Cutting AWS & GCP Cloud Costs: A Practical Guide

Cloud bills creep. A team ships fast, nobody owns cost, and six months later finance is asking why the AWS invoice doubled. I've helped teams cut cloud spend by 30-70%, not through one magic switch, but through a handful of high-leverage changes plus the habits that keep costs from creeping back.

Here's the practical playbook, for both AWS and GCP.

First: get visibility

You can't optimise what you can't see. Before changing anything:

• AWS: enable Cost Explorer and the Cost and Usage Report; tag resources by team/environment.
• GCP: use the Billing reports and BigQuery billing export; label resources.

Tagging/labelling is the unglamorous foundation, without it, you can't attribute cost to anything.

The biggest levers

1. Right-size compute (20-40% savings)

Most instances run at 10-20% utilisation. Use AWS Compute Optimizer or GCP Recommender to find over-provisioned VMs, then resize. This is usually the single biggest win.

2. Autoscale instead of over-provisioning (30-50% on variable load)

Static fleets sized for peak waste money off-peak. Use Auto Scaling Groups (AWS) or Managed Instance Groups (GCP), and on Kubernetes use the Cluster Autoscaler or Karpenter. Pay for what you actually use.

3. Spot / preemptible for fault-tolerant work (60-90% savings)

CI runners, batch jobs, and stateless workers don't need on-demand pricing.

• AWS: Spot Instances (and Spot for EKS node groups).
• GCP: Spot VMs / preemptible instances.

Just design for interruption, checkpoint long jobs and keep a small on-demand fallback.

4. Commit for steady-state (up to ~70%)

For the baseline you always run:

• AWS: Savings Plans or Reserved Instances.
• GCP: Committed Use Discounts.

Cover your stable floor with commitments, burst on spot/on-demand.

5. Storage tiering (15-30%)

• Move infrequently-accessed data to cheaper tiers (S3 Intelligent-Tiering / Glacier; GCS Nearline / Coldline) with lifecycle rules.
• On AWS, switch gp2 EBS volumes to gp3: cheaper and faster.
• Delete unattached volumes, old snapshots, and orphaned disks. They quietly accumulate.

6. Cut data-transfer and idle costs

• Use a CDN (CloudFront / Cloud CDN) instead of serving large assets from origin.
• Replace idle NAT gateways and unused load balancers, they bill by the hour whether used or not.
• Use VPC endpoints / Private Google Access to avoid egress charges to cloud APIs.

Make it stick: FinOps habits

One-time cleanups regress. What keeps costs down:

• Budgets + anomaly alerts. AWS Budgets / Cost Anomaly Detection; GCP budget alerts. Get pinged before a surprise, not after.
• Cost in code review. Treat a big new resource like any other change: reviewed, justified.
• Showback by team. When teams see their own spend (via tags/labels), behaviour changes.
• A regular cost review. A monthly 30-minute look at the top line items catches creep early.

A sensible rollout order

1. Week 1: visibility (tagging, cost reports, anomaly alerts) + delete obvious waste.
2. Weeks 2-4: right-size compute, switch gp2→gp3, add lifecycle rules.
3. Month 2: autoscaling + spot for fault-tolerant workloads.
4. Month 2-3: commitments for steady-state; embed the FinOps habits.

The mindset

Cost optimisation isn't a project you finish, it's a practice. The teams that stay efficient treat cloud spend as an engineering metric, owned and reviewed like latency or error rate. Done right, it funds more features instead of bigger invoices.

---

Need help getting your AWS or GCP bill under control? I do cloud cost optimisation as part of my consulting work, reach out.