
A VPC network in Google Cloud is a global resource, while each of its subnets is regional. One network can hold subnets in many regions at the same time, and instances in all of those subnets reach each other over their internal IP addresses without any extra setup. No gateway, tunnel or proxy sits in the path. The traffic stays on Google's backbone, and the only charge is the normal inter-region data transfer rate. This global scope is the starting point for many design decisions on the Professional Cloud Network Engineer exam.
The network is the unit that holds routing and firewall configuration, and it isn't tied to any one region. Subnets are where geography comes in. Each subnet belongs to one region and provides the IP range for the resources placed there. Extending a design to a new location means adding a subnet in that region to the same network, not building a new network.
Because the subnets share one network, they can already reach each other. Two instances in different regions can communicate over their private RFC 1918 addresses as soon as both are running in subnets of the same network. Nothing needs to be deployed to connect the two regions, and the traffic never leaves Google's backbone.
When workloads in several regions need to talk to each other, a single VPC network with a regional subnet in each location is the simplest design and the one Google recommends. The team manages one network, one set of firewall rules and one routing table, and there is only one set of networking resources to maintain.
Splitting workloads into separate VPC networks means something has to join them again, and every way of joining them carries its own overhead. HA VPN brings hourly gateway and tunnel charges, limits throughput to what the tunnels can carry and adds resources to operate. VPC Network Peering avoids gateways, but it is still configuration that a single network doesn't need, and peering is not transitive, so a network reaches only the networks it is directly peered with.
Other paths are built for different purposes. External IP addresses carry their own charges and expose workloads publicly, which gives up the private path a VPC network provides. A global external Application Load Balancer is designed to serve client traffic, so it is not a path for private service-to-service traffic, and it would add a proxy hop and cost.
Separate networks have their place when there is a requirement to keep things apart. When the only goal is regional reach for workloads that belong together, the extra networks and connections solve a problem that the global VPC network has already solved.
VPC networks are not limited to RFC 1918 space. Subnets can use other supported ranges, such as the 100.64.0.0/10 shared address space, which gives an address plan room when private ranges are already in use elsewhere in a connected environment. Those subnets behave like any others in the network, so subnets in different regions still route to each other without extra setup.
For the Professional Cloud Network Engineer exam, we would treat a single global VPC network as the default design whenever workloads in several regions belong together, and look for a specific requirement before adding a second network and a way to connect it. The facts that support that default are short: the network is global, subnets are regional, instances in different regions talk over internal addresses with no gateways or tunnels, and the only extra cost is inter-region data transfer.
Our Professional Cloud Network Engineer practice exams cover VPC network design alongside subnet addressing and VPC Network Peering, with detailed explanations that drill these distinctions.