Provider deep dive / Google Cloud Platform
GCP Cost Management in 2026: native tools, automatic discounts, and optimisation levers
GCP has the most distinctive cost model of the three majors. Sustained Use Discounts apply automatically. BigQuery billing export gives you SQL on hourly cost data. This is the GCP-specific reference: tools side by side, the SUD plus CUD discount stack, the gotchas, and the playbook.
Tool inventory / 6 native cost tools
Every GCP cost tool
| Tool | Cost | What it does | What it cannot do |
|---|---|---|---|
| Cloud Billing Reports | Free | Daily granularity, group by service, project, or label. | No anomaly detection in the basic UI. |
| Billing Budgets and Alerts | Free | Programmable threshold alerts with Pub/Sub fan-out. | Pub/Sub triggers require functions or workflows to take action. |
| Recommender Hub (Active Assist) | Free | Idle VM, oversized VM, idle disk, idle IP recommendations across projects. | Recommendation lag of 24-48 hours. No automatic action. |
| BigQuery billing export | BigQuery storage and query costs apply | Hourly granularity, FOCUS 1.3 schema, SQL on cost data. | Pay attention to slot consumption when running ad-hoc queries on full export tables. |
| Pricing Calculator | Free | What-if pricing with Sustained Use Discount preview. | List rate model only, does not reflect negotiated discounts. |
| GKE cost insights | Free in console | Per-namespace and per-workload cost allocation in GKE. | Requires cost allocation feature enabled per cluster. |
Discount stack / six commitment instruments
GCP discount mechanisms
Sustained Use Discounts (SUDs)
up to 30%- Commitment
- none, automatic
- Flexibility
- Apply to most general-purpose VM families.
Unique to GCP. Discount applies progressively across the month based on usage hours per resource type.
Committed Use Discounts, resource-based
up to 55% (1 year), up to 70% (3 year)- Commitment
- 1 or 3 years per machine type
- Flexibility
- Locked to family in a region.
Appropriate for stable workloads. Stack with SUDs for the steady-state floor.
Committed Use Discounts, spend-based
up to 28% (1 year), up to 46% (3 year)- Commitment
- 1 or 3 years, hourly $ commit
- Flexibility
- Across general-purpose families.
Closest GCP equivalent to AWS Compute Savings Plans.
Spot VMs
up to 80%- Commitment
- none
- Flexibility
- Eviction-based, dynamic pricing.
Replaced legacy Preemptible VMs. No 24-hour limit.
Preemptible VMs (legacy)
up to 80%- Commitment
- none
- Flexibility
- Maximum 24-hour runtime, 30-second shutdown notice.
Still available but Spot VMs are the default.
Free Tier and credits
varies- Commitment
- none
- Flexibility
- Always-free quotas on f1-micro, 5 GB Cloud Storage, 1 GB BigQuery analysis.
Useful for sandbox accounts and free CI tier.
Surprise line items / not on the pricing page
GCP-specific hidden cost catalogue
Eight charges that catch GCP customers, especially teams migrating from AWS who assume parity.
