GCP Cost Optimization Guide for Growing Companies (2025)

Your Google Cloud bill is growing faster than your revenue. Sound familiar?

Most GCP cost optimization guides are written for enterprises with dedicated FinOps teams and million-dollar cloud budgets. But what about growing companies—startups and scale-ups—who need practical solutions without the enterprise overhead?

This guide is for you. No enterprise jargon. No six-month implementation plans. Just practical strategies you can implement today to reduce your GCP costs.

Why GCP Costs Spiral for Growing Companies

Before diving into solutions, let's understand why costs get out of control:

1. No visibility: You're reacting to bills, not monitoring spend
2. Over-provisioned resources: You picked instance sizes based on guesses
3. Idle resources: That test VM from 3 months ago is still running
4. No commitment strategy: You're paying on-demand rates for predictable workloads
5. Missing discounts: SUDs, CUDs, and Spot VMs could save 30-70%

Quick Wins (Implement Today)

These take less than an hour and cost nothing:

1. Enable Billing Export to BigQuery

Why it matters: Without billing export, you're flying blind. GCP Console gives you summary views, but billing export gives you the raw data to actually analyze costs—down to individual resources.

Critical: Billing export data is NOT retroactive. If you enable it today, you only get data from today forward. Enable it now.

Step 1: Create a project for billing data

We recommend creating a dedicated project (e.g., billing-analytics) for your billing export. This keeps billing data isolated from production, which is important when granting access to cost optimization tools—you're only sharing billing data, not your production resources.

You can use an existing project, but a dedicated one simplifies access control and security reviews.

Multiple billing accounts? Repeat this setup for each one. See the Multi-Billing Account Visibility section below for managing costs across accounts.

Step 2: Create a BigQuery dataset

1. Open BigQuery in your billing project
2. Create a dataset (e.g., billing_export)
3. Choose a multi-region location (US or EU) for durability

Go to Billing → Billing export. You'll see four export options—but you only need one:

Enable "Detailed usage cost" — Click "Edit Settings", select your project and dataset, then save.

This export contains everything you need for cost optimization:

• Cost breakdown by service, project, SKU, and region
• Resource-level detail (which specific VM, disk, or query)
• CUD and SUD savings already applied
• Hourly granularity for trend analysis

Storage cost: Negligible—typically a few dollars per month. Worth it for the visibility.

Step 4: Wait for data

Data starts flowing within 24-48 hours. The first export includes the current month's data up to that point. GCP creates a table named gcp_billing_export_resource_v1_<YOUR_BILLING_ACCOUNT_ID> in your dataset.

2. Set Budget Alerts

Why it matters: Surprise bills happen when nobody's watching. Budget alerts are your early warning system.

Recommended setup:

• Set budget at 80% of expected monthly spend
• Alert at 50%, 90%, and 100% thresholds
• Send to engineering leads, not just finance

1. Billing → Budgets & alerts
2. Create budget
3. Set amount and alert thresholds
4. Add email recipients

3. Review Google Recommender Suggestions

Why it matters: Google already knows where you're wasting money. The Recommender API analyzes your usage and suggests optimizations.

Common recommendations:

• Idle VM instances (not used in 15+ days)
• Oversized instances (using <50% CPU)
• Unattached persistent disks
• Unused static IP addresses
• Idle Cloud SQL instances

Pro tip: Check recommendations weekly. New ones appear as usage patterns change. Or use a dashboard that surfaces these automatically.

Medium-Term Optimizations (This Month)

These require some planning but deliver significant savings:

4. Right-Size Your Compute Instances

Most companies over-provision by 40-60%. A VM that "might need" 8 vCPUs usually runs fine on 4.

