The Carbon Reduction Effect of AI Policy: Quasi-Experimental Evidence from China's National AI Innovation Pilot Zones

The Carbon Reduction Effect of AI Policy: Quasi-Experimental Evidence from China's National AI Innovation Pilot Zones | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article The Carbon Reduction Effect of AI Policy: Quasi-Experimental Evidence from China's National AI Innovation Pilot Zones Nanxun Liu, Shuqing Wang, Yuanhong Peng This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7917762/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 10 You are reading this latest preprint version Abstract The global low-carbon transition necessitates innovative policy interventions. Using staggered difference-in-differences estimation on a panel of 282 Chinese cities (2010–2023), this study provides causal evidence that China's National AI Innovation Pilot Zones (AIPZ) policy significantly reduces urban carbon emissions by 6.3% on average. Spatial econometric models reveal substantial negative spillovers, inducing an additional 8.6% reduction in contiguous cities, leading to a total abatement effect of 14.3%. Mechanism and heterogeneity analyses show that industrial upgrading and green innovation are key channels, with effects pronounced in the Pearl River Delta and non-resource-based cities, but short-run rebound effects occur in resource-dependent areas. This study demonstrates demonstrate that AI policies generate carbon co-benefits, yet their efficacy depends critically on local industrial context and spatial linkages, underscoring the importance of regional coordination in climate governance. Our findings underscore the importance of integrating AI policies into regional climate strategies to maximize carbon co-benefits. Earth and environmental sciences/Environmental social sciences Scientific community and society/Geography Social science/Geography artificial intelligence carbon emissions pilot zone difference-in-differences spatial spillover urban China Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction The extensive consumption of fossil fuels since the Industrial Revolution has led to a rapid ascent in atmospheric CO 2 concentrations, pushing global warming toward the critical 1.5°C threshold 1 . Cities, as hubs of economic activity, account for over 70% of global anthropogenic CO 2 emissions, making urban decarbonization a critical leverage point 1 . This challenge is particularly acute in China, the world's largest emitter, where coal-dominated energy structures and carbon-intensive process industries persist despite ambitious national goals of carbon peaking by 2030 and neutrality by 2060 2 . While traditional policy instruments like emissions trading have achieved partial success 3 ,4 , the transformative potential of digital technologies, particularly artificial intelligence (AI), has only recently entered the mainstream climate policy agenda 5 . AI is theorized to reduce carbon emissions primarily through two interrelated channels: (i) accelerating green technology innovation that improves energy efficiency and unlocks low-carbon solutions, and (ii) enabling industrial structure upgrading that shifts value-added from carbon-intensive sectors to cleaner, high-tech industries 6 , 7 . However, robust empirical evidence disentangling these mechanisms at scale remains scarce. Most existing studies rely on sector-level case analyses or firm-level correlations, which struggle to address endogeneity concerns such as self-selection and reverse causality. Consequently, a fundamental question persists: can public AI policies generate measurable, scalable carbon co-benefits through green innovation and industrial upgrading, especially in emerging economies central to global emission trajectories? China's National New-Generation AI Innovation Pilot Zones (AIPZ) policy, launched in 2019, provides an ideal quasi-experimental setting to address this question 8 . This policy designates specific cities as testbeds, providing them with supportive infrastructure and incentives without direct carbon mandates, thereby creating a plausibly exogenous treatment gradient. Leveraging this staggered rollout and a panel of 282 Chinese cities (2010–2023), we employ a difference-in-differences (DID) strategy augmented by spatial econometrics to quantify the policy's causal effect and its geographical spillovers. Our study aims to make four key contributions. First, we provide causal evidence on the efficacy of an AI-targeted industrial policy in reducing urban carbon emissions. Second, we empirically test the theoretical mediating roles of industrial structure upgrading and green technology innovation. Third, we examine the heterogeneity of effects across regions and city types. Fourth, we quantify substantial spatial spillover effects of the policy, documenting 8.6% emission reductions in neighboring cities and revealing 40% underestimation by conventional non-spatial models—a dimension often overlooked in environmental policy evaluations. The remainder of this paper is structured as follows. Section 2 reviews the extant literature and develops our theoretical framework and hypotheses. Section 3 outlines the research design and data. Section 4 presents the empirical results and robustness checks. Section 5 delves into mechanism analysis, heterogeneity, and spatial spillovers. Section 6 concludes with policy implications and limitations. Literature Review Carbon mitigation represents a critical pillar of sustainable development. Scholarly inquiry in this domain has largely progressed along two distinct tracks: policy evaluation and the analysis of underlying drivers. On the policy front, instruments including carbon trading pilots 9 , low-carbon city programs 10 , 11 , and eco-civilization zones 12 have been widely studied. A consensus confirms that these regulations curb emissions by optimizing resource allocation, incentivizing technological progress, and promoting cleaner energy sources 13 . While this work provides a robust methodological foundation for assessing environmental regulation, it has yet to account for the specific effects of digital-technology policies centered on artificial intelligence. Concurrently, research on emission drivers spans economic, demographic, and technological dimensions 14 – 16 . For instance, economic expansion remains a primary driver, albeit moderated by energy structure and efficiency gains 17 , 18 . Although this body of work illuminates the complex drivers of emissions and offers context for understanding AI's potential channels of influence, it falls short of quantifying the impact of AI as a general-purpose technology. The net environmental impact of AI is, in fact, characterized by a fundamental duality. On one hand, AI holds promise for reducing emissions through enhanced energy efficiency, structural economic optimization, and green innovation 19 – 21 . On the other, the substantial computational demands of AI training and inference, coupled with the lifecycle resource consumption of its infrastructure, present significant environmental costs. To illustrate, the carbon footprint from training a single large-scale model