| Item | List rate | How to avoid it |
|---|---|---|
| Block storage premium vs AWS/Azure | Persistent Disk SSD $0.170/GB/mo (vs S3-backed EBS gp3 ~$0.080/GB/mo on AWS) | Use Hyperdisk Balanced for IOPS-bound workloads. Right-size disks. Use snapshots aggressively for cold data. |
| GKE Standard control plane | $0.10/hour ($73/mo) per cluster | Consolidate dev clusters. GKE Autopilot includes the control plane in pod pricing. |
| Cloud NAT | $0.045/GB processed + $0.0014/hour per VM gateway endpoint | Use Private Service Connect for Google APIs. Co-locate egress through dedicated NAT subnets. |
| Persistent Disk snapshots | $0.026/GB/mo standard, $0.050/GB/mo regional | Snapshot lifecycle policies. Move long-term snapshots to Cloud Storage Coldline. |
| Network Intelligence Center flow logs | $0.04 per flow log entry | Sample VPC flow logs. Disable on subnets where not required for security or troubleshooting. |
| Cloud Logging ingestion | $0.50/GB above 50 GB/project/month free | Exclusion filters at sink level. Send container debug logs to Cloud Storage instead of Cloud Logging. |
| External IP addresses (static, unattached) | $0.01/hour ($7.30/mo) | Audit unattached static IPs monthly. Reserve only what is in use. |
| Cross-region Cloud Storage replication | Storage at multi-region price + replication egress | Use regional buckets for non-critical data. Reserve multi-region for compliance use cases. |
Optimisation playbook / six moves ranked by impact
GCP savings playbook
Let SUDs work before layering CUDs
Typical impact: 20-30% baseline before commitments
Sustained Use Discounts apply automatically and tier up the longer a resource runs in the month. Analyse 3 months of SUD discount data first. CUDs should cover the steady-state floor, not the discounted layer.
Move variable workloads to GKE Autopilot
Typical impact: 30-50% on container costs
Autopilot charges only for pod resources, not node overhead. The control plane is included. For variable workloads it removes both the control plane fee and the bin-packing wastage of Standard.
Run batch and training on Spot VMs
Typical impact: 60-80%
Spot VMs replaced Preemptible VMs. No 24-hour limit, dynamic pricing. Suitable for batch processing, ML training, render farms, and CI runners.
Tier Cloud Storage to Nearline, Coldline, Archive
Typical impact: 20-90%
Lifecycle rules move objects based on age or access. Autoclass automates the decision and removes retrieval fee surprises in mixed-access buckets.
Replace static IPs and disks identified by Recommender
Typical impact: 5-15%
Active Assist surfaces idle IPs, unattached disks, and oversized VMs. Apply quarterly. Memory-based rightsizing requires the Ops Agent on the VM.
Reserve BigQuery slots for predictable analytics
Typical impact: 10-40% on heavy BQ usage
On-demand BigQuery is $5-7.50 per TB scanned. Slot reservations (Editions: Standard, Enterprise, Enterprise Plus) make heavy analytics predictable and often cheaper.
Common questions
FAQ
Are GCP Sustained Use Discounts really automatic?+
Yes. SUDs apply progressively based on the percentage of the month a resource runs. There is no commitment, no purchase, and nothing to manage. The discount is calculated per resource family per region. For an N2 VM running the entire month, the effective discount tops out around 20-30%, depending on machine type.
When should I pick CUDs over SUDs?+
SUDs are automatic and apply to general-purpose families. CUDs require commitment and lock you to a machine type or hourly spend, but offer up to 70% off (resource-based, 3 years). The right pattern is to use SUDs as the baseline for everything, then layer CUDs on the steady-state floor identified by 3+ months of utilisation data.
Why is GCP block storage more expensive?+
Persistent Disk SSD pricing has historically run higher than AWS gp3 or Azure Premium SSD. Hyperdisk Balanced (introduced 2023) closes some of the gap with provisioned IOPS pricing. For SSD-heavy workloads, model the storage cost on each provider before committing to GCP.
Is GKE Autopilot cheaper than GKE Standard?+
It depends on cluster utilisation. Autopilot charges per pod request (vCPU, memory, ephemeral storage) and includes the control plane. Standard charges for nodes and the control plane separately. At low utilisation (under 50% bin-packing), Autopilot is usually cheaper. At high utilisation (over 70%), Standard with right-sized nodes can win.
What does BigQuery billing export give me that the UI cannot?+
Hourly cost data, custom dimensions through SQL joins to label tables, FOCUS 1.3 schema for cross-provider analysis, and unlimited retention. Watch the BigQuery slot consumption when querying full export tables: a poorly written query against the cost export can itself become a cost line item.
Continue reading