Google Compute Engine VM Cost: CPU, Disk, and Network

Start with the machine family, region, vCPU count, memory, and monthly running hours. Compute Engine bills vCPU and memory separately for most machine types, even when you select a predefined machine. Resources have a one-minute minimum and are then billed by the second, so a stopped VM avoids further compute charges but a continuously running service should be planned for the full month.

Do not size only for average CPU. Include memory pressure, short traffic peaks, operating-system overhead, and the number of replicas needed during maintenance or failure. Compare a right-sized general-purpose machine with a smaller custom machine before adding permanent capacity, and confirm current regional rates in the official calculator.

A VM estimate is incomplete without its attached resources. Add the provisioned capacity of boot and data disks, snapshot storage and operations, external IP addresses, licenses, GPUs, and any load balancing or monitoring used by the workload. Persistent Disk charges are based on provisioned space, not the bytes currently stored, and disks can keep accruing charges after a VM is stopped or deleted if the disks remain.

Model network paths separately: same-zone internal traffic, cross-zone or cross-region traffic, and internet data transfer can have different rates. Attribute outbound traffic to the sending VM, then multiply monthly GiB by the current price for the source and destination. Avoid embedding a single global egress rate in a long-term forecast.

Sustained use discounts can apply automatically to eligible resources used for more than 25% of a billing month, while committed use discounts exchange a one-year or three-year commitment for lower eligible rates. Spot VMs can be much cheaper but may be preempted, so use them only for fault-tolerant work that can resume safely. These discount types are not stacked on the same usage.

Build the baseline with on-demand rates first, then apply a discount only to the stable portion you expect to keep. Keep burst capacity on demand or Spot, monitor idle CPU and unattached disks, and compare the resulting VM total with managed options that may remove patching and availability work from your operating burden.

Compare the VM baseline with Cloud Run and App Engine runtime costs.