AI adoption among Chinese listed firms is linked to weaker corporate innovation as firms reallocate resources toward building and maintaining AI systems rather than R&D and technical equipment. State ownership, tougher competition, easier financing, high-tech sectors and substantial managerial shareholding help blunt this crowding-out effect.

Main Finding

AI application intensity significantly crowds out enterprise innovation in Chinese A-share firms (2013–2023). This inhibitory effect operates mainly by (1) reducing investment in innovation talent and (2) cutting the configuration of professional technical equipment. Certain firm characteristics — state ownership, intense market competition, low financing constraints, high technology intensity, and high management shareholding — mitigate the negative impact.

Key Points

• The paper frames the problem using resource allocation theory, principal–agent theory, and path-dependence theory: heavy AI spending can reallocate scarce resources away from traditional R&D and foster technological path lock-in.
• Empirical tests support three hypotheses:
  • H1: AI exerts a crowding-out effect on enterprise innovation.
  • H2: AI reduces innovation by lowering innovation-talent investment intensity (mediation).
  • H3: AI reduces innovation by lowering firms’ investment in professional technical equipment (mediation).
• Heterogeneity analysis: the negative AI–innovation relationship is weaker for firms with:
  • State ownership,
  • High market competition,
  • Low financing constraints,
  • High technology intensity,
  • High management shareholding.
• Policy/practice concern: excessive AI emphasis can produce short-term efficiency gains but damage long-term, sustainable innovation capabilities.

Data & Methods

• Sample: Chinese A-share listed companies, 2013–2023; final cleaned sample = 10,827 firm-year observations.
• Data sources:
  • Patent data: China Patent Research and Service System (CPRS) — invention patent applications used as innovation measure.
  • Annual-report text (AI-related mentions): Wind Financial Terminal; AI-application intensity index constructed via Python text analysis of annual reports and keyword extraction.
  • Financial and governance variables: CSMAR database; R&D personnel and related inputs from Wind and CSR reports.
• Sample cleaning: excluded financial firms, ST/*ST firms, major restructuring cases, and observations with missing key variables; continuous vars winsorized at 1%/99%.
• Outcome variable: LN_PAT = ln(1 + annual invention patent application count).
• Empirical strategy: panel regression analyses (firm-level) with robustness checks; mediation tests to identify channels through innovation-talent intensity and professional-technical-equipment configuration; heterogeneity/subsample analyses to test moderating firm characteristics.
• Robustness: alternative specifications and tests (details in paper) confirm main results.

Implications for AI Economics

• For firms:
  • Balance AI investment with sustained human-capital development and targeted equipment upgrades. Over-investing in AI platforms at the expense of specialist talent and equipment risks long-term innovation decline and technological lock-in.
  • Governance and incentive alignment (e.g., greater managerial ownership) can encourage managers to weigh long-term R&D trade-offs rather than pursue short-run AI-driven efficiency alone.
• For policymakers:
  • Provide targeted financing, subsidies, or tax incentives that preserve investment in basic R&D, specialized equipment, and talent training during digital transformation—especially for resource-constrained SMEs.
  • Support workforce reskilling and lifelong learning programs to maintain exploratory innovation capacity alongside AI adoption.
  • Monitor market structure and competition to ensure that AI adoption does not concentrate advantages that reduce broader innovation activity.
• For researchers and investors:
  • AI adoption is not uniformly pro-innovation; its net effect depends on resource trade-offs and institutional contexts. Investors should evaluate firms’ AI strategies alongside measures of sustained R&D and human-capital investment.
  • Future research should explore causal identification across contexts (other countries/industries), longer-term outcomes (innovation quality vs. quantity), and optimal mixed-investment strategies that combine AI platforms with traditional R&D capabilities.
• Caveats: findings are based on Chinese listed firms and patent-application measures; AI intensity is proxied by textual mentions, which may imperfectly capture actual resource commitments.