AI diffusion lowers industry concentration after initial rise

The study develops a general equilibrium model showing that AI diffusion initially increases sales concentration before reducing it once adoption becomes widespread.

In early stages, large incumbent firms leverage scale and data resources to expand market share through the intensive margin.

However, as AI supply grows and adoption costs decline, smaller firms enter as new adopters, triggering an extensive margin effect that erodes incumbent dominance.

Calibrated to U.S. economic data, the model identifies a turning point at an AI adoption rate of roughly 15 percent.

With current U.S. business adoption standing at approximately 18 percent, further AI diffusion is projected to lower market concentration across industries.

The impact on average markups depends on whether diffusion is driven by supply or demand factors.

Supply-driven progress, such as falling hardware costs, lowers marginal production costs and raises markups monotonically.

Conversely, demand-driven adoption via lower fixed costs yields a non-monotonic markup pattern.

Addressing market distortions from variable markups and monopolistic AI production, the paper demonstrates that a modest 3 percent revenue subsidy for AI adopters maximizes social welfare by balancing productivity gains against markup distortions.