Updated 26 Feb 2026 • 5 mins read
Google Cloud Pricing Explained (2026)
Google Cloud pricing is per-second compute, per-GB storage, per-TiB BigQuery, and per-GB egress, shaped by three discounts that stack: automatic sustained-use, committed-use, and Spot. This 2026 breakdown gives the main rates for Compute Engine, GKE, Cloud Storage, BigQuery, Cloud SQL, Cloud Run, and networking, the discount rules, the free tier, and worked examples.
Google Cloud pricing has the same shape as every hyperscaler's, per-second compute, per-gigabyte storage, per-gigabyte egress, and three features that make it behave differently from AWS and Azure in practice. One discount applies with no action at all, to some machine types and not others. The committed-use discount changed model in January 2026. And egress to the internet costs more than on either competitor. Understanding those three is the difference between an estimate that holds and one the first invoice embarrasses.
This guide explains how Google Cloud pricing works across the services that make up most bills, Compute Engine, GKE, Cloud Storage, BigQuery, Cloud SQL, Cloud Run, and networking, with 2026 rates for US regions, the three discounts and how they stack, the free tier, and worked examples. For the Kubernetes half, our GKE cost management best practices go deeper; for building an estimate, our Google Cloud cost calculator guide.
The short answer: Google Cloud bills per second for compute, per GB-month for storage, per TiB scanned or per slot for BigQuery, and per GB for network egress, by region. In us-central1 an e2-standard-2 VM costs about $49 a month on demand, balanced persistent disk $0.10 per GB-month, Standard Cloud Storage about $0.020 per GB-month, BigQuery on-demand $6.25 per TiB scanned with 1 TiB free, GKE $0.10 per cluster-hour, and internet egress $0.12 per GB for the first terabyte on the premium tier. Three discounts stack: sustained-use discounts apply automatically, up to about 30 percent, to N1, N2, and memory-optimized types running most of the month but not to E2, N4, C3, or C4; committed-use discounts run about 37 and 57 percent for resource-based one- and three-year commitments and about 28 and 46 percent for the spend-based Flexible model that became the default in January 2026; Spot VMs are 60 to 91 percent off. An Always Free tier covers one e2-micro, 5 GB of storage, 1 TiB of BigQuery queries, and more, permanently.
Key Takeaways
- Compute bills per second with a one-minute minimum; the rate depends on machine family, and family decides sustained-use eligibility.
- Sustained-use discounts are automatic and up to 30 percent, but only on N1, N2, N2D, C2, and memory-optimized types; E2, N4, C3, and C4 pay list.
- Committed-use discounts moved to the spend-based Flexible model in January 2026; resource-based CUDs still discount deeper for fixed shapes.
- BigQuery on-demand bills bytes scanned, not rows returned, at $6.25 per TiB; egress is $0.12 per GB, the highest of the big three.
- The project is the billing unit and the best allocation structure in the cloud, and the place waste accumulates when projects outlive their owners.
Compute Engine
Virtual machines bill per second with a one-minute minimum at a rate set by machine family and size, plus persistent disk, plus network. us-central1 list rates, October 2026, approximate.
| Machine type | vCPU / memory | Hourly | Monthly (730 h) | Sustained-use eligible | Notes |
|---|---|---|---|---|---|
| e2-micro | Shared / 1 GB | About $0.008 | About $6 (one free per month) | No | Always Free tier |
| e2-standard-2 | 2 / 8 GB | About $0.067 | About $49 | No | Cheapest general-purpose; no SUD |
| n2-standard-2 | 2 / 8 GB | About $0.097 | About $71; about $57 after SUD | Yes (about 20%) | Net rate close to E2 for steady use |
| n2-standard-8 | 8 / 32 GB | About $0.39 | About $284; about $227 after SUD | Yes | Common application tier |
| c3-standard-4 | 4 / 16 GB | About $0.20 | About $146 | No | Latest-generation compute |
| Custom machine types | Any shape | Priced per vCPU and GB | Varies | Family-dependent | Match the shape; a GCP advantage |
| Spot VMs (any type) | Any | 60 to 91% off on-demand | Varies | n/a | Interruptible with 30-second notice |
Persistent disk bills per GB-month provisioned: standard about $0.04, balanced about $0.10, SSD about $0.17, plus snapshots at about $0.026 per GB-month. Local SSD bills per GB-month attached. GPUs bill per GPU-hour on top of the VM, from about $0.35 for a T4 to several dollars for H100-class accelerators.