can equal the lifetime emissions of five conventional automobiles 22 . Consequently, the net effect of AI is contingent upon model scale, energy structure, and regional context, and a clear consensus remains elusive. Critically, the current discourse is dominated by theoretical speculations and micro-level case studies, lacking systematic, macro-level assessment of targeted AI policies. It is particularly unclear how such policies might leverage channels like industrial restructuring and innovation, or how their effectiveness varies across regions. This gap underscores the pressing need for large-sample, quasi-experimental evidence. Grounding our study in an integrated theoretical framework spanning environmental governance 23 , industrial economics 24 , and innovation economics 25 , we posit that the AIPZ policy serves as an institutional intervention. It is designed to catalyze a structural shift through a dual strategy of cultivating new, low-carbon industries while optimizing existing ones 26 , 27 . Furthermore, it can address market failures in green innovation 28 and use AI itself to de-risk and accelerate the R&D process 29 . This leads to our core hypotheses: H1: The implementation of the AIPZ policy significantly reduces urban carbon emissions. H2: Industrial structure upgrading mediates the carbon reduction effect of the AIPZ policy. H3: Enhanced green technology innovation mediates the carbon reduction effect of the AIPZ policy. By leveraging the AIPZ rollout as a quasi-natural experiment, this study moves beyond estimating the net treatment effect to rigorously unpack the theoretical black box through mediation analysis. Methods Empirical Strategy Baseline Model To identify the causal effect of the AIPZ policy on urban CO 2 emissions, we leverage its staggered adoption across cities as a quasi-natural experiment. Our baseline specification is a two-way fixed effects (TWFE) difference-in-differences (DID) model 30 .The geographic distribution of the treatment (pilot) and control cities is visualized in Fig. 1, which illustrates the spatial assignment of the AIPZ policy across China. $$\:{\text{l}\text{n}\text{C}\text{O}}_{2\text{i}\text{t}}=\:{\alpha\:}\:+{{\beta\:}\text{A}\text{I}\text{P}\text{Z}}_{\text{i}\text{t}}\:+\:{{\gamma\:}\text{X}}_{\text{i}\text{t}}\:+{{\mu\:}}_{\text{i}}\:+\:{{\lambda\:}}_{\text{t}}\:+\:{{\epsilon\:}}_{\text{i}\text{t}}$$ 1 where i and t denote city and year, respectively, spanning the period 2010–2023. The dependent variable, lnCO 2it , is the natural logarithm of total CO 2 emissions for city i in year t. The variable of interest, AIPZ it , is a policy dummy that equals 1 for city i in year t and all subsequent years once it is designated as a pilot zone. Specifically, for cities approved in the first half of a year, the dummy switches to 1 in that same year; for those approved in the second half, it switches to 1 in the following year. The vector X it represents a set of time-varying city-level control variables. City fixed effects ( µ i ) and year fixed effects ( λ t ) are included to account for time-invariant city heterogeneity and common temporal shocks, respectively. The error term, ε it , is clustered at the city level to robustly address potential serial correlation. Mechanism Inspection To investigate the potential mediating roles of industrial structure upgrading and green technology innovation, we employ a causal steps approach 31 . This involves estimating the following two models: $$\:{M}_{it}={\alpha\:}_{1}+a\times\:{AIPZ}_{it}+\gamma\:{X}_{it}+{\mu\:}_{i}+{\lambda\:}_{t}+{\epsilon\:}_{it}$$ 2 $$\:{lnCO}_{2it}={\alpha\:}_{2}+c\times\:{AIPZ}_{it}+b\times\:{M}_{it}+\gamma\:{X}_{it}+{\mu\:}_{i}+{\lambda\:}_{t}+{\epsilon\:}_{it}$$ 3 Here, M it denotes the mediating variables. A statistically significant coefficient a in Eq. ( 2 ) indicates that the AIPZ policy significantly influences the proposed mediator—a prerequisite for a mediation effect. Eq. ( 3 ) then assesses the association between the mediator and CO 2 emissions while controlling for the policy itself. Evidence of mediation is established if both coefficients a and b are statistically significant. Spatial Spillovers To examine whether the AIPZ policy induces extra-local impacts (i.e., spatial spillovers), we estimate a Spatial Durbin Difference-in-Differences (SDID) model, grounded in the spatial econometrics literature 32 – 34 . The empirical specification is formulated as follows: $$\:{lnCO}_{2it}={\alpha\:}_{0}+{\rho\:W\bullet\:lnCO}_{2it}+{\varphi\:}_{1}{AIPZ}_{it}+{\varphi\:}_{2}W\bullet\:{AIPZ}_{it}+\theta\:{X}_{it}+{\mu\:}_{i}+{\lambda\:}_{t}+{\epsilon\:}_{it}$$ 4 where W denotes the row-standardized spatial weight matrix. To ensure the robustness of our findings, we employ three alternative specifications for W : (1) W 1 – Rook contiguity (common border); (2) W 2 – geographic distance (inverse great-circle distance); (3) W 3 –Queen contiguity (common border or vertex) The coefficient ρ captures endogenous spatial dependence, and the coefficient φ 1 quantifies the magnitude of spatial spillovers from the policy. Variable Construction Dependent variable The construction of the prefectural-city-level CO 2 emissions panel dataset (2010–2023) was based on a spatial reassignment of the Emissions Database for Global Atmospheric Research (EDGAR) 2024 greenhouse gas emission inventory (at a 0.1° × 0.1° grid resolution) 35 . The administrative boundaries for Chinese prefectural cities, obtained from a 2019 vector map, served as the spatial framework upon which the gridded emissions data were aggregated. An area-weighting methodology was employed to sum the emissions from all grid cells located within each respective city's boundary, resulting in annual city-level emission totals. The use of a fixed-year administrative map ensures consistency and comparability across the panel. Core explanatory variable The core explanatory variable, AIPZ it , is a time-varying policy dummy that identifies the implementation of the AIPZ policy in a given city and year. Its construction is based on the official approval dates of the 18 pilot zones announced by the Ministry of Science and Technology from 2019 to 2021 36 . The variable is assigned according to the following rule: It takes the value of 1 for a city starting from the year of approval if the approval was granted in the first half of the year. It switches to 1 from the year following the approval if the approval was granted in the second half of the year. For all years prior to these respective effective years, and for cities never approved, the variable is coded as 0. Control variables Following common practice in the related literature to mitigate potential reverse causality, all control variables are lagged by one period. We include the following time-varying city-level characteristics: Population Density ( PD ): Measured as the natural logarithm of the number of permanent residents per square kilometer of land area. Affluence ( AGDP ): Represented by the natural logarithm of real per-capita gross domestic product. Financial Depth ( FDL ): Calculated as the ratio of the total balance of bank loans to the nominal GDP. Urbanisation