VPC Network Peering: How It Works, Its Limits and Why It Isn't Transitive

GCP Study Hub
November 8, 2026

VPC Network Peering connects two VPC networks so their instances communicate over internal IP addresses. It works across projects and across organizations, and each side creates its own peering configuration, so each team or organization keeps full control of its own network. Traffic stays on Google's network and never touches the public internet. Peering also has firm rules about address ranges, firewall rules, which routes are exchanged and how far those routes travel, and those rules shape many connectivity scenarios on the Professional Cloud Network Engineer exam.

Why peering is fast and inexpensive

There's no gateway or tunnel in the path of peered traffic. Peering gives the same latency and throughput as traffic within one VPC network, lower latency and higher throughput than a VPN, and no gateway or tunnel charges. That makes it the cheapest way to link two networks.

The comparison with HA VPN is the one that comes up most. A VPN between two VPC networks also keeps traffic private, but it brings hourly charges for each tunnel plus egress, and every tunnel has a bandwidth cap, none of which applies to peering.

Rules every peering follows

  • No overlapping ranges. The subnet ranges in the two networks must not overlap.
  • Routes, not open ports. Peering exchanges subnet routes, but it doesn't open any firewall ports. Each side still needs ingress allow rules for the other network's ranges. Egress is usually already allowed by the implied allow egress rule.
  • Only the two networks you connect. Peering exists only between the two networks in the peering. A third network stays isolated unless you peer it on purpose.
  • Peered networks exchange traffic. Joining two networks with peering breaks any requirement that the two must never exchange traffic.

Static routes aren't needed to make peering work, because peering already shares subnet routes between the two networks.

Firewall rules across a peering

Across a peering connection, the receiving network can't evaluate network tags or service accounts that belong to the other network. A rule that names a peer's network tag or service account as its source doesn't match peered traffic. Ingress rules for traffic from a peered network therefore use the peer network's IP ranges as the source.

Which routes peering exchanges

Subnet routes are exchanged automatically, but other routes need extra settings:

  • Custom routes, which include static routes and dynamic routes such as those learned over HA VPN, aren't carried to the peer unless custom route export and import are set up.
  • Subnet routes that use privately used public IPv4 (PUPI) addresses are exported by default but aren't imported unless the importing side turns that on. They are subnet routes, so they have their own export and import settings, separate from custom route exchange.

Peering is not transitive

VPC Network Peering is not transitive. If network A peers with network B, and network C also peers with network B, A and C each learn B's subnet routes, but they never learn each other's routes through B. To keep using peering, A and C have to be peered directly, so all three networks form a full mesh.

Custom route export and import don't change this, because they share routes across a single peering and never pass routes learned from one peer on to a different peer.

Non-transitivity affects several designs:

  • Hub-and-spoke peering. Spoke networks can't reach each other through a hub they are all peered with. The same property is useful when spokes should stay isolated from one another.
  • Two peering hops. An older GKE private cluster's control plane lives in a Google-managed VPC network that is peered to the cluster's network. A third network peered to the cluster's network has no route across both hops to the control plane.
  • Routes from a hybrid connection. Routes that one network learns over its own HA VPN connection don't reach a peer unless custom route export and import are configured on that peering.

Peering limits

A VPC network can have at most 25 peering connections by default. A hub network that needs more spokes than that can't be built with peering unless the limit is raised. Peering also won't connect networks whose subnet ranges overlap, and the per-network limit is well below the number of networks in a large estate. Designs at that scale need a different approach, such as Network Connectivity Center, which provides a managed hub that can connect a large number of VPC spokes.

When peering is the right choice

For the Professional Cloud Network Engineer exam, VPC Network Peering is the natural fit when two networks with non-overlapping ranges need private, internal-address connectivity at the lowest cost and with no extra hops or bandwidth caps, and each side wants to keep administering its own network. It stops being the right fit when ranges overlap, when more than the default number of connections is needed, when traffic must pass between networks that aren't directly peered, or when two networks must never exchange traffic.

Our Professional Cloud Network Engineer practice exams cover VPC Network Peering alongside Network Connectivity Center and HA VPN, with detailed explanations that drill these distinctions.

Get tips and updates from GCP Study Hub

arrow