GCP's pricing model is the cleanest of the big three — but that doesn't mean cheap. The two biggest GCP cost levers (Committed Use Discounts and BigQuery slot management) are dramatically underused. Here's the 11-pattern playbook.
1-year or 3-year commitments on vCPU + memory. Resource-based CUDs apply to any matching VM. Coverage target: 70% of baseline.
If you process >5TB/month, switch from on-demand to reserved slots. Edition: Standard for ad-hoc, Enterprise for production.
Partition + cluster every table >100GB. Use `SELECT *` only when you mean it. `INFORMATION_SCHEMA.JOBS_BY_PROJECT` reveals the worst queries.
Automatic for any VM running >25% of the month. Free money — just verify it's applying.
Cloud Run jobs, GKE node pools, Dataflow workers. 24h max lifetime — design for it.
GCP Recommender API surfaces resize candidates. `gcloud recommender recommendations list --recommender=google.compute.instance.MachineTypeRecommender`.
Lifecycle rules on buckets: Standard → Nearline at 30d, → Coldline at 90d, → Archive at 365d.
Pay per pod-second, not per node. Massive win for batch/event-driven workloads. Standard mode wins for steady-state.
HA = 2x cost. Enable only on prod. Right-size CPU/memory based on Cloud SQL Insights.
Cloud CDN cuts egress by 60% on cacheable content. Premium → Standard tier saves on intercontinental traffic.
Billing budget → Pub/Sub topic → Cloud Function that disables non-prod project at 95% threshold. Industry-leading guardrail.
CARTIE AI runs all 11 patterns against your GCP project. Read-only viewer role, no agents.
Get the auditTHE FINOPS BRIEF
Built for finance & engineering teams who are tired of paying for cloud they don't use. No fluff. Just what works.
Unsubscribe anytime. We never sell your data.