Number theory and algebraic geometry were developed for their own sake (i.e. "it is cool"), but later people found practical applications in cryptography.

So "useful math must be motivated by practice" is empirically false