one issue with arguments for the modular stack is that they work backwards from an ideal end state and assume the equilibrium has to converge there in the limit
realpolitech doesn't work like that
entrepreneurs are satisficers, they work in the present and try to find pmf as fast as possible
they'll go with the current best stack, not the one that could be the best stack in 10 years
the long run equilibrium (if it even exists?) is very much determined by chance and path dependence