AI nudges firms downstream in global value chains and makes technology more labor-augmenting, lifting workers' share and narrowing internal pay gaps but squeezing corporate profits; effects are most pronounced in private firms and when AI hardware platforms are deployed.

Main Finding

AI adoption shifts firms’ positions in global value chains (GVCs) and, through both direct and GVC-mediated channels, induces labor-biased technological progress. This labor bias raises firms’ labor income share and reduces internal wage inequality but lowers profitability. The effects are strongest for private firms and for use of AI hardware platform technologies.

Key Points

• The paper develops a task-based trade model linking AI, firm GVC position (downstream vs upstream), and the bias of technological progress.
• Empirical analysis uses panel data on Chinese A-share listed firms (2007–2021).
• Two complementary channels:
  • Direct channel: firms’ AI application directly promotes labor-biased technological progress.
  • GVC channel: AI encourages firms to embed more downstream in global value chains, which further catalyzes labor-biased technological progress.
• Heterogeneity:
  • Effects are concentrated in private (non-state) firms.
  • Stronger for firms deploying AI hardware platform technologies (vs other AI types).
• Economic consequences of the GVC-driven labor bias:
  • Increases firms’ labor income share.
  • Narrows internal salary gaps within firms.
  • Reduces firm profitability.
• The study highlights a complex interaction between technological change (AI) and globalization (GVC embedding) that shapes distributional and firm-performance outcomes.

Data & Methods

• Theoretical framework: a task-based trade model that maps how AI affects firm tasks, trade patterns, and position along GVCs, and how these in turn bias technological progress.
• Empirical setting: panel of publicly listed Chinese firms (A-share), 2007–2021.
• Empirical strategy (as reported):
  • Measure firms’ AI application and classify AI types (including hardware platform technologies).
  • Measure firms’ GVC position (extent of downstream embedding).
  • Estimate the effects of AI on technological progress bias, using panel techniques and mediation/channel analysis to separate direct and GVC-mediated effects.
  • Conduct heterogeneity analysis by ownership type (private vs state) and by AI technology type.
  • Examine downstream outcomes (labor income share, internal wage gap, profitability) to assess economic implications of the identified channels.
• (Note: the paper reports robustness checks and further tests consistent with the theoretical predictions; specifics of variable construction and identification strategies are in the full text.)

Implications for AI Economics

• Distributional effects of AI depend on firm-level GVC positioning: AI can be labor-augmenting (labor-biased) rather than uniformly capital-replacing, especially when firms move downstream in GVCs.
• Policy design should consider the joint role of digital technology and trade structure:
  • Labor-market policies: support training, reallocation, and social protections anticipating increased labor shares in some firms but potential profitability pressures.
  • Industrial policy: recognize that AI hardware platforms and private firms are key loci of these effects; targeted support or regulation may be warranted.
  • Trade/GVC policy: interventions that affect firms’ upstream/downstream positioning (e.g., trade costs, tariffs, sourcing rules) can alter the distributional consequences of AI.
• Research implications:
  • Future work should probe external validity beyond Chinese listed firms, distinguish AI subtypes more finely, and strengthen causal identification of the GVC channel.
  • The finding that AI can be labor-biased in some settings challenges simple narratives of universal labor displacement and suggests richer, context-dependent models in AI economics.