Level ( URB ): Defined as the share of the non-agricultural household-registered population in the total household-registered population. Openness ( OL ): proxied by the ratio of actually utilized foreign direct investment (FDI) to the nominal GDP. Mechanism Variables To uncover the potential channels through which the AIPZ policy influences urban carbon emissions, we focus on two mechanism variables: Green Technology Innovation ( GTI ): This variable is proxied by the number of green patent applications filed per 10,000 inhabitants. Green patents serve as a direct measure of a city's output in environmentally focused technological innovation. Industrial Structure Upgrading ( ISU ): We measure this using the ratio of the value-added of the tertiary sector (services) to that of the secondary sector (manufacturing and construction). A higher ratio indicates a more advanced economic structure that is less reliant on energy-intensive industrial production. Data Sources The data for the variables described above are assembled from multiple sources. The panel covers 282 prefecture-level cities over 2010–2023. The socioeconomic data for control and mechanism variables are compiled from the China City Statistical Yearbook, China Environmental Statistical Yearbook, China Regional Economic Statistical Yearbook, and China Statistical Yearbook for Regional Economy. All monetary variables are deflated to 2010 prices. Continuous variables are winsorised at the 1st and 99th percentiles to attenuate outlier influence. Empirical Results Baseline Estimates Table 1 reports the baseline difference-in-differences (DID) estimates of the impact of the National New-Generation Artificial Intelligence Innovation Pilot Zone (AIPZ) policy on urban CO 2 emissions. To examine the robustness of the estimates, we employ a strategy of progressively incorporating control variables across columns (1) to (6), with all models including both city and year fixed effects. Table 1 Benchmark regression results. Variables lnCO 2 lnCO 2 lnCO 2 lnCO 2 lnCO 2 lnCO 2 (1) (2) (3) (4) (5) (6) AIPZ − 0.050 *** (-2.84) -0.063 *** (-3.47) -0.061 *** (-3.37) -0.061 *** (-3.38) -0.062 *** (-3.46) -0.063 *** (-3.48) PD 0.133 *** (3.12) 0.121 *** (2.84) 0.122 *** (2.85) 0.110 *** (2.55) 0.108 *** (2.51) AGDP 0.045 *** (3.33) 0.048 *** (3.07) 0.049 *** (3.15) 0.052 *** (3.28) FDL 0.001 (0.29) 0.002 (0.45) 0.002 (0.45) URB -0.083 ** (-2.12) -0.081 ** (-2.06) OL -1.510(-1.05) Cons 16.984 *** (8743.89) 16.221 *** (66.26) 15.797 *** (57.34) 15.768 *** (53.57) 15.853 *** (53.39) 15.835 *** (53.24) City FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes R 2 0.9860 0.9860 0.9861 0.9861 0.9861 0.9861 Obs 3666 3666 3666 3666 3666 3666 Note: The dependent variable is the natural logarithm of city-level CO 2 emissions (lnCO 2 ). All models include city and year fixed effects. T-statistics based on standard errors clustered at the city level are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively. The results demonstrate that the estimated coefficient for the AIPZ policy is negative and statistically significant at the 1% level across all specifications. This finding indicates a consistent suppressive effect of the AI pilot zones on urban carbon emissions, which remains robust to the inclusion of various control variables. In the full model encompassing all controls (column 6), the coefficient for the AIPZ policy is -0.063. Given that the dependent variable is the natural logarithm of CO 2 emissions, this estimate implies that, on average, the establishment of an AI pilot zone led to a significant reduction of approximately 6.3% in CO 2 emissions for pilot cities compared to non-pilot cities, underscoring the substantial emission reduction potential of this policy. We further explore the dynamics of this effect by estimating a flexible specification that allows the treatment effect to vary by year relative to the policy adoption. The results reveal a pattern of escalating benefits: the emission reduction effect grows from 4.0% in the implementation year to 6.2%, 7.6%, and 9.3% in the first, second, and third year post-implementation, respectively. A joint test confirms the statistical significance of these dynamic post-treatment effects (F(4,281) = 4.43, p = 0.0017). This suggests that the carbon reduction capabilities of AI technologies strengthen over time, likely due to cumulative learning and gradual technological diffusion within the local economy. Parallel-Trend Validation To validate the key parallel trends assumption underlying our difference-in-differences design, we implement a dynamic event-study model. Figure 2 plots the estimated coefficients along with their 95% confidence intervals for four years before and after the implementation of the AIPZ policy, using the period immediately prior to the policy (t = − 1) as the reference. The empirical patterns provide strong support for the parallel trends assumption. As illustrated, all estimated coefficients for the pre-treatment periods (pre_4 to pre_1) are statistically indistinguishable from zero, indicating no systematic divergence in emission trends between the treatment and control groups before the policy intervention. Following the implementation of the AIPZ policy, the coefficients for post-treatment periods (post_1 to post_4) turn negative and become statistically significant, demonstrating a clear causal response. This dynamic path evidences a lagged yet persistent policy effect on urban carbon emission reduction. Robustness Checks Addressing Potential Bias in Staggered DID Design Recent econometric literature has highlighted that traditional two-way fixed effects (TWFE) estimators in staggered Difference-in-Differences (DID) designs may be biased if treatment effects are heterogeneous across cohorts and over time 37 . To ensure our baseline findings are not driven by this potential bias, we employ the Sun and Abraham (2021) estimator 38 . The results, presented in Column (1) of Table 2 , yield a coefficient of − 0.0635 (p < 0.01), which is remarkably close to our baseline estimate of − 0.063. This confirms that the core conclusion of a significant carbon reduction effect is robust to potential pitfalls in staggered DID estimation. Table 2 Other robustness tests. (1) Sun& braham (2) Excl. unicipalities (3) Winsor (4) Adj. window (5) psm-did AIPZ -0.064*** (0.0209) -0.067*** (-3.20) -0.065*** (-3.62) -0.052*** (-3.23) -0.063*** (-3.48) Controls YES YES YES YES YES City FE Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes R 2 0.9861 0.9854 0.9859 0.8492 0.8287 N 3666 3614 3666 2288 2309 Note:This table reports the coefficient of the AIPZ policy variable under different robustness checks. Column (1) employs the Sun & Abraham (2021) estimator to address potential bias in staggered difference-in-differences designs. Column (2) excludes direct-administered municipalities. Column (3) winsorizes all continuous variables at the 1st and 99th percentiles. Column (4) restricts the sample to a four-year window around the policy implementation. Column (5) uses Propensity Score Matching Difference-in-Differences (PSM-DID). All models include the full set of control