The Three Discounts, and How They Stack
| Discount | How it works | Rate | Applies to | Action required |
|---|---|---|---|---|
| Sustained-use (SUD) | Automatic reduction for VMs running more than 25 percent of the month, scaling to full-month usage | Up to about 30 percent (N1, memory-optimized); about 20 percent (N2, N2D, C2) | Eligible machine types only; not E2, N4, C3, C4 | None |
| Resource-based committed use (CUD) | Commit to vCPUs and memory in a region for 1 or 3 years | About 37 percent (1 year), 57 percent (3 years); up to 70 percent memory-optimized | One machine family and region | Purchase |
| Flexible committed use (CUD) | Commit to an hourly spend for 1 or 3 years | About 28 percent (1 year), 46 percent (3 years) | Across families, regions, GKE, Cloud Run; default model since January 2026 | Purchase |
| Spot VMs | Spare capacity, reclaimable with 30 seconds' notice | 60 to 91 percent off | Fault-tolerant workloads | Select Spot provisioning |
They stack in a specific way: SUDs reduce the on-demand rate first, so a committed-use discount on an eligible type adds its percentage to a rate already 20 to 30 percent lower, and the incremental saving from committing is smaller than the headline. For E2 and the newer families, which earn no SUD, the CUD is the whole discount. Spot replaces both for interruptible work. The sizing discipline for commitments is the same as on every cloud and is set out in our GCP cost optimization guide: commit to the trough of sustained usage, after right-sizing, with Flexible CUDs where the shape may change.
Google Kubernetes Engine
GKE charges $0.10 per cluster-hour, about $73 a month, for the control plane, with a free tier credit that covers one zonal or Autopilot cluster per billing account. Standard mode bills the node VMs at Compute Engine rates, so the node bill, its SUDs, CUDs, and Spot, is the cluster bill; Autopilot bills per pod's requested vCPU, memory, and ephemeral storage per second, with no nodes to manage, at a premium over well-packed Standard nodes and a saving over poorly packed ones. Pod requests set the cost in both modes, which is the first lever in our GKE guide.
Cloud Storage
| Class | Per GB-month (US) | Minimum duration | Retrieval per GB | Best for |
|---|---|---|---|---|
| Standard | About $0.020 | None | None | Active data |
| Nearline | About $0.010 | 30 days | $0.01 | Read about monthly |
| Coldline | About $0.004 | 90 days | $0.02 | Read about quarterly |
| Archive | About $0.0012 | 365 days | $0.05 | Long-term retention; still millisecond access |
Operations bill per 10,000, with class A (writes, lists) costing more than class B (reads), and cooler classes charging more per operation. Autoclass moves objects between classes automatically for a per-object fee; lifecycle rules do it by age for free. Google's Archive is unusual in being instantly readable, where S3 Deep Archive and Azure Archive take hours. The three-cloud comparison is in our cloud storage pricing guide.
BigQuery
Two pricing models. On-demand bills $6.25 per TiB of data scanned by queries, first TiB a month free, and bills bytes read rather than rows returned, so a query over an unpartitioned 10 TB table costs about $62 however small its result. Capacity pricing (editions) bills reserved slots per hour, Standard, Enterprise, or Enterprise Plus, with autoscaling, and is predictable; it typically beats on-demand above roughly 100 TB scanned a month. Storage bills about $0.02 per GB-month active and about $0.01 long-term (90 days unmodified). Partitioning and clustering reduce bytes scanned directly, and per-user byte limits stop a runaway dashboard.
Cloud SQL, Cloud Run, and the Rest
- Cloud SQL bills the instance per hour by vCPU and memory, storage per GB-month, and backups, with high availability roughly doubling the instance and storage lines; committed-use discounts apply.
- Cloud Run bills per vCPU-second and GiB-second while a container handles requests, plus per request, with 2 million requests a month free and scale-to-zero; Flexible CUDs apply.
- Cloud Functions bills invocations, compute time, and memory like Cloud Run, with a free allowance.
- Vertex AI and Gemini bill per token or per hour of accelerator, on their own meters, covered in our Gemini API pricing guide.
Networking
Ingress is free. Egress to the internet on the premium tier, the default, costs $0.12 per GB for the first terabyte a month, $0.11 for the next 9 TB, and $0.08 above 10 TB, higher than AWS's $0.09 and Azure's $0.087 at the first tier. The standard tier, which routes over public internet rather than Google's backbone, is cheaper for traffic that can tolerate it. Egress to Google services in the same region is free; traffic between regions bills per GB; traffic between zones in a region bills at a low per-GB rate that adds up on chatty multi-zone deployments. Cloud CDN in front of public content cuts the egress line substantially, and Private Google Access and Private Service Connect keep traffic to Google services off the metered path.
The Free Tier
Google Cloud's Always Free tier is permanent and per billing account: one e2-micro VM in eligible US regions, 30 GB of standard persistent disk, 5 GB of Standard Cloud Storage, 1 TiB of BigQuery queries and 10 GB of BigQuery storage, 2 million Cloud Run requests, and allowances on Cloud Functions, Pub/Sub, and others. New accounts also receive a time-limited trial credit. The permanent allowances cover real small production use; the trial credit covers the first weeks of a larger estate.
