China's digital expansion first raises and then lowers per-capita city emissions while hurting then improving emission efficiency; meaningful carbon cuts materialize only once green-technology innovation crosses a critical threshold, implying digitalization must be sequenced with innovation policy to deliver low-carbon outcomes.

Main Finding

Using 2011–2022 panel data for 278 Chinese cities and a mix of panel fixed-effects, mediating-effect, and threshold-regression models, the study finds that the digital economy (DE) has non-linear and context-dependent impacts on urban carbon outcomes: DE exhibits an inverted-U relationship with per capita carbon emissions (PCE) and a U-shaped relationship with carbon emission efficiency (CEE). Crucially, the carbon-reduction benefits of DE emerge only after green technology innovation passes a critical threshold; otherwise DE mainly operates by changing CEE. Effects vary across city agglomeration types, implying the need for differentiated, innovation-focused policy sequencing.

Key Points

• Data: panel of 278 Chinese cities, 2011–2022.
• Models: panel fixed-effects, mediating-effect (to test channels), and threshold-regression (to test non-linear moderation by green technology innovation).
• Non-linear patterns:
  • DE vs. per capita carbon emissions (PCE): inverted U — DE initially associated with rising PCE, then with falling PCE after a turning point.
  • DE vs. carbon emission efficiency (CEE): U-shaped — at low DE levels efficiency worsens then improves as DE expands further.
• Role of green technology innovation:
  • Green innovation is a critical moderator/threshold variable.
  • When green-tech innovation is low, DE’s main measurable effect is on CEE (efficiency channel), but DE does not yet reduce per capita emissions.
  • Once green-tech innovation exceeds a threshold, DE begins to produce stronger direct carbon-reduction effects (reducing PCE).
• Heterogeneity by agglomeration type:
  • In “optimization and upgrading” agglomerations, DE reduces carbon emissions but with a later/timed effect.
  • In “growth and expansion” agglomerations, DE’s impact is more pronounced on improving CEE.
• Policy recommendations (as reported): prioritize innovation-driven digitalization, accelerate diffusion of green technologies, and optimize region-specific mechanisms for coordinated low-carbon growth.

Data & Methods

• Sample: 278 Chinese prefecture-level cities observed annually from 2011 to 2022.
• Dependent variables:
  • Per capita carbon emissions (PCE) — quantity/scale measure.
  • Carbon emission efficiency (CEE) — an efficiency/quality measure (likely output per unit of emissions or efficiency index).
• Main explanatory variable: level of digital-economy development (DE) at city level (treated as continuous, allowing non-linear effects).
• Empirical strategy:
  • Panel fixed-effects models control for time-invariant city heterogeneity and common time effects.
  • Mediating-effect models test whether CEE mediates the DE → PCE relationship.
  • Threshold-regression models use a measure of green-technology innovation as the threshold variable to detect regimes in which DE’s effects differ.
• Identification approach relies on longitudinal variation and fixed effects; non-linearities and regime-specific effects are identified via threshold estimation. (Robustness tests and exact variable constructions are not detailed here.)

Implications for AI Economics

• Complementarity matters: Digitalization (including AI adoption) alone is insufficient to guarantee carbon reductions. AI-driven productivity or service expansion can initially raise emissions unless paired with strong green-technology innovation. For AI economics, this implies modelling complementarities between AI adoption and green-tech R&D/investment when evaluating net environmental impacts.
• Sequencing and policy design:
  • Prioritize policies that couple AI/digital diffusion with incentives for green innovation (R&D subsidies, standards, public–private partnerships).
  • In early-stage digitalization contexts, emphasize efficiency-raising AI applications (e.g., process optimization, demand-response) to improve CEE, while building green-innovation capacity to unlock later emissions reductions.
• Regional and sectoral heterogeneity:
  • Economic geography matters: cities at different development/agglomeration stages will experience different AI–emissions trade-offs. AI deployment strategies and carbon policy should be tailored to local industrial structure and innovation capacity.
• Measurement and evaluation:
  • Empirical studies of AI impacts on emissions should allow for non-linearities and threshold effects (e.g., interaction with local green-innovation intensity).
  • Use both scale (emissions quantities) and efficiency (emissions per unit output or welfare) metrics to capture distinct channels.
• Research directions for AI economists:
  • Decompose which digital/AI components (infrastructure, platforms, automation, AI energy use) drive the inverted-U vs. U-shaped patterns.
  • Study causal mechanisms with quasi-experimental variation in AI/digital adoption and green-innovation shocks.
  • Explore policy instruments that accelerate the green-innovation threshold (e.g., targeted R&D, technology diffusion programs), and quantify welfare trade-offs across regions and income groups.
• Practical policy takeaway: to realize the environmental benefits of AI and broader digitalization, pair digital investments with policies that raise green-technology capacity and diffusion; otherwise, digital expansion risks temporary increases in emissions or limited efficiency gains only.