variables, city fixed effects, and year fixed effects. Robust standard errors are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively. Placebo Test To rule out spurious correlation, we conducted a placebo test by randomly assigning "pseudo" pilot zone status 500 times and re-estimating Eq. ( 1 ). As depicted in Fig. 3, the distribution of the resulting placebo coefficients is tightly centered around zero, and its kernel density curve closely aligns with a normal distribution reference line centered at zero. The actual estimated coefficient from our baseline model (-0.063) falls far into the left tail of this placebo distribution, demonstrating that the observed emission reduction effect is highly unlikely to be driven by chance and confirming that our baseline result is attributable to the genuine policy intervention. Exclusion of Municipalities To address concerns that the direct-administered municipalities (Beijing, Shanghai, Tianjin, and Chongqing) might exert undue influence due to their unique administrative status and resource advantages, we excluded them from the sample. As presented in Column (2) of Table 2 , the re-estimated coefficient for the AIPZ policy is -0.067, which is statistically significant at the 1% level. This coefficient is marginally larger in magnitude than the baseline estimate, reinforcing that our primary finding is not driven by these atypical metropolitan entities. Winsorization To mitigate potential bias induced by outliers, all continuous variables were winsorized at the 1st and 99th percentiles. The result of this procedure, reported in Column (3) of Table 2 , yields an AIPZ coefficient of -0.065, virtually identical to our baseline estimate. This consistency indicates that our core results are robust to the influence of extreme values. Shortened Event Window We tested the sensitivity of our results to the chosen time frame by restricting the sample to a narrower event window of four years before and after the policy implementation (i.e., [-4, + 4]). The estimated coefficient within this window, shown in Column (4) of Table 2 , is -0.052 and remains statistically significant at the 1% level. This confirms that the identified negative effect of the policy is not sensitive to the length of the study period. Propensity Score Matching with DID (PSM-DID) To account for potential non-random selection into the treatment group, we employed a PSM-DID approach. Propensity scores were estimated using all pre-treatment covariates, and kernel matching was applied to construct a balanced control group. The DID estimate from the matched sample, presented in Column (5) of Table 2 , is -0.063, closely aligning with the baseline result. This robustness check effectively alleviates concerns regarding selection bias. Endogeneity Treatment To address potential endogeneity from reverse causality or omitted variables, we implement a two-stage least squares (2SLS) instrumental variable (IV) approach 39 . Our instrument exploits the interaction between a city's historical fixed-telephone density in 1984 and the post-policy period dummy (Phone1984 × Post). We posit that this historical communication endowment shaped a city's long-term technological trajectory, thereby influencing its contemporary suitability for AI adoption, while being plausibly exogenous to transient shocks in current CO 2 emissions given the substantial temporal distance. The first-stage results in Column (1) of Table 3 confirm a strong, statistically significant relationship between the instrument and the endogenous AIPZ variable. The first-stage F-statistic of 23.7 comfortably surpasses the Stock-Yogo critical value for weak instruments 40 , decisively rejecting weak-instrument concerns. The second-stage results in Column (2) yield an AIPZ coefficient of − 0.071 (p < 0.01), indicating a significant causal effect. The fact that this IV estimate is slightly larger in magnitude than its OLS counterpart further corroborates a robust negative causal impact and suggests that any attenuation bias in the OLS estimate is effectively corrected. Table 3 2SLS regression results. Variables First Stage Second Stage (1) AIPZ (2) lnCO 2 AIPZ -0.071*** (0.020) Iv 0.0002*** (0.00004) Controls Yes Yes City FE Yes Yes Time FE Yes Yes Constant -1.620** (0.634) 16.347*** (0.302) R 2 0.6342 0.0091 N 2592 2592 Note: The dependent variable is lnCO 2 The instrument is the interaction between a city's fixed-telephone density in 1984 and the post-policy period dummy (Phone1984 × Post). The first-stage F-statistic is 23.7, and the Kleibergen-Paap rk LM statistic is 20.4 (p = 0.000), comfortably rejecting the null of weak instruments. Robust standard errors are in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively. Mechanism Analysis To rigorously identify the causal mechanisms underlying the carbon reduction effects of the AIPZ policy, we employ a mediation analysis framework complemented by bootstrap testing with 5,000 replications 41 . The results, summarized in Table 4 , provide robust evidence for two distinct transmission channels. Table 4 Mediating effect test results. Variable Green Technology Innovation Industrial Structure Upgrading (1) GTI (2) lnCO 2 (3) ISU (4) lnCO 2 AIPZ 0.162 *** (0.016) -0.056 *** (0.018) 0.194 *** (0.035) -0.050 *** (0.018) GTI -0.047 ** (0.019) ISU -0.069 *** (0.009) Cons -0.286 (0.264) 15.821 *** (0.297) 7.439 *** (0.581) 16.347 *** (0.302) Controls Yes Yes Yes Yes City FE Yes Yes Yes Yes Time FE Yes Yes Yes Yes Proportion of indirect effect 11.83% 21.54% R2 0.687 0.986 0.880 0.987 OBS 3666 3666 3666 3666 Note: The dependent variable in columns (1) and (3) is the mediator (Green Technology Innovation and Industrial Structure Upgrading, respectively); in columns (2) and (4) it is lnCO₂. The significance of the indirect mediating effects is confirmed by bootstrap analysis with 5,000 replications. Robust standard errors are in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively. Green Technology Innovation: As shown in columns (1) and (2) of Table 4 , the AIPZ policy significantly enhances green patent intensity (GTI), with a coefficient of 0.162 (p < 0.01). The subsequent negative and statistically significant coefficient on GTI in the emission equation (β = -0.047, p < 0.05) indicates that green innovation serves as a meaningful pathway for emission abatement. The bootstrap analysis confirms the significance of this indirect effect (β = -0.00756, p = 0.014), with a 95% confidence interval of [-0.01359, -0.00152] excluding zero, accounting for approximately 12.0% of the total policy effect. Industrial Structure Upgrading: Columns (3) and (4) demonstrate that the policy fosters a transition toward a more service-oriented economy, significantly increasing the tertiary-to-secondary output ratio (ISU) by 0.194 (p < 0.01). This structural transformation is strongly associated with reduced emissions (β = -0.069, p < 0.01). Bootstrap testing provides robust validation for this mediating pathway, revealing a significant indirect effect (β = -0.01335, p < 0.001) with a 95% confidence interval of [-0.02043, -0.00628] excluding zero, mediating 21.2% of the total emission reduction effect. The