Published September 14, 2026. Data verified from Google Cloud Pricing pages, GCP Committed Use Discounts docs, BigQuery pricing, and Active Assist documentation.
GCP cost optimization in 2026 is driven by Committed Use Discounts (CUDs), Sustained Use Discounts (SUDs), Spot VMs, and BigQuery slot commitments. Resource-based CUDs deliver up to 65% off vCPU and memory on Compute Engine; spend-based CUDs deliver up to 52% off across any GCP service for a 3-year commit. BigQuery flat-rate pricing is cheaper than on-demand above 100 TB scanned per month. Layered, mid-size teams routinely cut GCP bills 35-50% within two quarters (Google Cloud Pricing, September 2026).
Last verified: Sep 14, 2026.
At a glance
- Resource-based CUDs: up to 65% off vCPU and memory for 3-year commit
- Spend-based CUDs: up to 52% off any GCP service for 3-year commit
- Sustained Use Discounts: automatic up to 30% off for full-month VMs
- GCP Spot VMs: up to 91% off on-demand
- BigQuery flat-rate: $10-$110/slot/month depending on edition
- Active Assist: free FinOps recommendations across Compute, Storage, GKE
The GCP cost optimization playbook for 2026
The first lever is Committed Use Discounts (CUDs) because they cover steady-state workloads with no code change. Resource-based CUDs commit to a specific vCPU and memory amount and deliver the highest discount, up to 65% off on-demand for a 3-year commit. CUDs are flexible across instance families within the same region. Spend-based CUDs commit a dollar amount that applies to any GCP service in a region and are easier to manage when the workload mix shifts (GCP Committed Use Discounts docs, September 2026).
The recommended CUD commit size is 60-70% of trailing 30-day on-demand spend. CUDs can be purchased via the GCP console, gcloud CLI, or the Cloud Billing Budget API. They auto-renew at the end of the term unless canceled.
GCP CUD discount ladder (us-central1, September 2026)
| CUD type | Commit | 1-year discount | 3-year discount | Flexibility |
|---|---|---|---|---|
| Resource-based CUD (vCPU/memory) | vCPU + GB | ~37% | ~65% | Any family in region |
| Resource-based CUD (sole-tenant) | vCPU + GB | ~35% | ~57% | Any sole-tenant in region |
| Spend-based CUD (compute) | Dollar amount | ~28% | ~52% | Any service in region |
| Spend-based CUD (BigQuery) | Dollar amount | ~25% | ~50% | BigQuery only |
| Sustained Use Discount | Automatic | ~20% (full month) | ~30% (full month) | Auto, N1/N2 only |
Source: GCP Committed Use Discounts documentation, Google Cloud Pricing (September 2026).
Spot VMs, N2/N2D pricing, and custom machine types
GCP Spot VMs (formerly Preemptible VMs) are the highest-discount compute lever on GCP, delivering up to 91% off on-demand prices. Spot VMs can run for up to 24 hours before being preempted with 30 seconds of notice. Spot is suitable for batch processing, fault-tolerant Kubernetes pods, dev/test, CI runners, and stateless workloads. As of September 2026, n2-standard-4 in us-central1 lists $0.1942/hour on-demand but around $0.0227/hour Spot, roughly 88% off (GCP Spot VM pricing, September 2026).
Custom Machine Types on GCP let teams specify exact vCPU and memory for workloads with non-standard sizing. This is unique to GCP and can save 25-50% over pre-defined machine types for memory-heavy or compute-heavy workloads.
GCP Compute Engine on-demand vs Spot (us-central1, September 2026)
| Machine type | vCPU | RAM | On-demand | Spot (typical) | Spot discount |
|---|---|---|---|---|---|
| n2-standard-2 | 2 | 8 GB | $0.0971 | $0.0116 | 88% |
| n2-standard-4 | 4 | 16 GB | $0.1942 | $0.0227 | 88% |
| n2-highmem-8 | 8 | 64 GB | $0.6067 | $0.0728 | 88% |
| n2-highcpu-16 | 16 | 16 GB | $0.7769 | $0.1034 | 87% |
| c2-standard-30 | 30 | 120 GB | $2.955 | $0.591 | 80% |
Source: GCP Compute Engine pricing, Google Cloud Pricing API (September 2026).
BigQuery pricing: on-demand vs flat-rate vs editions
BigQuery has three pricing models: on-demand, flat-rate (slot-based), and the newer Enterprise Plus edition. On-demand charges $6.25 per TiB scanned (or $0.04 per TiB storage per month). Flat-rate charges a fixed monthly fee per BigQuery slot; as of September 2026, Standard edition is $10 per slot per month with a 100-slot minimum, Enterprise edition is $50 per slot per month, and Business Critical edition is $110 per slot per month (BigQuery pricing, Google Cloud, September 2026).
Flat-rate is cheaper than on-demand for predictable workloads above 100 TB scanned per month. BigQuery Editions (Standard, Enterprise, Enterprise Plus) layer on top of slots: Enterprise adds 99.99% SLA, customer-managed encryption keys (CMEK), and BigQuery Omni cross-cloud queries.
