China’s rise in AI appears to narrow regional carbon-emissions gaps by improving energy efficiency and environmental monitoring, but deep structural divides persist—green innovation remains coastal and has not yet reduced inland–coastal emission inequality.

Artificial intelligence, green innovation, and regional carbon inequality: evidence from Chinese provincial data

Xiangjun Fan · March 30, 2026 · Humanities and Social Sciences Communications

openalex correlational medium evidence 7/10 relevance DOI Source PDF

Using a 2003–2021 provincial panel for China, the paper finds that higher regional AI development is associated with reduced interprovincial carbon-emissions inequality (notably by the Gini index) through improved energy efficiency, monitoring, and resource allocation, while green innovation has not yet narrowed disparities because it remains concentrated in coastal provinces.

China’s pursuit of carbon peaking and carbon neutrality has heightened the urgency of addressing regional differences in carbon emissions. While artificial intelligence (AI) is increasingly regarded as a transformative force for industrial upgrading and environmental management, its impact on the spatial distribution of carbon emissions remains insufficiently understood. This study empirically examines how AI development and green innovation affect carbon inequality using a provincial panel dataset spanning 2003–2021. Carbon inequality is measured using both the Gini index and the Theil index to capture different dimensions of regional disparities. The results reveal three main findings. First, AI development significantly reduces carbon inequality, particularly when measured by the Gini index. This effect operates primarily through improved energy efficiency, enhanced environmental monitoring, and more efficient resource allocation, thereby disproportionately benefiting lagging regions and helping narrow interprovincial emission gaps. However, the inequality-reducing impact of AI is weaker when assessed using the Theil index, suggesting that deep-rooted structural divides between advanced and less developed regions remain difficult to overcome. Second, green innovation does not yet significantly reduce carbon inequality, largely because innovative activities are concentrated in coastal provinces and have not effectively diffused to inland areas. Overall, this study provides new insights into the mechanisms through which digital technologies shape environmental inequality and highlights the importance of coordinated digital green development strategies to promote a more balanced and inclusive transition toward China’s dual-carbon goals.

Summary

Main Finding

AI development in China (2003–2021, provincial panel) significantly reduces regional carbon-emissions inequality—especially when inequality is measured by the Gini index—primarily via improved energy efficiency, enhanced environmental monitoring, and more efficient resource allocation that disproportionately benefit lagging provinces. By contrast, green innovation (as currently distributed) does not yet significantly reduce carbon inequality because innovative activity is concentrated in coastal/advanced provinces and has not diffused effectively to inland regions. The inequality-reducing impact of AI is weaker when measured by the Theil index, indicating persistent deep structural divides between advanced and less-developed regions.

Key Points

Scope: Provincial panel data for China, 2003–2021. Carbon inequality measured with both Gini and Theil indices to capture different dimensions of spatial disparity.
Main empirical result: AI development → significant decline in carbon inequality (stronger for Gini). Theil-based results show a weaker effect, implying persistent structural gaps.
Mechanisms identified: AI reduces inequality primarily through (i) raising energy efficiency, (ii) improving environmental monitoring and governance enforcement, and (iii) enabling more efficient resource and production allocation that helps lagging regions catch up.
Role of green innovation (GI): GI is an expected mediator between AI and carbon outcomes, but current GI benefits are spatially concentrated in coastal, higher-capacity provinces and thus do not (yet) significantly mediate reductions in carbon inequality nationwide.
Heterogeneity: Effects vary by region (Eastern vs Central/Western/Northeastern provinces). Regions with stronger digital infrastructure, R&D, human capital, and institutional quality capture more of AI’s benefits.
Theoretical framing: Path dependence of regional emissions, absorptive-capacity conditionality (human capital, institutions, infrastructure), and regional innovation-system strength determine whether AI and GI produce equalizing effects.

Data & Methods

Data: Provincial panel covering 2003–2021. Dependent: provincial carbon-emissions inequality (Gini and Theil indices). Key covariates (in logged form): GDP and GDP^2, AI development indicator, green innovation (GRIN) indicator, urbanization (URB). (The manuscript combines AI and GI indicators and inequality metrics in a unified empirical framework; exact indicator definitions are those used by the paper.)
Empirical model: Log-linear panel specification linking CI_it to ln(GDP), ln(GDP^2), ln(AI), ln(GRIN), ln(URB) with province and time variation.
Inference: Fixed-effects-style panel regressions with Driscoll–Kraay standard errors to adjust for heteroskedasticity, serial correlation, and cross-sectional dependence common in provincial environmental data.
Additional analyses: Mediation tests for green innovation as a transmission channel; spatial/region-specific heterogeneity checks across Eastern, Central, Western, Northeastern provinces; robustness checks reported (index sensitivity: Gini vs Theil).