consistent findings from both traditional mediation analysis and bootstrap testing substantiate that industrial structure upgrading and green technology innovation represent significant mechanisms through which the AIPZ policy achieves its carbon mitigation objectives. Further Discussion Heterogeneity Analysis To delineate the nuanced impacts of the AIPZ policy and uncover potential variations across different contexts, we conduct a heterogeneity analysis from two pivotal dimensions: geographic configuration (major urban agglomerations) and local economic structure (resource dependence). The results, synthesized in Fig. 4 and Fig. 5, reveal substantial disparities in the policy’s effectiveness, underscoring that the local context is a critical determinant of environmental outcomes. Heterogeneity across Major Urban Agglomerations Given the extensively documented regional disparities in China's economic development, industrial composition, and environmental enforcement 42 , we conduct a heterogeneity analysis by partitioning our sample based on the five major national-level urban agglomerations officially designated by the National Development and Reform Commission. This classification provides a robust and widely-adopted framework for analyzing regional policy impacts in China 43 . The DID estimates for each urban agglomeration are visually summarized in Fig. 4. The results unveil a pronounced heterogeneity in the policy's impact: Pearl River Delta (PRD): The AIPZ policy induced a statistically significant reduction in CO 2 emissions by 12.7% (β = -0.127, *p* < 0.01). This success is likely attributable to the PRD's mature, export-oriented economy, stringent environmental regulations, and a well-functioning carbon emissions trading scheme. These conditions collectively create a conducive ecosystem for the rapid integration of AI with advanced low-carbon technologies, thereby amplifying the carbon-saving effect. Chengdu-Chongqing(CC): In stark contrast, the policy resulted in a significant increase in CO 2 emissions of approximately 10.7% (β = 0.107, p < 0.01). This adverse short-term effect likely stems from the region's well-documented heavy reliance on traditional heavy industries and its role as a manufacturing hub in the interior 44 . In this context, initial AI adoption appears to be oriented towards enhancing production capacity and operational efficiency within these incumbent, carbon-intensive sectors, rather than facilitating energy-saving retrofits or a green transition. The concurrent underdeployment of complementary green technologies further exacerbates this issue. Beijing-Tianjin-Hebei (BTH), Yangtze River Delta (YRD), and Mid-Yangtze River (MYR): The estimated coefficients for these three agglomerations are statistically indistinguishable from zero. We propose two non-mutually exclusive explanations. First, substantial intra-cluster developmental heterogeneity—encompassing both highly developed mega-cities and less developed hinterland cities—may cancel out the average treatment effect when estimated for the agglomeration as a whole. Second, these more economically advanced regions might already be operating closer to the existing technological frontier in terms of emission efficiency for their current industrial mix, leaving limited scope for further immediate abatement from this specific policy intervention. Heterogeneity by Resource Dependence We further probe heterogeneity by categorizing cities into resource-based (RC) and non-resource-based (NRC) types, following the seminal official classification from China's Ministry of Natural Resources 45 . The regression results, visually summarized in Fig. 5, reveal a striking contrast that underscores the pivotal role of a city's initial industrial structure. Resource-based Cities (RC): The implementation of the AIPZ policy is associated with a significant 14.3% increase in CO 2 emissions (β = 0.143, p < 0.05; Fig. 5). This finding suggests that in cities specialized in carbon-intensive foundational sectors (e.g., mining, smelting, petrochemicals), early-stage AI application is primarily geared towards optimizing extraction, refining, and processing efficiency. Without concurrent stringent carbon constraints or proactive industrial diversification, this efficiency gain effectively lowers the cost and boosts the output of fossil fuel-based production, potentially inducing a short-run "rebound effect" that increases overall emissions. Non-Resource-based Cities (NRC): In stark contrast, non-resource cities experienced a significant 7.9% decrease in CO 2 emissions (β = -0.080, p < 0.01; Fig. 5). Their industrial base, characterized by lighter and more diversified manufacturing, high-tech industries, and services, is inherently more flexible and amenable to adopting AI for energy management, smart logistics, and precision emission monitoring. This structural advantage allows them to harness AI for direct carbon abatement and operational optimization, yielding more immediate and positive environmental dividends. Spatial Spillover Effects The quasi-experimental difference-in-differences (DID) design identifies the average treatment effect on the treated (ATT) but may fail to capture spillovers to neighboring regions. Given the geographic contiguity of cities and the propensity for knowledge diffusion in technological fields like AI 46 , the AIPZ policy might exhibit significant spatial externalities. To account for this, we estimate a Spatial Durbin Model (SDM) 47 , employing three alternative spatial weight matrices: a rook contiguity matrix (W1), an inverse-geodistance matrix (W2), and a queen contiguity matrix (W3). Model specification and selection tests Diagnostic tests confirm the presence of spatial dependence and justify our model choice. Moran's I test 48 soundly rejects the null hypothesis of spatial independence (p < 0.01). Furthermore, both the Lagrange Multiplier (LM) and robust LM tests for lag and error dependence 49 are statistically significant (p < 0.01). Critically, likelihood-ratio and Wald tests 50 reject the null that the SDM can be simplified to a spatial lag (SAR) or spatial error (SEM) model (p ≤ 0.05). Consequently, we adopt the SDM specification with city and year fixed effects. Results Table 5 reports the direct, indirect (spillover), and total effects derived from the Spatial Durbin Model. The AIPZ policy demonstrates robust direct effects across all specifications, reducing local carbon emissions by 5.7–5.8% (p < 0.01). Crucially, we identify substantial negative spatial spillovers under the rook contiguity matrix, with the policy reducing CO 2 emissions in geographically contiguous neighbors by 8.6% (p < 0.05). This translates to a total emission abatement effect of 14.3% when accounting for both direct and indirect impacts. Table 5 The results of the spatial spillover effect. Variable Rook Contiguity Matrix Geographic Distance Matrix Queen Contiguity Matrix lnCO 2 lnCO 2 lnCO 2 Direct -0.057*** (0.018) -0.058*** (0.018) -0.057*** (0.018) Indirect -0.086** (0.036) 0.343 (0.465) -0.086** (0.036) Total -0.143*** (0.042) 0.284 (0.465) -0.143*** (0.042) Spatial rho (ρ) 0.117*** -1.387*** 