We study optimal policy in a dynamic general equilibrium model where heterogeneous monopolistic competitive firms pay a fixed cost to adopt a frontier technology that grows exogenously. Using Mean Field Games tools, we show that the optimal policy consists of exactly two time-invariant subsidies: one correcting the static misallocation from market power, and one correcting the dynamic under-incentive to adopt. This holds outside of balanced growth paths, for any initial distribution of technology gaps. We analyze a simplified version of the model that aggregates to a Neoclassical Growth Model with an S-shaped production function whenever complementarities are strong, and fully characterize when the optimal policy uniquely implements the first best. When it does not, two novel results emerge: the efficient allocation prescribes escaping a poverty trap-providing an explicit optimality foundation for a Big Push-and, more surprisingly, escaping an abundance trap, where dismantling adopted technologies is optimal. In both cases, a temporary, costless supplementary policy restores unique implementation.