Worked Examples
us-central1, monthly, on demand unless stated.
| Workload | Components | Monthly |
|---|---|---|
| Small web application | 2 x e2-standard-2 ($98), 200 GB balanced disk ($20), Cloud SQL 2 vCPU HA (about $250), 500 GB egress ($60) | About $430 |
| Same application on N2 with sustained use | 2 x n2-standard-2 after SUD ($114), same disk, SQL, and egress | About $445; cheaper than E2 only with a CUD |
| Analytics platform | GKE cluster ($73) with 12 n2-standard-8 nodes after SUD ($2,720), BigQuery scanning 40 TB ($244), 50 TB Standard storage ($1,000), 2 TB egress ($230) | About $4,270 |
| Same platform with 3-year Flexible CUD on the nodes and lifecycle rules on storage | Nodes about $1,470; storage about $600 with 60 percent in Nearline and Coldline | About $2,620 |
How Opslyft Helps
Google Cloud's project model is both the cleanest allocation structure in the cloud and the reason spend accumulates in projects nobody owns, and the discounts above only save money if someone measures which machine types earn them and what the commitments actually cover. Opslyft reads the GCP billing export, shows spend by project, folder, team, and product beside AWS, Azure, Kubernetes, and Snowflake, and surfaces the idle resources and uncovered baseline the rate card cannot; what to do about them is in our GCP cost optimization guide.
The Bottom Line
Google Cloud pricing rewards workloads that run steadily on the right machine family, commit their baseline, tier their storage, partition their tables, and keep their traffic off the internet, and charges list price for everything else. Learn the three discounts and which types earn them, decide between resource-based and Flexible commitments on shape stability, model BigQuery from bytes scanned, put a CDN in front of public content, and give every project an owner. The levers that follow are in our GCP cost optimization guide; the comparison with the other two clouds is in our AWS versus Azure versus GCP guide; and an estimate of your own configuration starts with the Google Cloud pricing calculator and our guide to using it well.
FAQs
Google Cloud bills per second for compute, per GB-month for storage, per TiB scanned or per slot for BigQuery, and per GB for network egress, with rates that vary by region. Its distinguishing feature is three discounts that stack: sustained-use discounts applied automatically to eligible machine types that run most of the month, committed-use discounts for one- or three-year commitments, and Spot VMs at 60 to 91 percent off for interruptible work. There is no upfront fee and an Always Free tier with permanent allowances.
In us-central1, an e2-standard-2 (2 vCPU, 8 GB) costs about $0.067 an hour, about $49 a month on demand; an n2-standard-2 about $0.097, about $71 before sustained-use discounts bring it near $50; a c3-standard-4 about $0.20. Persistent disk adds $0.04 per GB-month for standard, $0.10 for balanced, and $0.17 for SSD. Spot VMs cut the compute rate by 60 to 91 percent.
Automatic discounts Google applies to eligible Compute Engine machine types that run for more than 25 percent of a month, rising to about 30 percent off for N1 and memory-optimized types and about 20 percent for N2, N2D, and C2 at full-month usage, with no commitment and nothing to configure. E2, N4, C3, and C4 machine types are not eligible, which is the detail most estimates miss.
Discounts for committing to a resource or a spend level for one or three years. Resource-based CUDs commit to vCPUs and memory in a region for about 37 percent (one year) or 57 percent (three years) off, up to 70 percent on memory-optimized types; spend-based Flexible CUDs commit to an hourly spend for about 28 or 46 percent off and apply across machine families, regions, GKE, and Cloud Run. Google moved customers to the Flexible model in January 2026.
On-demand BigQuery bills $6.25 per TiB of data scanned by queries, with the first TiB each month free, plus storage at about $0.02 per GB-month active and $0.01 long-term. Capacity pricing (editions) bills reserved slots by the hour instead and typically becomes cheaper above roughly 100 TB scanned a month. Partitioning and clustering tables reduce bytes scanned and therefore on-demand cost directly.
Internet egress on the premium network tier costs $0.12 per GB for the first terabyte a month, stepping down to $0.08 above 10 TB, which is the highest of the three major clouds. Egress to Google services in the same region is free, inter-region transfer bills per GB, and the standard network tier offers lower rates for traffic that does not need Google's global backbone. Cloud CDN in front of public content reduces egress substantially.
An Always Free tier with permanent monthly allowances: one e2-micro VM in eligible US regions, 30 GB of standard persistent disk, 5 GB of standard Cloud Storage, 1 TiB of BigQuery queries and 10 GB of storage, 2 million Cloud Run requests, and allowances across other services. New accounts also receive a time-limited trial credit. The allowances are per billing account and do not expire.