0.117*** Log-likelihood 2892.20 2875.33 2892.20 Controls Yes Yes yes City FE Yes Yes Yes Time FE Yes Yes Yes Obs. 3666 3666 3666 Notes: Direct, indirect and total effects from Spatial Durbin Models. The AIPZ policy reduces local emissions by 5.7–5.8% (direct effects) and generates 8.6% additional reductions in contiguous cities (spillover effects). All models include controls and two-way fixed effects. Standard errors in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. The spatial spillovers, however, exhibit notable heterogeneity across weight matrices. While statistically significant under both rook and queen contiguity (both showing − 8.6% spillover effects), the indirect effects become insignificant when using the geographic distance matrix, and even show a positive but statistically insignificant coefficient.This pattern indicates that the carbon reduction benefits of AI policies diffuse primarily through contiguous regional linkages rather than following simple distance decay or economic similarity channels. The finding reinforces the importance of geographic proximity in technology and knowledge diffusion 51 ,52 . In aggregate, conventional non-spatial models would underestimate the total carbon abatement benefits of the AIPZ policy by approximately 40%. This substantial underestimation highlights the critical importance of incorporating spatial dimensions in environmental policy evaluation, particularly for technology-oriented interventions with inherent cross-border externalities. Conclusions and Policy Implications This study constructs a theoretical framework linking general-purpose AI technologies to urban carbon mitigation and empirically evaluates the causal effect of China’s National New-Generation AI Innovation Pilot Zones (AIPZ) using a staggered difference-in-differences design combined with spatial econometric models. Analysis of a panel dataset covering 282 Chinese cities from 2010 to 2023 yields four principal findings. First, the AIPZ policy significantly reduces urban carbon emissions by an average of 6.3%. This effect remains robust across multiple sensitivity checks, including recent estimators for staggered DID designs, placebo tests, propensity score matching, instrumental variable estimation, and sample adjustments, indicating that the finding is not attributable to selection bias or unobserved confounding factors. Second, mediation analysis identifies two substantive transmission channels: industrial structure upgrading and green technology innovation. The former accounts for approximately 22% of the total emission reduction effect, while the latter explains about 11.8%. Together, these pathways fully capture the policy’s abatement mechanism, with no significant residual channels detected. Third, the policy’s impact exhibits considerable heterogeneity across regions and city types. Pronounced carbon reductions are observed in the Pearl River Delta (− 12.7%) and non-resource-based cities (− 7.9%). In contrast, short-term emission increases are identified in resource-dependent cities (+ 14.3%) and the Chengdu-Chongqing economic zone (+ 10.7%). These divergent outcomes appear closely tied to local industrial composition and the maturity of green supporting infrastructure. Fourth, spatial econometric estimates reveal substantial negative spillover effects, with contiguous cities under both rook and queen adjacency experiencing an additional 8.6% reduction in CO 2 emissions. Policy Implications The findings offer three targeted insights for policymakers seeking to align technological innovation with climate goals: Mainstream Carbon Reduction in Digital Policy Frameworks The strategic orientation of innovation policies should be expanded beyond economic productivity to explicitly incorporate carbon reduction as a measurable objective. Integrating verifiable emission targets into the performance evaluation and fiscal incentive systems for pilot zones could ensure that technological development directly supports China's "Dual Carbon" goals. Our results demonstrate that well-designed innovation policies can achieve total emission reductions of 14.3% when accounting for both direct and spatial effects. Strengthen Mechanism-Specific Policy Interventions Policy instruments should be refined to actively foster the identified mediating channels. Targeted measures could include linking resource allocations to green innovation outputs, designing fiscal incentives tailored to industrial servitization and intelligentization, and establishing green finance facilities conditioned on carbon performance metrics. The documented mediation effects—21.54% through industrial structure upgrading and 11.83% through green technology innovation—provide clear leverage points for policy design. Develop Differentiated and Coordinated Regional Implementation Strategies Policy implementation must account for regional disparities and leverage spatial interconnections. In technologically advanced eastern regions, policy should encourage applications in smart energy management and carbon accounting systems. In resource-based and industrial relocation zones in central and western China, policy should couple technological transformation with just transition measures, supporting energy efficiency retrofits alongside workforce reskilling programs. Critically, our spatial analysis reveals that inter-city coordination can amplify total carbon reduction benefits by approximately 40% compared to isolated implementations.* This underscores the importance of establishing regional innovation-carbon governance networks, cross-jurisdictional benefit-sharing mechanisms, and coordinated performance evaluations. Fostering open-source platforms for technological solutions and shared digital infrastructure can institutionalize the identified spatial spillovers, transforming local innovations into regional decarbonization outcomes. Limitations and Future Research Directions Notwithstanding its contributions, this study is subject to several limitations that warrant attention in future research. A primary limitation stems from the city-level analytical scale, which cannot fully uncover the micro-level mechanisms of firm-level AI adoption and adaptation. Future research combining firm-level surveys with in-depth case studies could provide finer-grained insights into corporate decision-making and implementation processes. The use of downscaled EDGAR emission data also presents a constraint, as it does not permit precise attribution of emission changes directly to AI activities.Integrating high-resolution, real-time emission monitoring data with establishment-level AI adoption metrics could substantially improve causal attribution in future analyses. Furthermore, while this study examines heterogeneity across industrial and regional dimensions, it does not systematically account for other potentially influential contextual factors such as the quality of digital infrastructure or the stringency of local environmental regulations. Subsequent studies could develop more comprehensive contingency frameworks to better understand how local conditions moderate policy effectiveness. Finally, the spatial analysis, while