Macroeconomics & International Finance Seminar Series Presents: Helen Popper

ABSTRACT:  This paper builds a simple general equilibrium model in which an AI producer is a monopolist who both learns by doing and uses AI recursively as an input. These mechanisms link today’s scale to tomorrow’s costs, so pricing is dynamic: the firm sets a price below the static monopoly benchmark to expand capacity and speed learning. Final goods are produced by monopolistic competitors with constant returns to scale each period. We first use Cobb–Douglas technologies to solve for a generalized balanced growth path that pins down the condition for stable, nonexplosive growth. On this path, AI output grows faster than final output, the relative price of AI falls persistently, real wages rise with overall output, and the specialized–to–nonspecialized wage ratio is flat. We then analyze CES versions of both sectors and derive a closed form effective demand elasticity for AI that combines input substitution in production with final-goods market substitution across varieties. Finally, simulations link adoption and distribution to elasticities, and they allow us to explore the dynamics. When final-goods inputs are complements, adoption is learning-first and capital-light before scaling; when they are substitutes, adoption is scale-first and the two-phase pattern attenuates. On the distribution side, the specialized–to–nonspecialized wage premium is lowest with complements and rises with substitutes. Greater substitutability in AI production amplifies these patterns without changing their sign.