Not a Typical Firm: Capital–Labor Substitution and Firms' Labor Shares
Source: https://www.semanticscholar.org/paper/5445ed2bad62488a70392c845891a2c5b94bea00 ↗
This paper resolves a puzzle that sits at the heart of the automation debate: if machines are replacing workers, why do most firms show rising labor shares? The answer is heterogeneity — large firms automate and drive down the aggregate labor share, while small firms remain labor-intensive because automation has fixed costs.
For product directors this is a model of how technology adoption creates winner-take-all dynamics: the same AI capabilities that help large firms reduce headcount may be economically inaccessible to smaller competitors.
The mathematics are rigorous but the insight travels: automation is not a uniform tide but a differentiating force that reshapes industry structure.
Essential reading alongside Brynjolfsson for understanding how AI transforms not just work but competition.
Central argument
Hubmer and Restrepo resolve the apparent contradiction between aggregate labor share decline and firm-level labor share growth by introducing firm heterogeneity as the key mechanism. Large firms automate aggressively, substituting capital for labor and compressing their labor shares, while simultaneously expanding market share — pulling the aggregate share down. Smaller firms remain labor-intensive not by choice but because automation carries fixed costs that make adoption economically irrational at their scale. The aggregate trend thus masks a structural divergence: automation is a large-firm phenomenon that reshapes competitive dynamics, not a sector-wide transformation.
Critique
The model's explanatory power rests heavily on fixed costs as the barrier to small-firm automation, but this assumption may increasingly misfit a world where AI capabilities are delivered as cloud APIs with near-zero fixed costs and consumption-based pricing. If the fixed-cost barrier erodes — as SaaS and foundation model APIs suggest it already is — the predicted divergence between large and small firms could attenuate or reverse, which the paper does not adequately address. There is also limited treatment of how regulatory environment, firm age, or organizational capability gaps (rather than pure economics) independently constrain adoption.
Why it matters for product
For a CPO deciding where to invest in AI-assisted product capabilities, this paper reframes the build-vs-buy question as a structural one: if automation confers compounding scale advantages, delaying adoption is not a neutral decision but a relative competitive loss against larger, better-resourced product organizations. It also has direct implications for team design — the paper implies that labor-intensive small organizations are not merely under-resourced but may be structurally locked out of the productivity compounding that reshapes industry concentration, suggesting that product leaders in mid-market firms should treat AI tooling access as a strategic priority rather than an efficiency line item.