How to right-size:

1. Check actual utilization: Console → Compute Engine → VM instances → Observability
2. Look for patterns: If average CPU is under 40%, consider downsizing
3. Use Recommender: It suggests specific machine types based on actual usage
4. Test first: Resize in staging before production

• n2-standard-8 → n2-standard-4: ~50% savings on that instance
• Across 10 oversized VMs: $500-2,000/month savings

5. Clean Up Idle Resources

Idle resources are pure waste. Common culprits:

1. Run Recommender report
2. List resources older than 90 days
3. Check with owners before deleting
4. Delete or tag for retention

6. Implement Cost Labels

Labels let you see WHERE money goes. Without them, you just see THAT money went.

Recommended labels:

• environment: prod, staging, dev, test
• team: engineering, data, platform
• project: customer-api, data-pipeline, internal-tools
• owner: email of responsible person

1. Create labeling policy document
2. Add labels to Terraform/IaC templates
3. Use Cloud Asset Inventory to audit unlabeled resources
4. Set up alerts for resources missing required labels

Committed Use Discounts: When They Make Sense for SMBs

CUDs offer significant savings (28-55%) but require commitment. Here's when they make sense for growing companies:

When to Buy CUDs

Buy CUDs if:

• You have stable, predictable workloads
• Monthly GCP spend exceeds $3,000
• Workloads will exist for 12+ months
• You can commit to specific regions

• Spend is under $2,000/month
• You're still figuring out architecture
• Workloads change frequently
• You might migrate to another cloud

CUD Types Explained

Resource-based CUDs:

• Commit to specific vCPUs and memory
• Up to 55% savings (3-year)
• Best for: Stable production workloads
• Risk: Locked to specific machine family and region

• Commit to hourly spend amount
• Up to 46% savings (3-year)
• Best for: Variable workloads across services
• Benefit: Applies to Compute, GKE, Cloud SQL

Quick CUD Calculator

For a $5,000/month GCP compute spend:

Pro tip: Analyzing your usage patterns to find the right CUD coverage takes time. Tools that analyze your billing data can recommend optimal CUD purchases based on actual usage.

Advanced: Multi-Billing Account Visibility

Many growing companies end up with multiple billing accounts:

• Acquired company kept their account
• Different accounts for different clients
• Separate accounts for dev vs prod

Solutions:

1. Manual consolidation: Export all billing data to one BigQuery dataset (complex)
2. Third-party tools: Use tools like GCP FinOps that aggregate across billing accounts automatically (recommended)
3. Organization-level billing: Migrate all projects to one billing account (disruptive)

Tools Comparison for SMBs

Free Native Tools

SMB-Friendly Third-Party Tools

Enterprise Tools (Often Overkill)

CloudHealth, Apptio, Spot.io, CAST AI—these are enterprise-priced and require lengthy implementation projects. Great for large organizations, but often overkill for growing companies spending under $50K/month.

Monthly Cost Review Checklist

Run this checklist monthly:

Week 1: Quick Review (30 min)

• Check Recommender for new suggestions
• Review budget alert history
• Compare spend to last month

Week 2: Resource Audit (1 hour)

• Identify idle VMs (CPU < 5% for 14+ days)
• List unattached disks
• Check for unused IP addresses
• Review snapshot age and necessity

Week 3: Optimization Actions (2 hours)

• Resize oversized instances
• Delete confirmed idle resources
• Update labels on new resources
• Schedule resources that don't need 24/7

Week 4: Planning (30 min)

• Review CUD utilization
• Plan next month's optimizations
• Update stakeholders on savings achieved

Getting Started

1. Today: Enable billing export (5 minutes)
2. This week: Set up budget alerts, run Recommender
3. This month: Audit idle resources, right-size top 10 VMs
4. Next month: Evaluate CUDs for stable workloads

Key Takeaways

• Enable billing export immediately—data isn't retroactive
• Check Recommender weekly—Google tells you what's wasted
• Right-size aggressively—most VMs are over-provisioned
• Start CUDs conservatively—cover 30-50% of stable spend
• Monthly audits matter—idle resources accumulate silently

I'm building GCP FinOps to help growing companies optimize GCP costs without enterprise complexity. Have questions? Reach out on LinkedIn.