Very original thought...

I think in the near future (10-50 years) ASICs at 10nm-1nm will be energy efficient enough.

Also, I do not think that Bitcoin requires infinitely increasing number of nodes to function, only needs large number of nodes with sufficient hashing power to prevent someone from mounting a 51% attack.

The currently established model, having few large pools containing majority of the hashing power, I think will be an adequate match to the network's growth challenge to quite a degree.