Atlas requires that the use of Private Service Connect for a geo-replicated cluster, requires PSC to be configured in all regions that cluster is in.

The best practice for a DR configuration in Atlas is to have a 2-2-1 configuration over 3 regions so that the cluster fails over automatically in the case of a region outage and there is no split-brain possibility.

The applications themselves really only need to be in 2 regions. The requirement for a PSC in the 3rd region, along with a customer-side project to host that connection is wasteful. This situation is especially problematic in those geo-political areas where the cloud provider has only 2 regions, like GCP in Canada and Australia. Using PSC and having to meet this PSC-in-every-region requirement is wasteful and not particularly useful to clients. For example, apps deployed in 2 GCP regions would not be able to see the 5th member in AWS, even if there was a PSC (or the AWS equivalent) configured in that 5th region.

I suggest that MongoDB add a redundancy node type whose sole purpose is to provide that region outage resiliency, but is hidden. Hidden members can vote. Additionally, this redundancy node does not require a PSC connection in the case where PSC connections are used for the other two regions.