Integration of data and AI shrinks China’s provincial development gaps, working mainly through innovation, industry upgrading and improved resource allocation; once financial development passes a 12.7% threshold the payoff from DIAI jumps markedly, especially in western regions and digital-economy pilot zones.

Main Finding

The integration of data factors and artificial intelligence (DIAI), measured by a Data‑Intelligence Fusion Index (DIFI), significantly narrows provincial development gaps in China (2012–2023). DIAI reduces dispersion of per‑capita GDP (coefficient of variation) and inequality (Theil L). Its impact operates mainly through three mediating channels — innovation, industrial‑structure upgrading, and resource‑allocation efficiency — and is strongly amplified once financial development exceeds a threshold (12.7%). Effects are largest in western provinces and national digital‑economy pilot zones.

Key Points

• Data and sample: Panel of 30 Chinese provinces (2012–2023); Tibet excluded.
• Core effect: A one‑unit increase in DIFI is associated with a reduction in the coefficient of variation of per‑capita GDP by 0.038 and the Theil index by 0.042 (two‑way fixed effects).
• Mediation (contribution shares):
  • Innovation mechanism: indirect effect −0.012; contributes 42.1% of DIFI’s total effect.
  • Industrial upgrading: indirect effect −0.009; contributes 28.6%.
  • Resource allocation (TFP/factor matching/public service equalization): indirect effect −0.015; contributes 29.3%.
  • Total indirect effect ≈ −0.036 (three channels explain ~72.3% of the integration effect).
• Threshold: Financial development (share of financial industry value added in GDP) shows a single threshold at 12.7%.
  • Below 12.7%: marginal DIFI effect ≈ −0.011.
  • At/above 12.7%: marginal DIFI effect ≈ −0.043 (≈3.9× larger).
• Regional heterogeneity:
  • Western China: DIFI coefficient −0.051 (stronger catching‑up effect).
  • National digital economy pilot zones: DIFI coefficient −0.058 (largest effect).
• Robustness: results hold when (i) substituting core indicators (number vs. density of AI firms), (ii) IV estimation using number of provincial data trading platforms (first‑stage F = 18.7), and (iii) excluding 2015–2017.
• Limitations noted by authors: micro‑level mechanisms (firm/household behavior) insufficiently explored; Xizang data missing.

Data & Methods

• Outcome variables for regional balance: coefficient of variation of per‑capita GDP; Theil‑L index (provincial GDP shares).
• Core explanatory variable: Data‑Intelligence Fusion Index (DIFI) built by entropy weights on:
  • Data transaction volume,
  • AI enterprise density (enterprises per 10k people),
  • AI patent applications.
• Mediators:
  • Regional innovation index (standardized R&D expenditure, patent grants, S&T personnel),
  • Industrial structure upgrading (tertiary industry value added / GDP),
  • Total factor productivity (TFP) estimated via Levinsohn‑Petrin.
• Controls: human capital (higher ed enrollment), urbanization rate, government intervention (fiscal expenditure/GDP), etc.
• Econometric strategy:
  • Two‑way fixed effects for baseline causal association.
  • Mediation analysis with bootstrap (5,000 resamples) to quantify indirect effects.
  • Hansen threshold regression to detect nonlinear moderation by financial development; likelihood ratio test supports a single threshold at 12.7%.
  • Robustness checks: alternative DIFI composition, IV estimation, sample‑period exclusions; winsorization and multiple imputation applied for data cleaning.

Implications for AI Economics

• Complementarity: Data and AI form a synergistic production factor — policy and investment that treat them jointly yield larger regional convergence effects than acting separately. Economists should model data and AI as complementary inputs (nonlinear returns).
• Role of finance as an amplifier: Financial development is a key enabling condition. There are nonlinear returns to DIAI where financial system scale/depth materially increases the payoff to data+AI investments. Empirical and theoretical work should incorporate financial market depth as a moderator of AI adoption returns.
• Mechanisms to target: Most of DIAI’s convergence effect comes via innovation (knowledge spillovers), but substantial shares flow through industry upgrading and factor reallocation. Cost‑benefit analysis of AI investments should therefore include spillover magnitudes (regional knowledge diffusion), structural change effects, and matching frictions reductions.
• Policy design and spatial targeting: Late‑developing regions (e.g., western China) can realize larger marginal gains from DIAI — suggesting high social returns to targeted public investments in digital infrastructure, data platforms, and sectoral AI pilots. Pilot zones accelerate diffusion; scaling pilots can be an effective policy instrument.
• Measurement and valuation of data: The study operationalizes a composite DIFI; AI economics must standardize measurement of “data as a factor” (transaction volume, tradability, platform presence) to better estimate elasticities and general equilibrium effects.
• Distributional and labor market considerations: While DIAI narrows regional GDP dispersion, implications for wages, employment composition, and skill premiums are implied but not resolved. Micro‑level research is needed to assess whether convergence arises with inclusive labor market outcomes or via capital/skill‑biased gains that may require complementary human‑capital policies.
• Research agenda: Important open questions include firm‑level adoption dynamics, heterogeneous returns across industries, optimal sequencing of financial and digital reforms, and welfare implications of data market architectures. Causal identification at micro scales (e.g., randomized pilots, firm‑level panel IVs) would strengthen inference about mechanisms.

Summary takeaway: Joint investments in data markets and AI can be a potent policy lever for regional convergence, but their effectiveness depends critically on financial development and policy design that fosters knowledge spillovers, industrial upgrading, and efficient factor reallocation.