Shared VPC Explained: Host Projects, Service Projects and Where Resources Live

GCP Study Hub
November 10, 2026

Shared VPC lets one host project hold a VPC network, along with its subnets, routes and firewall rules, while other projects, called service projects, deploy resources such as Compute Engine VMs and GKE clusters into the shared subnets. The network belongs to the host project and the workloads belong to the service projects. That split explains most Shared VPC behavior, from who can change a firewall rule to which project a Cloud Router must be created in, and it runs through many design scenarios on the Professional Cloud Network Engineer exam.

Host projects and service projects

Each application or group keeps its own service project, so billing and IAM stay separate per project. A central network team administers the host project and keeps sole control of the network. Developers get compute permissions in their service projects plus the Compute Network User role on the subnets they need. That combination lets them launch VMs and clusters on shared subnets without being able to change subnets, routes or firewall rules.

One host project can serve many service projects. Applications in service projects attached to the same host project share one VPC network and can reach each other over internal addresses. Creating a separate host project for every group only multiplies the networks the central team must maintain, because the point of Shared VPC is one network administered in one place.

When Shared VPC doesn't fit

Shared VPC solves a specific problem, central administration of a network used by many projects, and its limits matter as much as its features:

  • One organization. The host project and its service projects must belong to the same organization, so Shared VPC can't link two companies' separate organizations.
  • Central control. Shared VPC hands network control to the host project's administrators, so it doesn't fit teams that must each administer their own network.
  • Existing instances would move. Workloads run on host project subnets, so existing instances would have to move onto them. Shared VPC doesn't fit when instances must stay on their current networks.

When one of those limits applies, the design keeps separate networks and connects them instead of sharing one, so each network stays under its own administrators.

Where resources live in a Shared VPC

In a Shared VPC, the network and all of its subnets belong to the host project, even when the VMs that use a subnet live in a service project. Anything configured on the network itself is therefore configured in the host project. The same rule decides where each networking resource goes.

Cloud Router, VLAN attachments and Interconnect connections

A Cloud Router must be created in the same project as the VPC network it advertises routes for, so in a Shared VPC it goes in the host project. Once it learns on-premises routes over BGP, those dynamic routes are added to the Shared VPC network, and every service project that uses the shared subnets can reach the data center with no extra configuration.

VLAN attachments follow the Cloud Router. An attachment must be in the same project as the Cloud Router it uses, and that Cloud Router must belong to the VPC network being connected, so the attachments go in the host project too. Attachments can't be created in service projects and pointed at the host project's network.

The physical Dedicated Interconnect connections are more flexible. They can live in their own project, separate from the network that uses them, or in the host project. Either way, every service project reaches on-premises through the one shared network.

Routers or attachments created in service projects can never attach to the shared network, so they add resources with no connectivity benefit. Keeping hybrid connectivity in the host project also keeps it under the central network team, which is the usual Shared VPC model.

Flow logs, DNS and network appliances

  • VPC Flow Logs is a setting on the subnet, so it is enabled from the host project. Once it is on, it records sampled flows for every VM interface in the subnet, whichever service project owns the VM, and sends the records, including byte and packet counts, to Cloud Logging. A service project admin can't change it.
  • Cloud DNS forwarding zones are created in the host project and authorized for the shared network, so every service project VM can use them. DNS configuration for the shared network belongs in the host project rather than in each service project.
  • Network appliances that must be owned in the networks' project belong in the host project, where they stay under the network team's control while service projects keep using the shared subnets.

Reasoning through Shared VPC designs

A reliable way to work through Shared VPC scenarios on the Professional Cloud Network Engineer exam is to ask which project owns the thing being configured. Subnet settings such as flow logs, routes, firewall rules, Cloud Routers, VLAN attachments and DNS forwarding zones for the shared network all live with the network in the host project. Workloads, and the billing and IAM that go with them, live in service projects. When a design calls for separate administrative control, a link between organizations or instances that can't move, Shared VPC is the wrong tool, and the design should connect separate networks instead.

Our Professional Cloud Network Engineer practice exams cover Shared VPC alongside networking IAM roles and hybrid connectivity with Cloud Router, with detailed explanations that drill these distinctions.

Get tips and updates from GCP Study Hub

arrow