China’s digital economy boosts provincial carbon productivity both locally and across borders, while industrial intelligence depresses local carbon productivity but produces strong positive spillovers that make its net effect beneficial; impacts vary sharply by region.

Main Finding

Using panel data for 30 Chinese provinces (2010–2023), the paper finds that both the digital economy and industrial intelligence (AI-driven industrial technologies) improve regional total factor carbon productivity (TFCP) primarily through spatial spillovers. The digital economy has a significant positive local (direct) effect and a strong positive spatial spillover — a “local driving + spatial radiation” pattern. Industrial intelligence shows an insignificant or slightly negative local effect but a highly significant positive spatial spillover (1% level), yielding an overall significantly positive total effect. TFCP rose nationally over the sample but with widening regional gaps (“higher in the east, lower in the west”) and positive spatial autocorrelation.

Key Points

• Outcome measured: Total factor carbon productivity (TFCP) — GDP relative to CO2 emissions accounting for capital, labor, energy.
• Data: Panel of 30 Chinese provincial-level regions, 2010–2023.
• Measurement: TFCP estimated by an undesirable-output super-efficiency SBM (Slack-Based Measure) DEA model.
• Empirical model: Spatial Durbin Model (SDM) following confirmation of spatial autocorrelation (Moran’s I).
• Main empirical findings:
  • Digital economy: significant positive direct effect on local TFCP and strong positive indirect (spillover) effect to neighbors.
  • Industrial intelligence (AI/industrial internet/robotics): insignificant negative direct effect on local TFCP, but significant positive indirect (spillover) effect at 1% — total effect significantly positive.
  • Regional heterogeneity:
    • Industrial intelligence: negative direct effect in the east, positive direct effect in the central region, insignificant direct effect in the west; indirect (spillover) effects positive in east and west.
    • Digital economy: east shows dual local+spillover drive; central is local-dominated; west is spillover-dominated.
  • Controls: coal-heavy energy structure and secondary-industry-dominated industrial structure strongly and negatively affect TFCP with negative spatial spillovers. Green finance shows an insignificant positive effect. FDI has an insignificantly positive local effect but a significantly negative indirect effect (evidence of a “pollution haven” spillover).
• Interpretation: Industrial intelligence’s primary carbon gains materialize through interregional coordination and technology diffusion rather than immediate local gains (likely because adoption/transition costs or rebound effects reduce local short-term benefits). The digital economy more immediately raises TFCP locally and also radiates benefits regionally.

Data & Methods

• Sample: 30 provincial-level Chinese regions (2010–2023).
• TFCP measurement:
  • Inputs: capital (regional fixed capital stock via perpetual inventory method), labor (year-end employed persons per area), energy (total energy converted to standard coal).
  • Desired output: real regional GDP (2010 base).
  • Undesired output: CO2 emissions (IPCC method aggregating 8 fossil fuel types).
  • Efficiency model: super-efficiency SBM DEA (Tone, 2002) with undesirable outputs and weak disposability assumptions; ρ* ≥ 1 indicates super-efficiency.
• Spatial analysis:
  • Spatial autocorrelation tested by global Moran’s I (found significant positive spatial clustering of TFCP).
  • Spatial Durbin Model (SDM) estimated to capture both direct and indirect (spatial spillover) effects of industrial intelligence and the digital economy.
  • Spatial weight matrix used to construct neighbor relations (paper discusses construction but full details not shown in the provided excerpt).
• Robustness and heterogeneity: Regional subsample analyses (east/central/west) to identify heterogeneous effects.
• Caveats mentioned by authors: manuscript unedited pre-publication; theoretical considerations include possible rebound effects for industrial intelligence and the carbon footprint of the digital economy (tipping-point behavior).

Implications for AI Economics

• AI (industrial intelligence) operates as a spatial externality: local deployment may not immediately improve a province’s TFCP but raises TFCP in neighboring regions and yields net positive total effects. Policy design should therefore:
  • Encourage interregional coordination and mechanisms to internalize cross-border benefits (e.g., interregional green transfers, joint R&D, coordinated industrial parks).
  • Support diffusion channels (labor mobility, supply-chain linkages, shared digital/compute infrastructure) so local adoption benefits are realized more fully.
• Rebound risks and short-term local negative effects: AI-driven efficiency gains can trigger scale-up or substitution effects that temporarily reduce local TFCP. Policies should:
  • Combine AI deployment with energy/industrial policy (e.g., cleaner energy mix, emissions standards) to limit rebound.
  • Incentivize adoption of low-carbon AI applications (e.g., predictive maintenance, process optimization) while disincentivizing expansions that raise emissions intensity.
• Digital economy as a dual engine: The digital economy both directly raises local TFCP and radiates benefits, so investments in digital infrastructure, data centers, cloud and edge computing can have high leverage for carbon productivity — but:
  • Account for the carbon footprint of digital infrastructure (data centers, networks). Promote energy-efficient data centers, renewable power sourcing, and lifecycle assessments.
  • Track the “tipping point”: ensure digital sector emissions are offset by broader efficiency/structural gains.
• FDI and structural risks: Negative spillovers from FDI (pollution haven effect) underline the need to attach environmental performance standards to foreign investment and prevent relocation of dirty industries to neighboring regions.
• Policy targeting by region:
  • East: address possible short-term local negative effects of industrial intelligence; leverage digital economy both locally and regionally.
  • Central: promote local adoption of industrial intelligence to realize direct gains.
  • West: emphasize receiving spillovers — build absorptive capacity (skills, infrastructure) to capture benefits.
• Research and measurement recommendations for AI economics:
  • Investigate causal channels and firm-level mechanisms (microdata) to separate adoption costs, rebound, scale effects.
  • Quantify lifecycle emissions of AI/digital infrastructures and model thresholds/tipping points.
  • Design metrics that capture both local and spatial spillovers when evaluating AI and digital investments.

Limitations to keep in mind: this is an unedited manuscript (possible errors), potential endogeneity concerns (reverse causality between TFCP and digital/AI deployment), and details of the spatial weight matrix and robustness checks are not fully visible in the excerpt. Further causal identification and micro-level analyses would strengthen policy prescriptions.