confirming the presence of spillovers, does not distinguish between specific diffusion channels such as knowledge spillovers, industrial linkage effects, or labor mobility. Constructing specialized spatial weight matrices based on patent citation networks, inter-firm supply chain data, or skilled labor flows could help disentangle the precise mechanisms underlying the observed spatial externalities. Declarations Data Availability Statement The gridded carbon emission data from the EDGAR inventory used in this study are publicly available from the EDGAR website at https://edgar.jrc.ec.europa.eu/. The socioeconomic data for Chinese cities were compiled from publicly available statistical yearbooks, including the China City Statistical Yearbook and the China Statistical Yearbook for Regional Economy. The city-level panel dataset of CO 2 emissions (2010-2023) generated in this study is available from the corresponding author on reasonable request. The full raw dataset is not publicly available due to copyright restrictions on the commercial statistical yearbooks but can be acquired by following the methodology described in the 'Data Sources' section. We strongly encourage future researchers to replicate the findings using the described methodology, as the core data sources are all public. Author contributions Nanxun Liu: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Writing - Original Draft, Writing - Review & Editing. Shuqing Wang: Data Curation, Software, Visualization, Writing - Review & Editing. Yuanhong Peng: Supervision, Project administration, Funding acquisition. Competing interests The authors declare no competing interests. Ethics approval This study used only publicly available, anonymized data at the city level and did not involve human participants or animals. Therefore, ethical approval was not required. References IPCC. Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change (Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 2021). https://doi.org/10.1017/9781009157896 State Council of the People's Republic of China. Action Plan for Carbon Peaking Before 2030 (Central People's Government of the People's Republic of China, 2021). https://www.gov.cn/zhengce/content/2021-10/26/content_5644984.htm Liu, Z. et al. Reduced carbon emission estimates from fossil fuel combustion and cement production in China. Nature 524 , 335–338 (2015). https://doi.org/10.1038/nature14677 Auffhammer, M. & Carson, R. T. Forecasting the path of China's CO2 emissions using province-level information. J. Environ. Econ. Manage. 55 , 229–247 (2008). https://doi.org/10.1016/j.jeem.2007.04.003 Riahi, K. et al. The Shared Socioeconomic Pathways and their energy, land use, and greenhouse gas emissions implications: An overview. Glob. Environ. Change 42 , 153–168 (2017). https://doi.org/10.1016/j.gloenvcha.2016.05.009 Rolnick, D. et al. Tackling climate change with machine learning. ACM Comput. Surv. 55 , 1–96 (2022).https://doi.org/10.1145/3485128 Vinuesa, R. et al. The role of artificial intelligence in achieving the Sustainable Development Goals. Nat. Commun. 11 , 233 (2020).https://doi.org/10.1038/s41467-019-14108-y Ministry of Science and Technology of the People's Republic of China. Guidelines for the Construction of the National New Generation Artificial Intelligence Innovation and Development Pilot Zone (2019). Dewaelheyns, N., Schoubben, F., Struyfs, K. & Van Hulle, C. The influence of carbon risk on firm value: Evidence from the European Union emission trading scheme. J. Environ. Manage. 344, 118293 (2023). https://doi.org/10.1016/j.jenvman.2023.118293 Lee, C. C., Feng, Y. & Peng, D. A green path towards sustainable development: The impact of low-carbon city pilot on energy transition. Energy Econ. 115, 106343 (2022). https://doi.org/10.1016/j.eneco.2022.106343 Calel, R. & Dechezleprêtre, A. Environmental Policy and Directed Technological Change: Evidence from the European Carbon Market. Rev. Econ. Stat. 98 , 173–191 (2016). https://doi.org/10.1162/REST_a_00470 Hübler, M. & Löschel, A. The EU decarbonisation roadmap 2050—what way to travel? An informational basis for the 'Energy Strategy 2050' debate. Energy Policy 55, 190–207 (2013). https://doi.org/10.1016/j.enpol.2012.11.038 Stavins, R. N. The future of US carbon-pricing policy. Energy J. 43, 1–22 (2022). https://doi.org/10.5547/01956574.43.1.rsta Ge, X., Liu, X. & Zhong, M. From aging to greener homes: Understanding the link between population aging and household carbon emissions in China. Environ. Impact Assess. Rev. 106, 107459 (2024). https://doi.org/10.1016/j.eiar.2024.107459 Shao, S., Fan, M. & Yang, L. Economic structure adjustment, green technological progress and China's low-carbon transition development. Manag. World 38 , 46–69 (2022). https://doi.org/10.19744/j.cnki.11-1235/f.2022.0056 Wang, Q. & Su, M. Drivers of decoupling economic growth from carbon emission — An empirical analysis of 192 countries using decoupling model and decomposition method. Environ. Impact Assess. Rev. 81 , 106356 (2020). https://doi.org/10.1016/j.eiar.2019.106356 Wang, Q. & Wang, L. Why does China's carbon intensity decline and India's carbon intensity increase? A decomposition analysis on the sectors. J. Clean. Prod. 265, 121569 (2020). https://doi.org/10.1016/j.jclepro.2020.121569 Zhong, Y., Zhao, H. & Yin, T. Unleashing the power of AI: How industrial intelligence impacts carbon productivity? Technol. Forecast. Soc. Change 188, 122327 (2023). https://doi.org/10.1016/j.techfore.2022.122327 Huang, Y., Li, J. & Chen, J. AI and green innovation: Evidence from Chinese firms. Technol. Forecast. Soc. Change 200 , 123181 (2024). https://doi.org/10.1016/j.techfore.2023.123181 Li, Z., Huang, Z. & Su, Y. How does artificial intelligence affect the carbon emissions of urban agglomerations? The role of green technology innovation and spatial spillover. J. Clean. Prod. 434, 140002 (2024). https://doi.org/10.1016/j.jclepro.2023.140002 Chen, J., Gao, M., Cheng, S., Hou, W., Song, M., Liu, X., Liu, Y. & Shan, Y. County-level CO2 emissions and sequestration in China during 1997–2017. Sci. Data 7 , 391 (2020). https://doi.org/10.1038/s41597-020-00736-3 Strubell, E., Ganesh, A. & McCallum, A. Energy and policy considerations for deep learning in NLP. in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics 3645–3653 (2020). https://doi.org/10.18653/v1/2020.acl-main.334 Cheng, Y. & Yao, X. Carbon intensity reduction assessment of renewable energy technology innovation in China: A panel data model with cross-section dependence and slope heterogeneity. Renew. Sust. Energ. R ev. 135, 110157 (2021). https://doi.org/10.1016/j.rser.2021.112052 Zhou, X., Zhang, J. & Li, J. Industrial structural transformation and carbon dioxide emissions in China. Energy Policy 57 , 43–51 (2013). https://doi.org/10.1016/j.enpol.2012.07.017 Bresnahan, T. F. & Trajtenberg, M. General purpose technologies: 'Engines of growth'? J. Econometrics 65 , 83–108 (1995). https://doi.org/10.1016/0304-4076(94)01598-T Brevini, B. Is AI Good for the Planet? (Polity Press, 2021). Geels, F. W. From sectoral systems of innovation to socio-technical systems: Insights about dynamics and change from sociology and institutional theory. Res. Policy 33, 897–920 (2004). https://doi.org/10.1016/j.respol.2004.01.015 Acemoglu, D., Aghion, P., Bursztyn, L. & Hemous, D. The environment and directed technical change. Am. Econ. Rev. 102 , 131–166 (2012). https://doi.org/10.1257/aer.102.1.131 Chen, J. & Lin, S. How does artificial intelligence drive the transformation and upgrading of industrial structure? Econ. Model. 