Implications for AI Economics

Distributional effects matter: Beyond aggregate emissions, AI adoption has important spatial distributional impacts. Economists studying AI should incorporate inequality metrics (Gini, Theil) to capture who benefits from AI-enabled environmental gains.
Conditionality and absorptive capacity: AI’s equalizing potential depends on complementary investments—digital infrastructure, human capital, R&D capacity, and institutional quality. Cost–benefit analyses of AI deployment should account for heterogeneous returns across regions and the need for public investment to enable equitable gains.
Technology diffusion & policy design: Green innovation tends to cluster in advanced regions; without targeted policies (technology transfer, subsidies, cooperative R&D, infrastructure investment), AI-enabled innovations may reinforce rather than reduce regional disparities. Policy interventions can align AI deployment with inclusive green outcomes.
Measurement nuance: Different inequality metrics reveal different dimensions (Gini captures relative spread; Theil is more sensitive to between-group/structural gaps). AI economics work should use multiple inequality measures when assessing distributional impacts of digital technologies.
Research agenda: Further causal work (e.g., quasi-experimental identification), micro-level firm/sector analyses of AI→GI pathways, and studies on policy instruments that foster diffusion of AI-enabled green innovations to lagging regions are important next steps.

Note: This summary is based on an unedited manuscript (Article in Press). The paper’s precise variable constructions and some estimation details are as reported by the authors; consult the final published version for definitive measures and full robustness results.

Assessment

Paper Typecorrelational Evidence Strengthmedium — Long panel (2003–2021) across Chinese provinces, multiple inequality metrics, and mediation analyses strengthen the associative evidence; however, absence of a clear exogenous source of variation in AI (or credible IV/DID design) leaves potential for omitted variable bias and reverse causality, limiting causal claims. Methods Rigormedium — Appropriate use of panel methods, fixed effects, alternative inequality measures, and mechanism tests indicates careful empirical work, but reliance on aggregated provincial proxies for AI and green innovation and lack of a convincing exogenous identification strategy weaken internal validity. SampleAnnual provincial-level panel for China covering 2003–2021 (≈30 provinces), with observations on provincial carbon emissions used to compute interprovincial Gini and Theil indices, measures of AI development (province-level AI activity/capacity proxy), measures of green innovation (e.g., patent or innovation activity concentrated in coastal provinces), and standard socioeconomic and energy-related controls drawn from Chinese statistical sources. Themesinequality innovation adoption IdentificationUses a provincial panel (2003–2021) and exploits within-province over-time variation in AI development in panel regressions with province and year fixed effects and standard controls; conducts robustness checks using alternative inequality measures (Gini and Theil) and explores mediation/interaction channels (energy efficiency, monitoring, resource allocation). No clearly described exogenous shock or instrumental variable is reported to deliver clean causal identification. GeneralizabilityFindings are China-specific and may not generalize to other institutional or energy contexts., Province-level aggregation masks within-province heterogeneity (city/firm/household-level dynamics)., AI development proxy may capture broader digitalization or economic development rather than AI-specific causal effects., Green-innovation measures (e.g., patents) are geographically concentrated and may not reflect effective diffusion or deployment in lagging regions., The 2003–2021 period may underrepresent the most recent rapid advances and diffusion of frontier generative AI since ~2019–2021.

Claims (7)

Claim	Direction	Confidence	Outcome	Details
AI development significantly reduces carbon inequality, particularly when measured by the Gini index. Inequality	negative	high	carbon inequality (Gini index)	0.3
The inequality-reducing impact of AI is weaker when carbon inequality is measured by the Theil index, implying persistent structural divides between advanced and less developed regions. Inequality	mixed	high	carbon inequality (Theil index)	0.3
AI reduces carbon inequality primarily through improved energy efficiency, enhanced environmental monitoring, and more efficient resource allocation, disproportionately benefiting lagging regions and narrowing interprovincial emission gaps. Inequality	negative	high	carbon inequality (interprovincial emission gaps)	0.3
Green innovation does not yet significantly reduce carbon inequality. Inequality	null_result	high	carbon inequality	0.3
Green innovation is concentrated in coastal provinces and has not effectively diffused to inland areas, limiting its ability to reduce regional carbon inequality. Adoption Rate	negative	high	geographic concentration of green innovation (diffusion to inland areas)	0.3
AI's disproportionate benefits for lagging regions help narrow interprovincial emission gaps. Inequality	negative	high	interprovincial emission gaps (carbon inequality)	0.3
Coordinated digital green development strategies are important to promote a more balanced and inclusive transition toward China’s dual-carbon goals. Governance And Regulation	positive	medium	balanced and inclusive transition to carbon peak and neutrality goals	0.03