big 3 can save 50-80% before any optimization.
- • Combine strategies: reserved/committed for baseline, spot for burst, and smaller providers for non-critical workloads like CI/CD and staging.
Cloud bills grow faster than most teams expect. These strategies work across all 8 major providers to reduce spending without sacrificing performance. For Big 3 providers (AWS, Azure, GCP), the focus is on commitment discounts and rightsizing. For smaller providers (DigitalOcean, Hetzner, Linode, Oracle, Vultr), the strategy shifts to choosing the right provider and tier from the start. Each strategy includes specific actions and typical savings you can expect.
Side-by-Side Comparison
| Feature | AWS | Azure | GCP | DO | Hetzner | Linode | OCI | Vultr |
|---|---|---|---|---|---|---|---|---|
| Cost management tool | Cost Explorer + Budgets | Cost Management + Advisor | Cost Management + Recommender | Billing Dashboard | Cloud Console Billing | Cloud Manager Billing | Cost Analysis + Budgets | Billing Dashboard |
| Rightsizing tool | Compute Optimizer | Azure Advisor | VM Rightsizing Recommender | Monitoring + manual | Manual (Grafana integration) | Cloud Manager metrics | Operations Insights | Manual (monitoring API) |
| Commitment discounts | Savings Plans / RIs | Reservations / Savings Plans | CUDs / Sustained Use Discounts | — | — | — | Universal Credits | — |
| Spot / preemptible | Spot Instances (up to 90% off) | Spot VMs (up to 90% off) | Spot VMs (up to 91% off) | — | — | — | Preemptible (50% off) | — |
| ARM instances | Graviton (20-40% cheaper) | Ampere Altra (up to 50% cheaper) | Tau T2A (up to 40% cheaper) | Premium AMD only | Ampere Altra (CAX series) | — | Ampere A1 (1c/6GB free) | — |
| Auto-scaling | Auto Scaling Groups | VM Scale Sets | Managed Instance Groups | Droplet Autoscale | — (API-based) | NodeBalancer + API | Autoscaling configs | Load Balancer + API |
| Storage tiers | S3 Intelligent-Tiering | Blob Lifecycle Policies | Autoclass | Spaces (single tier) | Object Storage (single tier) | Object Storage (single tier) | Standard/Infrequent/Archive | Object Storage (single tier) |
Strategy 0: Choose the Right Provider from the Start
Before optimizing within a provider, consider whether you're on the right provider. For many workloads — web applications, APIs, databases, CI/CD — smaller providers offer dramatically lower prices with comparable performance. This is the single highest-impact optimization for small and medium workloads.
Example comparison for 4 vCPU / 16 GB compute: AWS m5.xlarge on-demand: $140/mo. AWS with 1yr Savings Plan: $88/mo. Hetzner CX41: $15/mo. DigitalOcean Premium: $48/mo. The Hetzner server is 89% cheaper than AWS on-demand and 83% cheaper than AWS with a Savings Plan.
When to stay on Big 3: You need provider-specific managed services (DynamoDB, BigQuery, Azure AD integration). You need 50+ global regions. You have compliance requirements that mandate a specific provider. You need enterprise SLAs and support contracts.
1. Rightsize Your Instances
Most cloud instances are over-provisioned. Check actual CPU and memory utilization — if averages are below 40%, downsize. Use each provider's tools: AWS Compute Optimizer, Azure Advisor, GCP VM Rightsizing Recommender. On smaller providers, check CPU/memory graphs in the dashboard and downgrade if consistently underutilized.
Example: Downsizing from m5.xlarge (4 vCPU, 16 GB, $140/mo) to m5.large (2 vCPU, 8 GB, $70/mo) saves $70/mo per instance. If you have 10 over-provisioned instances, that's $700/mo ($8,400/yr) saved with zero architecture changes. Typical savings: 20-40%. Start with dev/test environments, then move to production. Review quarterly as usage patterns change.
2. Use Commitment Discounts for Baseline Usage
For workloads that run 24/7 on AWS/Azure/GCP, commitment discounts offer 30-72% savings. Commit for at least your baseline usage — the portion that never drops below a certain level. Use Savings Plans (AWS/Azure) for flexibility across instance families. Reserve 60-70% of baseline and use on-demand for peaks.
Example: 10 m5.xlarge instances running 24/7 on AWS: On-demand: $1,400/mo. With 1yr Compute Savings Plan (no upfront): $883/mo (37% savings = $517/mo). With 3yr all-upfront RI: $577/mo (59% savings = $823/mo). On GCP, Sustained Use Discounts apply automatically after 25% monthly usage, saving up to 30% without any commitment.
3. Leverage Spot Instances for Fault-Tolerant Workloads
Spot instances offer 60-90% savings for workloads that can handle interruptions: batch jobs, CI/CD, containerized services, ML training. Mix spot with on-demand to maintain minimum capacity.
Example: Running 20 c5.xlarge instances for ML training: On-demand cost: $0.170/hr × 20 = $3.40/hr ($2,482/mo). Spot cost: ~$0.040/hr × 20 = $0.80/hr ($584/mo). Monthly savings: $1,898. Alternative: Run the same training on 20 Hetzner CCX23 instances at $0.031/hr × 20 = $0.62/hr ($452/mo) — cheaper than spot with zero interruption risk.
4. Optimize Storage Classes
Don't store cold data in hot storage. Use lifecycle policies to automatically move data to cheaper tiers: S3 Intelligent-Tiering (AWS), Blob Lifecycle Management (Azure), Autoclass (GCP). Oracle offers Standard, Infrequent Access, and Archive tiers. Smaller providers (DO, Hetzner, Linode, Vultr) typically have single-tier object storage at $0.02/GB/mo — already cheaper than Big 3 hot storage.
