
Private NAT is the mode of Cloud NAT that translates the source addresses of VM traffic headed to internal destinations, rather than to the internet. It takes the translated addresses from a subnet whose purpose is PRIVATE_NAT. It rewrites only the source address, so the destination stays the same, and it involves no VMs to deploy, patch or scale. Both of its forms address networks whose internal IP ranges overlap. Private NAT is regional, and a gateway only translates traffic from VMs in its own region. For the Professional Cloud Network Engineer exam, the distinctions to get right are which form of Private NAT applies to a given traffic path and how Hybrid NAT is laid out across regions.
Because Private NAT changes only the source, the VMs keep connecting to the real addresses of the servers they need. The far side sees only the NAT addresses, so it can route replies back to them, as long as the NAT range is one it can route to Google Cloud.
Source-only translation is also what sets Private NAT apart from other networking tools. Route advertisements decide which prefixes a peer learns, but they never rewrite packet addresses, so on their own they can't resolve an overlap. A proxy places its own virtual IP between client and server, so clients stop connecting to the real server addresses. The subnet types differ too. Private NAT draws on a subnet with the PRIVATE_NAT purpose, while the REGIONAL_MANAGED_PROXY purpose belongs to proxy-based load balancers, not NAT.
Private NAT for Network Connectivity Center spokes translates traffic that flows through a Network Connectivity Center hub. That covers traffic between VPC spokes and traffic between VPC spokes and hybrid spokes. It lets many VPC networks that reuse the same internal ranges reach each other's unique ranges. The reused ranges are excluded from export, and traffic leaving the overlapping subnets is translated.
Hybrid NAT is the type of Private NAT built for traffic that leaves a VPC network toward an on-premises network over Cloud VPN or Cloud Interconnect, following dynamic routes that a Cloud Router has learned. It fits when VPC subnets overlap ranges that on-premises already uses, but the on-premises servers the VMs need sit in non-overlapping ranges. On-premises sees only the NAT addresses, so it can route replies back, while the VMs still connect to the real server addresses that arrived as BGP routes.
The two forms are not interchangeable. Private NAT for Network Connectivity Center spokes only translates traffic that passes through the hub. If the VPN tunnels to on-premises are not hybrid spokes on the hub, traffic to on-premises never passes through it, and that gateway would not translate it. Hybrid NAT on a Cloud NAT gateway in the VPC network itself handles that case.
Every Hybrid NAT design needs a subnet whose purpose is PRIVATE_NAT to supply the translated addresses, and the translation applies to traffic that follows dynamic routes learned from on-premises over Cloud VPN or Cloud Interconnect. Beyond that, the hybrid connections can be laid out in more than one way.
One setup configures Hybrid NAT on a Cloud NAT gateway in the VPC network itself, where the tunnels or VLAN attachments and their Cloud Router live. This works even when those hybrid connections are not attached to a Network Connectivity Center hub. For hybrid traffic over Cloud Interconnect, another approach adds the VLAN attachments to Network Connectivity Center as hybrid spokes, and Cloud NAT then translates traffic that follows the dynamic routes learned from on-premises. Either can fit, depending on how the rest of the network is built. What matters is that the translating gateway actually sees the traffic, so whether the hybrid connections are attached to a hub is worth checking in any design that mixes Network Connectivity Center and on-premises connectivity.
The layout of Hybrid NAT follows from how the translation works once a network spans more than one region or joins a Network Connectivity Center hub.
A Hybrid NAT gateway matches destinations against the local dynamic routes that the Cloud Router in its own region learned, and it does not use routes learned by a Cloud Router in a different region. Combined with the regional scope of Private NAT, that means each region with VMs that start hybrid connections needs its own gateway and NAT subnet.
The NAT subnets are the addresses on-premises replies to, so they belong in a range that on-premises routes back to Google Cloud. They can't be carved from the on-premises network's own range, because replies to those addresses would stay on-premises.
When the VPC network is a Network Connectivity Center spoke, its export filters decide which ranges the hub learns. An include export range filter that lists only the NAT subnets advertises them and keeps non-routable application subnets out, so the ranges that replies depend on are the ones that get shared.
For Professional Cloud Network Engineer scenarios, a useful way to approach these designs is to trace where the traffic goes. Traffic between VPC networks that reuse ranges, flowing through a Network Connectivity Center hub, points to Private NAT for Network Connectivity Center spokes. Traffic leaving one VPC network for on-premises over VPN or Interconnect, following routes learned by Cloud Router, points to Hybrid NAT, with a gateway in every region that has workloads and NAT subnets that on-premises can route back to. In both forms, Private NAT rewrites the source and leaves the destination alone, which is what separates it from changes to route advertisements and from proxies.
Our Professional Cloud Network Engineer practice exams cover Private NAT and Hybrid NAT alongside Network Connectivity Center and Public NAT, with detailed explanations that drill these distinctions.