It’s highly dependent on infa-setup whether there’s a cost reduction, and in many setups it’s actually possible that there will actually be a cost increase. Why? Because if you are isolating your setups, as you consolidate validators (stake) into 1 validator client, you need to increase the amount of pods / “instances” you are running (i.e. VC + EL+CL combos), because essentially for blast-radius minimization and containment purposes, you will likely not want to be running more than 1 to 8 2048 ETH (as an example, 8x2048 ~= 500x32) validators per VC.
Ultimately, it depends on how an NO is running their infrastructure. One possibility is that an NO could utilize the same EL+CL pair for more VCs than before, but if for example your infra paradigm is to keep all pods isolated then this doesn’t work.