AI development at Chinese listed firms widens wage gaps: firms with more AI-related patents exhibit higher skill premiums and lower labor income shares, with stronger effects in non-state firms, more digitalized firms, and less concentrated industries.

Main Finding

AI development at the firm level (measured from AI-related patenting by Chinese listed firms) significantly increases the firm-level skill premium — i.e., it raises earnings differences between high- and low-skilled workers. This effect is robust to multiple checks and instrumental-variable (IV) treatment, and is driven by AI substituting for low-skilled labor and by AI-related productivity, capital-deepening, and technological-upgrading effects. AI adoption is also associated with a decline in firms’ labor income share and larger income disparities across industries.

Key Points

• Primary result: Firm-level AI development significantly raises the skill premium within firms.
• Mechanisms identified:
  • Direct substitution of AI for low-skilled labor (reducing demand or relative wages for low-skill workers).
  • Productivity gains that raise returns to more-skilled workers (scale/complementarity with high skills).
  • Capital deepening (more and higher-quality capital complementary to high-skill labor).
  • Technological upgrading that favors skilled tasks and occupations.
• Robustness: results hold after an IV approach to address endogeneity and in many sensitivity checks:
  • Alternative constructions/measurements of the dependent variable (skill-premium variants).
  • Alternative AI measures.
  • Policy-shock tests (AI pilot zone tests).
  • Alternative sample restrictions.
• Heterogeneity: effects are stronger in non-state-owned firms, firms with higher digitalization levels, and firms in industries with lower market concentration.
• Distributional spillovers: AI development is linked to a lower labor income share at the firm level and wider inter-industry income disparities.

Data & Methods

• Data: AI-related patent data for Chinese listed firms used to construct a firm-level AI development measure; matched to firm financials and labor/wage data for firm-level analysis (sample details and years are reported in the paper).
• Dependent variable: firm-level skill premium (earnings/wage gap between high- and low-skilled workers; the paper tests several constructions).
• Empirical strategy:
  • Panel regression models estimating the relationship between firm AI development and the firm-level skill premium, with appropriate controls (firm and time controls are applied in the paper).
  • Instrumental-variable (IV) approach employed to mitigate endogeneity concerns (paper provides IV strategy and diagnostics).
• Mechanism tests:
  • Labor composition and wage responses to identify substitution effects for low-skilled workers.
  • Productivity, capital intensity, and measures of technological upgrading to test complementary channels.
• Robustness checks: multiple alternative variable definitions, alternative AI measures, policy-shock (AI pilot zone) tests, and alternative sample selections.

Implications for AI Economics

• Distributional impact: AI adoption can exacerbate within-firm wage inequality (higher skill premium) and contribute to an overall decline in labor’s share of income — important for macro-distributional dynamics and inequality modeling.
• Policy responses required:
  • Strengthen worker skill formation, retraining, and adaptability to reduce displacement of low-skilled workers and to allow broader sharing of AI gains.
  • Reconsider income-distribution mechanisms (tax, transfer, minimum wages, bargaining institutions) to address widening gaps and falling labor shares.
  • Promote inclusive technology diffusion and policies that encourage complementarities between AI and a broader set of occupations (not only high-skill roles).
• Firm and industry strategy:
  • Non-state firms and digitally advanced firms may be front-runners in AI-driven wage divergence; regulators and policymakers should monitor sectoral concentration and competitive effects.
  • Industrial policy and competition policy matter: less concentrated industries see larger skill-premium effects, suggesting market structure interacts with AI’s distributional consequences.
• Research directions for AI economics:
  • Longer-run impacts of AI on career trajectories, skill accumulation, and intergenerational mobility.
  • Detailed task-level analyses to map which tasks are automated versus augmented and how wages adjust across the skill distribution.
  • Better instruments and exogenous shocks to further pin down causality; cross-country and non-listed firm evidence to assess external validity.
  • Quantifying welfare trade-offs between productivity gains and distributional outcomes to inform balanced policy design.