BigQuery pricing model comparison (September 2026)
| Model | Unit | Price | Best for |
|---|---|---|---|
| On-demand | TiB scanned | $6.25/TiB | Ad-hoc, spiky workloads |
| Standard edition (flat-rate) | Slot-month | $10/slot/mo (100 minimum) | Predictable analytical workloads |
| Enterprise edition | Slot-month | $50/slot/mo | High-throughput, CMEK, cross-cloud |
| Business Critical edition | Slot-month | $110/slot/mo | Regulated industries, HIPAA |
| Storage (active) | TiB-month | $0.04/TiB/mo | Frequently queried data |
| Storage (long-term) | TiB-month | $0.02/TiB/mo | Data not modified in 90 days |
Source: BigQuery pricing, Google Cloud BigQuery documentation (September 2026).
GKE cost optimization and Active Assist recommendations
GKE (Google Kubernetes Engine) cost optimization focuses on node pools, Spot VMs, Autopilot, and bin-packing. GKE Standard bills $0.10 per cluster per hour for the control plane; GKE Autopilot charges per pod vCPU and memory, removing the need to manage node pools. As of September 2026, GKE Autopilot vCPU is $0.0345/vCPU-hour and memory is $0.0037/GB-hour on top of the underlying Compute Engine costs (GKE pricing, Google Cloud, September 2026).
Active Assist provides GKE rightsizing recommendations via the GKE Rightsizing Recommender. The recommender analyzes pod CPU/memory utilization over 7 days and suggests smaller node pool sizes, Spot VM migration, or Autopilot adoption.
Recommended GCP Active Assist cost recommendations (September 2026)
| Recommender | Workload | Typical savings |
|---|---|---|
| Idle VM recommender | VMs <5% CPU over 14 days | 100% of VM cost |
| Idle persistent disk recommender | Disks detached >15 days | 100% of disk cost |
| Rightsizing recommender | Over-provisioned N2/N1 VMs | 20-50% per VM |
| Committed use discount recommender | Steady-state workloads | 25-52% on target service |
| BigQuery slot recommender | BigQuery usage patterns | 30-50% vs on-demand |
| GKE rightsizing recommender | Underutilized GKE node pools | 25-40% per node pool |
Source: GCP Active Assist documentation, Google Cloud Recommender API (September 2026).
Cloud Storage, networking, and committed-spend discounts
Cloud Storage tiering and lifecycle management deliver 30-80% storage cost savings. GCP Cloud Storage has Standard, Nearline, Coldline, and Archive classes. Standard lists $0.020/GB/mo; Nearline $0.010/GB/mo; Coldline $0.004/GB/mo; Archive $0.0012/GB/mo. Lifecycle policies move objects between classes automatically. Nearline has 30-day minimum and per-GB retrieval; Archive has 365-day minimum and graduated retrieval fees (Cloud Storage pricing, September 2026).
Custom private pricing for committed-spend customers with $1M+ per year is available through Google Cloud sales. Private pricing typically delivers 5-25% off list on Compute Engine, BigQuery, GKE, and Cloud Storage.
FAQs
Is GCP cheaper than AWS or Azure in 2026?
GCP's list price for N2, N2D, C2, and M2 instance families is typically 5-15% lower than AWS or Azure equivalents. BigQuery on-demand ($6.25/TiB) is competitive with Snowflake credit pricing but harder to predict. Realized cost depends on commitment discounts, regional pricing, and data transfer rates. Most FinOps teams run a TCO model per workload rather than rely on list price (Google Cloud Pricing vs AWS, Azure, September 2026).
Can I use CUDs with Spot VMs?
Yes. CUDs apply to the on-demand rate first, then Spot is applied on top. A Spot VM that would have cost $0.0971/hour on-demand in us-central1 is $0.0116/hour Spot, and any CUD applies as a further discount on the Spot rate. CUDs on Spot are typically rare because Spot discount exceeds CUD discount; CUDs are most useful for steady-state on-demand workloads (GCP CUD docs, September 2026).
What is the difference between GKE Standard and GKE Autopilot?
GKE Standard requires manual node pool management: teams choose instance types, configure autoscaling, and manage OS patches. The control plane costs $0.10/cluster/hour. GKE Autopilot is fully managed: Google configures nodes, applies security patches, and bills per-pod vCPU and memory. Autopilot vCPU is $0.0345/vCPU-hour; memory is $0.0037/GB-hour on top of Compute Engine. Autopilot is typically 20-40% more expensive per vCPU but eliminates node pool management overhead (GKE pricing, GKE Autopilot docs, September 2026).
How do I set up BigQuery flat-rate?
In the GCP console, navigate to BigQuery, then 'Reservations'. Click 'Create Reservation', choose a region, edition (Standard, Enterprise, Business Critical), and slot count. Auto-scaling reservations automatically scale slots based on query load. Baseline slots are the minimum committed spend; autoscaling slots are pay-per-use. Auto-scaling slots bill $0.04 per slot-hour for Standard edition (BigQuery Reservations docs, September 2026).
References and next steps
GCP cost optimization in 2026 rewards teams that layer CUDs, Spot VMs, and BigQuery slot commitments. Begin by enabling Active Assist and reviewing idle VM, idle disk, and rightsizing recommendations. Commit 60-70% of trailing compute spend to Resource-based CUDs, then layer Spot VMs on batch and CI workloads. For BigQuery, switch from on-demand to flat-rate above 100 TB scanned per month, choosing Standard or Enterprise edition based on compliance needs. For annual commits above $1M, negotiate private pricing through Google Cloud sales. Track realized savings monthly with Billing Reports + BigQuery + Looker Studio.






