The first issue is a technical one I think can be solved (think Fediverse-style setups with easily transferable entry into meshes, or multi-mesh joining). N! connections obviously can't be handled so not everything can be connected to everything, but making node transfers fluid and hierarchical with easy swap can make for a robust network. Sort of like how we currently route with DNS + IP4/6, but simpler and broadcasted DNS node provisioning and more flexibility from DNS down to subnetworks. If you can set up an entire mesh node from your cell phone with the click of a button, its sort of hard to shut down an internet. Add decentralized ledger tech (no, not a shitcoin :D) and you have a hard time shutting everything down outside an EMP.
For the second, ease of transition is how to overcome existing network effects. As an example, ADP is bleeding customers to Gusto because they make it so damn easy that the only reason you stay with ADP is because they provide a service (like PEO) that Gusto doesn't yet offer. (plus, less data leakage and sales). You can view Gusto/ADP as B2B providers, but they actually operate as platforms between companies and their employees/contractors and thus the network effect arguments apply. Network effects aren't something to fear or use as an excuse to not build, they're a strategy game.
For the second, ease of transition is how to overcome existing network effects. As an example, ADP is bleeding customers to Gusto because they make it so damn easy that the only reason you stay with ADP is because they provide a service (like PEO) that Gusto doesn't yet offer. (plus, less data leakage and sales). You can view Gusto/ADP as B2B providers, but they actually operate as platforms between companies and their employees/contractors and thus the network effect arguments apply. Network effects aren't something to fear or use as an excuse to not build, they're a strategy game.