AI initially boosts China’s low‑altitude economy but delivers diminishing returns: provincial evidence shows an inverted‑U relationship within regions and a U‑shaped spatial spillover across neighbors, while a Big Data Pilot Zone shock supports the robustness of the pattern; human‑capital and innovation channels appear weak or counterproductive in the short term due to resource misallocation.

Main Finding

AI development has a non-linear, spatially heterogeneous impact on the Low‑altitude economy (Lae) in China. Using 2012–2022 panel data for 30 provinces, the authors find a significant inverted U‑shaped direct effect of AI on Lae (AI boosts Lae up to a turning point, after which marginal returns decline) and a U‑shaped spatial spillover effect (early-stage AI in neighbouring regions can depress local Lae but becomes beneficial past a threshold). Short‑term mediating channels through human capital and technological innovation are weak or even negative—attributed to resource misallocation and skill mismatches—while AI ranks as the single most important predictor of Lae performance in artificial neural network (ANN) prediction exercises. Results are robust to a multi‑period spatial DID using the National Big Data Pilot Zone designation as an exogenous shock.

Key Points

• Nonlinear direct effect: AI → Lae follows an inverted U shape (positive effect up to a turning point; diminishing/negative marginal returns thereafter).
• Spatial dynamics: AI shows a U‑shaped spatial spillover on neighbouring regions’ Lae (initially negative spillovers that turn positive as neighbouring AI reaches higher levels).
• Mediating channels: Human capital and technological innovation channels are weaker than expected, and may generate short‑term negative effects due to:
  • resource reallocation away from existing Lae activities,
  • skill mismatches between workers and new AI‑enabled tasks,
  • short‑run disruption from infrastructure and regulatory lags.
• Robustness: Findings hold in a quasi‑experimental spatial DID using the National Big Data Comprehensive Pilot Zone policy as an exogenous shock.
• Predictive importance: In ANN models, AI is the single most important predictor of provincial Lae outcomes, corroborating econometric findings from a forecasting perspective.
• Contribution: Integrates spatial econometrics and machine learning (SDM + spatial mediation + ANN) and constructs entropy‑based composite indices for AI and Lae.

Data & Methods

• Data: Panel dataset covering 30 Chinese provincial regions (2012–2022). Composite indices for AI development and Lae constructed using the entropy method. GIS maps and kernel density estimates depict spatiotemporal patterns.
• Literature mapping: CiteSpace bibliometric analysis (1995–2025) used to identify research themes and guide variable selection.
• Econometric strategy:
  • Spatial Durbin Model (SDM) with a quadratic term for AI to capture nonlinearity and spatial dependence; estimates direct and spillover effects.
  • Spatial mediation analysis incorporating human capital, technological innovation, and government intervention as potential channels.
  • Multi‑period spatial difference‑in‑differences (DID) exploiting designation of National Big Data Pilot Zones as a quasi‑natural experiment for robustness.
• Machine learning:
  • Artificial Neural Network (ANN) used to rank predictive importance of AI relative to other covariates for Lae forecasting.
• Diagnostic/visual methods: Kernel density estimation, GIS mapping of indices, CiteSpace network visuals.
• Note: Manuscript provided is an unedited preprint (Article in Press).

Implications for AI Economics

• Policy must be stage‑sensitive: Because AI’s benefits follow an inverted U, policymakers should recognize diminishing returns at advanced stages and avoid one‑size‑fits‑all promotion. Continued investment must be complemented by policies that sustain complementarities (e.g., infrastructure, standards).
• Manage short‑run disruption: Weak/negative short‑term effects through human capital and innovation channels imply active measures are needed to reduce workforce displacement and skill mismatches—targeted retraining, education, and labor market matching are critical.
• Coordinate regional policy to harness spillovers: The U‑shaped spatial effect suggests that early AI buildout in some regions can harm neighbours (through competition for resources or talent) until neighbouring regions also scale AI; coordinated regional planning and shared infrastructure (airspace management, data platforms) can accelerate positive spillovers.
• Use mixed methods for evaluation: Combining spatial econometrics with predictive ML (ANN) offers complementary causal and predictive insights for technology policy evaluation—advocated for future AI economic assessments.
• Monitor and measure thresholds: Given the nonlinearity and spatial thresholds, policymakers and firms should track region‑level AI indicators and Lae outcomes to identify turning points where marginal returns change sign, and adjust incentives accordingly.
• Research implications: Future work should quantify the turning points and examine longer‑term dynamics beyond 2022, explore firm‑ and city‑level heterogeneity, and assess how regulatory design (airspace rules, safety, privacy) conditions AI–Lae interactions.

Reference (from manuscript): Yin, J., Chen, W., & Wang, F. (2026). A study of the impact of artificial intelligence on the low‑altitude economy. Humanities and Social Sciences Communications. DOI: 10.1057/s41599-026-07561-w

If you want, I can: - Extract the estimated turning‑point values and key coefficient estimates if they appear later in the paper; - Draft short policy recommendations tailored to central, coastal, and inland provincial contexts; or - Convert this summary into a one‑page policy brief. Which would be most useful?