I don't have a general theory of monorepo versus polyrepo. I can tell you what we chose on the Cloud Platform team, and why it has held up.

Application landing zones live in a monorepo. Each application gets a folder, and each folder has a pipeline. That was fine for us. The app configurations all call a single wrapper module, and that wrapper calls the shared Terraform modules. We call those modules ETMs internally; I'll just say modules here.

We looked at splitting app configs into separate repos and didn't see an advantage. The configurations aren't independent enough that separate histories would help, and a repo per application would have been more pipeline and permissions work for the same result.

Modules get their own repos

Modules are a different story. Those live in their own repositories. That keeps module development, commit history, and release cadence separate from the landing-zone repo. We use semantic versioning and tags. Most of the consumers are application teams. About 180 landing-zone pipelines call these modules, including a few we use ourselves in the landing-zone repo. Semver is what makes that tolerable: a breaking change or a new feature ships as a version, and teams upgrade when they're ready.

If you're deciding for your own org, the useful question is which parts need to move together and which parts need their own release cycle. For us, the app folders belong together. The modules do not.