Do firms substitute labor for AI at different rates?

The question of whether firms substitute labor for AI is asked constantly but rarely with direct evidence. This study provides some of the first firm-level data on how businesses actually spend on AI tools and how that interacts with the labor force, focusing on the segment most exposed: jobs contracted through online labor marketplaces.

The findings are specific and non-uniform. Increasing exposure to AI shocks causes firms to substitute away from labor marketplaces faster and at a higher rate — and the magnitude differs across firms rather than being uniform. To the authors' knowledge it is the first study to show firm-level differences in substitution patterns. Substitution is not just faster but cheaper: higher AI-exposed firms substitute relatively more labor for AI at a lower cost.

The interpretive keeper is the asymmetry, not the substitution itself. That high-exposure firms substitute both faster and cheaper suggests something beyond uniform technology diffusion — candidate mechanisms include returns to scale in building internal AI capability, or a natural advantage in AI adoption that compounds. The paper is careful that it identifies the pattern but not the mechanism. For writing about AI and labor, the durable point is that AI's labor-market effect is heterogeneous across firms — concentrating capability and cost advantages in already-exposed firms — which reframes the displacement debate from "will AI replace workers" to "which firms capture the substitution surplus." This sharpens Does concentrated AI exposure enable workers to adapt and reallocate? from the firm side rather than the worker side.