Cloud and platforms
AWS NAT Gateway vs Azure NAT Gateway vs Cloud NAT: What Each Bills
In short
All three bill an hourly gateway charge plus data processed, but they count the hour differently. AWS bills each gateway-hour, and a regional gateway bills per Availability Zone. Azure bills from creation, whatever is attached. Google bills per VM using the gateway, capped at 32. Public IP addresses and ordinary data transfer are billed on top.
What is each hourly charge actually counting?
AWS bills each hour that a NAT gateway is provisioned and available, and a partial hour is billed as a full hour. A regional NAT gateway is billed per Availability Zone: one running across three zones for an hour is three NAT Gateway-hours. When it drops a zone, billing for that zone stops.
Azure starts billing when the resource is created. The pricing FAQ says the hourly rate applies whether or not the gateway has subnets or public IPs attached, and a partial hour is billed as a full hour. Standard and StandardV2 cost the same.
Google does not charge per gateway. The hourly price for Public NAT depends on how many VMs use the gateway, counting each VM that gets a NAT assignment, and it is capped at 32 VMs. Beyond that, the gateway is charged at the maximum hourly rate.
This is the largest structural difference. On AWS and Azure the gateway count drives the hourly line. On Google the VM count does, up to the cap.
| Meter | AWS NAT Gateway | Azure NAT Gateway | Google Cloud NAT |
|---|---|---|---|
| Hourly unit | Per gateway-hour; regional mode per Availability Zone | Per gateway resource-hour, from creation | Per VM using the gateway, capped at 32 VMs |
| Data processed | Per GB, any source or destination | Per GB, outbound and return | Per GiB, inbound and outbound |
| Public IPs | Public IPv4 billed per address-hour | Public IP is its own billed resource | Hourly price per external IP the gateway uses |
| Billed separately | Standard data transfer | Bandwidth | Data transfer out |
What does data processed include?
AWS says processing is charged for each gigabyte through the gateway regardless of the traffic source or destination. Azure says all outbound data and the return data, meaning responses to outbound connections, is charged per GB. Google lists a per-GiB price for inbound and outbound data transfer processed. On paper, all three count return traffic.
None of the three replaces ordinary network charges. AWS adds standard data transfer, Azure adds bandwidth pricing and Google adds data transfer out. The AWS example sends 1 GB from an EC2 instance to S3 in the same region and zone: the NAT processing charge applies and the data transfer charge is zero. Across Availability Zones, AWS says transfer charges between the gateway and the instance apply.
How are the public IP addresses billed?
On AWS a public NAT gateway needs an Elastic IP at creation, and public IPv4 addresses are billed hourly in one-second increments with a 60-second minimum. The page lists separate in-use and idle rates and says the price is the same for both. A private NAT gateway cannot have an Elastic IP.
On Azure the gateway uses public IP addresses or a prefix, up to 16 IPv4 addresses (StandardV2 also takes up to 16 IPv6), each giving 64,512 SNAT ports. The IPs are separate resources on their own pricing page. A Standard static IPv4 address is billed per hour, and a public IP prefix is billed per address per hour from the moment the prefix is created.
On Google, Public NAT bills an hourly price for each external IP the gateway uses, and Private NAT uses private addresses, which the pricing example calls free. With automatic allocation, the gateway adds IPs based on the network tier, the number of VMs and the ports reserved for each VM. The default minimum is 64 ports per VM for static port allocation and 32 for dynamic. Each IP offers 64,512 TCP and 64,512 UDP source ports.
I read that as: on Google, VM count and the ports setting can move the IP line as well as the hourly line. On AWS and Azure you choose the IP count yourself.
Checked against: source 1, source 4, source 6, source 7, source 8, source 9, source 10, source 11
Why can the same traffic cost differently?
Three documented mechanisms explain most gaps. First, the zone. AWS says a regional gateway can take up to 60 minutes to expand to a new zone, and until then traffic from that resource is processed across zones. A zonal gateway in a different zone from its clients also adds cross-zone transfer.
Second, which path the traffic takes. AWS says gateway VPC endpoints have no hourly or data processing charges, so traffic routed through one does not reach the NAT gateway meters. Google routes traffic to Google APIs through Private Google Access even when the VM uses Public NAT. Azure goes the other way: a NAT gateway takes precedence over load balancer outbound rules, instance-level public IPs and Azure Firewall for new connections, so attaching one moves that traffic onto its meter.
Third, the hour. A Google gateway serving 40 VMs is billed at the 32-VM maximum rate, while the AWS and Azure pages I read state the hourly charge per gateway and say nothing about how many VMs sit behind it.
Checked against: source 1, source 2, source 3, source 4, source 6, source 9, source 10
What do the docs say about reducing the bill?
The vendors give different amounts of guidance, so this list marks where a page is silent.
- AWS: keep resources in the same Availability Zone as the NAT gateway, or create one gateway per zone that has resources. Use interface or gateway endpoints for AWS services that support them.
- AWS: delete the gateway when you no longer want the charge.
- Azure: the pages I read give no reduction tips. They do say the hourly rate runs while the gateway has nothing attached, so an unused gateway keeps billing.
- Google: keep Google API traffic on Private Google Access, which the overview says it uses even with Public NAT.
What else do people ask?
Does a private NAT gateway cost less than a public one?
The pages I read do not compare them. AWS cannot attach an Elastic IP to a private NAT gateway. Google says Private NAT uses private IP addresses, and its pricing example treats them as free. Both still bill hours and data processed.
Is Cloud NAT billed per gateway like the AWS and Azure gateways?
No. Google bases the hourly price on the number of VMs using the gateway, up to 32, plus the data processed and the external IPs it uses.
Sources
- Amazon VPC pricing, NAT Gateway section, fetched 2026-11-19
- Amazon VPC User Guide, Pricing for NAT gateways, fetched 2026-11-19
- Amazon VPC User Guide, Regional NAT gateways, fetched 2026-11-19
- Amazon VPC User Guide, NAT gateways, fetched 2026-11-19
- Azure NAT Gateway pricing, fetched 2026-11-19
- Microsoft Learn, What is Azure NAT Gateway, fetched 2026-11-19
- Microsoft Learn, NAT gateway resource, fetched 2026-11-19
- Azure IP addresses pricing, fetched 2026-11-19
- Google Cloud NAT pricing, fetched 2026-11-19
- Google Cloud NAT overview, fetched 2026-11-19
- Google Cloud NAT addresses and ports, fetched 2026-11-19
How was this page made?
I drafted this page with AI assistance from the official documents listed under Sources. This page compares vendor documents that I fetched on 19 Nov 2026. I did not deploy or run a NAT gateway on any of the three clouds for it. It does not cover prices, logging charges, egress rates by destination, or how to size ports and IP counts. Found a mistake? Tell me.