Example: 10 TB on AWS S3 Standard: $230/mo. Moved to S3 Glacier Instant Retrieval: $40/mo (83% savings). On DigitalOcean Spaces: $50/mo flat (250 GB included) + $0.02/GB. For archival data on Big 3 providers, lifecycle policies are essential. For active data, compare against smaller providers' flat-rate storage.
5. Minimize Data Transfer Costs
Egress (data transfer out) is the hidden cloud cost that surprises most teams. Big 3 providers charge $0.08-0.12/GB for internet egress, while smaller providers are dramatically cheaper: Hetzner includes 20 TB/mo free, DigitalOcean includes 1-6 TB depending on plan, Vultr includes 1-4 TB.
Example: 5 TB/mo egress on AWS: ~$450/mo. Same on Hetzner: $0 (included). Same on DigitalOcean: typically $0 (included with most plans). Use CDN (CloudFront, Azure CDN, Cloud CDN, Cloudflare) to serve content from edge locations. Keep traffic within the same region and AZ. Use VPC endpoints / Private Link to avoid NAT gateway charges on Big 3.
6. Use Serverless for Bursty Workloads
If your workload has idle periods, serverless (Lambda, Functions, Cloud Functions) can be cheaper than always-on instances. For APIs with fewer than 10M requests/month, serverless is almost always cheaper. For containers, use serverless options (Fargate, Container Apps, Cloud Run) that scale to zero.
Compare against equivalent always-on compute: A Lambda function handling 1M requests/mo (128 MB, 200ms avg) costs about $0.20/mo. A t3.micro running 24/7 to do the same costs $7.50/mo. But at 100M requests/mo, Lambda costs ~$20/mo while a t3.medium at $30/mo might handle the load with less complexity.
7. Delete Unused Resources
Zombie resources are pure waste: unattached EBS volumes, idle load balancers, stopped instances with provisioned storage, unused Elastic IPs, orphaned snapshots. Run a monthly audit using each provider's tools. Tag all resources with an owner and project, and delete anything that hasn't been used in 30 days.
This applies to all providers. On smaller providers, common waste includes: forgotten Droplets/Linodes running staging environments, unused block storage volumes, old backups. Typical savings: 5-15% of total bill. Set up a monthly calendar reminder to audit.
8. Use ARM-Based Instances
ARM processors offer 20-40% better price-performance than equivalent x86. AWS Graviton (m7g, c7g series), Azure Ampere Altra (Dpsv5), GCP Tau T2A, and Hetzner CAX series all use ARM chips. Oracle's Ampere A1 shapes include a generous Always Free allocation (4 OCPU, 24 GB).
Example: AWS m5.xlarge (x86, 4 vCPU, 16 GB): $0.192/hr. AWS m7g.xlarge (Graviton, 4 vCPU, 16 GB): $0.163/hr — 15% cheaper with better performance. Hetzner CAX31 (Ampere ARM, 8 vCPU, 32 GB): ~$0.023/hr. Most workloads run on ARM without modification: web servers, containerized apps, databases, CI/CD.
9. Set Budget Alerts
Catch unexpected spend before it becomes a problem. Set budget alerts at 50%, 80%, and 100% of your expected monthly spend. All Big 3 providers offer free budget alerting (AWS Budgets, Azure Cost Alerts, GCP Budget Alerts). Oracle offers budget alerts in Cost Management. Smaller providers have simpler billing dashboards but you can set up alerts via their APIs or use third-party tools.
This doesn't save money directly but prevents surprise bills. A misconfigured auto-scaling group or forgotten GPU instance can easily cost $1,000+ in a single weekend.
10. Multi-Provider Strategy
The most cost-effective approach for many organizations is using multiple providers for different purposes. Run your core application on the provider whose managed services you depend on. Offload commodity compute to a cheaper provider.
Example multi-provider setup: Production API on AWS (using DynamoDB, SQS, Lambda): optimized with Savings Plans. CI/CD runners on Hetzner: $0.021/hr vs $0.170/hr on AWS = 88% savings. Staging environment on DigitalOcean: $48/mo vs $140/mo on AWS. Static site hosting on Cloudflare Pages: free vs S3+CloudFront complexity. This blended approach can cut total cloud spend by 30-50% compared to running everything on a single Big 3 provider.
Frequently Asked Questions
What is the easiest way to reduce cloud costs?
Rightsizing instances and deleting unused resources are the easiest wins — they require no code changes. Check your instance utilization and downsize anything consistently below 40% CPU/memory. Then audit for zombie resources. For the biggest single impact, evaluate whether a cheaper provider would work for some of your workloads.
How much can I save with cloud cost optimization?
Most organizations can reduce their cloud bill by 20-40% through rightsizing, commitment discounts, spot instances, and storage optimization. Organizations willing to adopt a multi-provider strategy can save 50-80% on commodity compute by using providers like Hetzner or DigitalOcean for appropriate workloads.
Should I use reserved instances or spot instances?
Use both for different workloads. Reserved instances (or Savings Plans) are for your baseline workload that runs 24/7 — save 30-60%. Spot instances are for variable and fault-tolerant workloads — save 60-90%. A typical Big 3 setup: 60% reserved, 20% on-demand, 20% spot. Alternatively, run stable workloads on a cheaper provider at fixed on-demand rates.
Which cloud provider is cheapest overall?
There is no single cheapest provider for all workloads. For commodity compute, Hetzner is consistently the cheapest (80-90% below AWS on-demand). For managed services, GCP is often cheapest with sustained use discounts. For enterprise deals, Azure can be cheapest with EA agreements. For specific services (like data warehousing), compare BigQuery vs Redshift vs Synapse pricing directly. Use our comparison pages to check specific services.