Network Expansion Limits

The special magic of the network effect: Adding node C to path AB not only makes journeys from B to C, but also A to C. So there is an inherent non-linear scaling to adding links. However, while each link adds more because of the network effect, but each link is less-good than the last, because you always build the best links first. So at some point, any network faces declining marginal returns on expansion. 

Further, in any infrastructure network, expansion is perilous, because infrastructure has non-linear maintenance, and those maintenance costs are end-loaded--cheap initially, but with a ruinously expensive (and unavoidable) balloon payment for reconstruction at the end of their useful life*.

*Rail miles, like road miles, are more discrete than things like bridges, so the necessary 'balloon payment' is much smaller, and it's possible to make a lot of low-quality 'patches' that neither work well nor are cost efficient. (If the patches were either, you'd have built the how system out of 'patches' from the beginning).