Google Cloud’s compute model differs from its competitors in two ways that matter for cost: you can define custom machine shapes, and volume discounts apply automatically without any commitment.
The machine families
| Family | Built for |
|---|---|
| E2 | Cost-optimised general purpose. Good default for web and application tiers. |
| N2 / N2D | Balanced general purpose with more consistent performance. N2D is AMD. |
| C3 / C2 | Compute optimised for CPU-bound work. |
| M | Memory optimised for large in-memory databases. |
| A / G | Accelerator families for GPU training and inference. |
| T2A / T2D | Arm and AMD options that often win on price for performance. |
Custom machine types
This is the genuinely distinctive feature. Rather than picking from fixed vCPU and memory pairings, you specify each independently within the family’s allowed ratios.
It matters when your workload is unbalanced. A cache or search node that needs a lot of memory and little CPU would, on a fixed-shape platform, force you up a tier and waste the cores. A custom shape bills you for what you asked for.
If your requirements are balanced, custom types buy nothing, and a standard shape is simpler to reason about.
Sustained use discounts
Run an instance for a significant portion of the month and Google applies a discount automatically. There is no reservation to buy and nothing to forecast. Committed use discounts exist separately for deeper savings on a genuinely steady baseline.
This changes cost comparisons against AWS and Azure, where equivalent savings usually require commitment planning. Compare the platforms directly in AWS vs Google Cloud and Google Cloud vs Azure.
Preemptible and Spot VMs
Spot VMs are heavily discounted instances that Google can reclaim with short notice. They suit batch processing, rendering, CI and anything checkpointed. They do not suit stateful services that cannot tolerate sudden termination.
Quotas still apply
As on any platform, provisioning is capped by quota per region, and quotas are separate from any credit balance on the account. GPU quotas in particular start at zero on many new projects and must be requested.
Our Google Cloud listings cover seven configurations, including credit-backed tiers from $300 up to $25,000. Which tier suits you is a burn rate question, covered in Google Cloud credits explained.
Before you deploy
Change the delivered credentials, enable multi-factor authentication, create scoped service accounts rather than working as project owner, and set a budget alert. The full sequence is in the day one checklist.
Common questions
Can I resize a running instance?
Stop it, change the machine type, start it again. Custom shapes can be adjusted the same way.
Is E2 or N2 better?
E2 is cheaper and fine for most general workloads. N2 gives more predictable performance under sustained load.
Do sustained use discounts need enabling?
No. They are applied automatically based on usage.