Practical guide
Reduce surprises in Cloud NAT cost forecasts
Identify the Cloud NAT usage signals that change a bill: assigned VMs, processed GiB, IPs, egress, and logging.
Updated · Sources checked
The variable is often traffic, not gateway count
Cloud NAT estimates fail when they model a flat gateway but omit processed bytes and external IP time. Google lists those charges for Public NAT and states the VM-based gateway rate is capped at 32 assigned instances. Track all three quantities and the separate outbound transfer destination.
For hypothetical math, a workload grows from 100 to 900 processed GiB while VM assignments and IP-hours stay unchanged. The gateway and IP subtotals remain flat, while the processing component becomes nine times larger. Conversely, doubling assigned VMs below the cap changes the gateway component even if traffic stays fixed. This identifies the usage metric to investigate.
Separate NAT from application egress. A container-image pull, OS package update, analytics export, and customer response can all appear as traffic but have different reduction options. Co-locate services where appropriate, cache artifacts, and inspect logs before changing network architecture. Do not remove required NAT or logging just to improve a spreadsheet.
Use the Cloud NAT Cost Calculator to model current counts and the network egress calculator for destination pricing.
FAQ
Is a gateway charged once per project?
Pricing depends on the configured NAT type and usage; use the official table for the deployed configuration.
Are external IPs free with NAT?
Google lists an hourly external-IP charge for Public NAT.