133 , 106666 (2024). https://doi.org/10.1016/j.econmod.2024.106666 Dechezleprêtre, A. & Sato, M. The impacts of environmental regulations on competitiveness. Rev. Environ. Econ. Policy 11, 183–206 (2017). https://doi.org/10.1093/reep/rex013 Barbieri, N., Marzucchi, A. & Rizzo, U. Knowledge sources and impacts on the eco-innovation: Evidence from European firms. Res. Policy 49 , 104070 (2020). https://doi.org/10.1016/j.respol.2020.104070 Rennings, K. Redefining innovation—eco-innovation research and the contribution from ecological economics. Ecol. Econ. 32 , 319–332 (2000). https://doi.org/10.1016/S0921-8009(99)00112-3 Li, G. & Wang, X. Can artificial intelligence enhance green innovation efficiency? Evidence from China's manufacturing sector. J. Environ. Manage. 351, 119658 (2024). https://doi.org/10.1016/j.jenvman.2023.119658 Ferraris, A. et al. The role of AI in reducing uncertainty in innovation processes: A resource orchestration perspective. Technovation 132, 102981 (2024). https://doi.org/10.1016/j.technovation.2024.102981 Crippa, M. et al. CO2 emissions of all world countries 2024 Report. Publications Office of the European Union (2024). https://edgar.jrc.ec.europa.eu/report_2024 Ministry of Science and Technology of the People's Republic of China. Notice of the Ministry of Science and Technology on the Issuance of the 'Guidelines for the Construction of National New Generation Artificial Intelligence Innovation and Development Pilot Zones'. MoST (2019). de Chaisemartin, C. & D'Haultfœuille, X. Two-way fixed effects estimators with heterogeneous treatment effects. Am. Econ. Rev. 110 , 2964–2996 (2020). https://doi.org/10.1257/aer.20181169 Sun, L. & Abraham, S. Estimating dynamic treatment effects in event studies with heterogeneous treatment effects. J. Econom. 225, 175–199 (2021). https://doi.org/10.1016/j.jeconom.2020.09.006 Anselin, L. Spatial Econometrics: Methods and Models (Kluwer Academic Publishers, 1988). LeSage, J. P. & Pace, R. K. Introduction to Spatial Econometrics (CRC Press, 2009). Preacher, K. J. & Hayes, A. F. Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behav. Res. Methods 40, 879–891 (2008). https://doi.org/10.3758/BRM.40.3.879 Crippa, M. et al. GHG emissions of all world countries - 2023 Report. Publications Office of the European Union (2023). Fang, C. & Yu, D. Urban agglomeration: An evolving concept of an emerging phenomenon. Landsc. Urban Plan. 162 , 126–136 (2017). https://doi.org/10.1016/j.landurbplan.2017.02.014 Liu, J., Zhao, M. & Wang, Y. Impact of digital economy on urban green innovation: Evidence from China. J. Clean. Prod. 434 , 139996 (2024).https://doi.org/10.1016/j.jclepro.2024.139996 Aghion, P. et al. Carbon taxes, path dependency, and directed technical change: Evidence from the auto industry. J. Polit. Econ. 124 , 1–51 (2016).https://doi.org/10.1086/684581 Angrist, J. D. & Pischke, J.-S. Mostly Harmless Econometrics: An Empiricist's Companion (Princeton University Press, 2009). Duranton, G. & Turner, M. A. Urban growth and transportation. Rev. Econ. Stud. 79 , 1407–1440 (2012).https://doi.org/10.1093/restud/rdr042 Stock, J. H. & Yogo, M. Testing for weak instruments in linear IV regression. in Identification and Inference for Econometric Models: Essays in Honor of Thomas Rothenberg (eds Andrews, D. W. K. & Stock, J. H.) 80–108 (Cambridge University Press, 2005). https://doi.org/10.1017/CBO9780511614491.006 Jiang, Z. & Ding, P. Causal mediation analysis with latent mediators. Psychometrika 88, 1093–1115 (2023).https://doi.org/10.1007/s11336-023-09924-7 Conley, T. G., Hansen, C. B. & Rossi, P. E. Plausibly exogenous. Rev. Econ. Stat. 94, 260–272 (2012).https://doi.org/10.1162/REST_a_00139 Jaffe, A. B., Trajtenberg, M. & Henderson, R. Geographic localization of knowledge spillovers as evidenced by patent citations. Q. J. Econ. 108 , 577–598 (1993).https://doi.org/10.2307/2118401 Audretsch, D. B. & Feldman, M. P. R&D spillovers and the geography of innovation and production. Am. Econ. Rev. 86 , 630–640 (1996). Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 06 Jan, 2026 Reviews received at journal 04 Jan, 2026 Reviews received at journal 15 Dec, 2025 Reviewers agreed at journal 03 Dec, 2025 Reviewers agreed at journal 18 Nov, 2025 Reviewers invited by journal 17 Nov, 2025 Editor invited by journal 31 Oct, 2025 Editor assigned by journal 26 Oct, 2025 Submission checks completed at journal 26 Oct, 2025 First submitted to journal 21 Oct, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7917762","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":534001134,"identity":"91435eb1-4aa3-41b6-a5ae-762feefe8b02","order_by":0,"name":"Nanxun Liu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAyklEQVRIiWNgGAWjYPACCx4G9gYgXcDAIEGkFgkeBp4DDAwHDEjQAkQJRGoxZz97TJp3h4SMwc3Hx6Q/GNjISTYwP3x0A48Wy568ZGPeMxI8BrfT0iQOGKQZSzOwGRvn4NFicCDH8DFvG0hLjhlQy+HEeQw8bNJ4tZx/Y3AYrOXmGWK13IDZcoMHomU2YS1vjA3nArVInklLtjgD9ItkMyG/nAd64W2bjT3f8cMHb1RU2MhJHG9++BifFiyAmTTlo2AUjIJRMAqwAABauUPgRh/MkQAAAABJRU5ErkJggg==","orcid":"","institution":"Party School of Tongliang District Committee of C.P.C","correspondingAuthor":true,"prefix":"","firstName":"Nanxun","middleName":"","lastName":"Liu","suffix":""},{"id":534001135,"identity":"b59fc0c1-b489-4ad2-915a-1d413869ca97","order_by":1,"name":"Shuqing Wang","email":"","orcid":"","institution":"Party School of Tongliang District Committee of C.P.C","correspondingAuthor":false,"prefix":"","firstName":"Shuqing","middleName":"","lastName":"Wang","suffix":""},{"id":534001136,"identity":"319e1a22-01fa-4b4f-b456-715a82f33c04","order_by":2,"name":"Yuanhong Peng","email":"","orcid":"","institution":"Party School of Yongchuan District Committee of C.P.C","correspondingAuthor":false,"prefix":"","firstName":"Yuanhong","middleName":"","lastName":"Peng","suffix":""}],"badges":[],"createdAt":"2025-10-22 08:11:24","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7917762/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7917762/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":94273074,"identity":"48bfeede-f475-4323-b370-e9a763e38f90","added_by":"auto","created_at":"2025-10-26 07:01:10","extension":"jpg","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":123036,"visible":true,"origin":"","legend":"","description":"","filename":"Fig1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/b14985d94685091a6345e01f.jpg"},{"id":94272834,"identity":"e9368172-e44b-4102-ab0f-800d61cfa0ab","added_by":"auto","created_at":"2025-10-26 06:49:56","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":55971,"visible":true,"origin":"","legend":"","description":"","filename":"ManuscriptLiuetalScientificReports.docx","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/234f2c85df4bada10e20bf2f.docx"},{"id":94272838,"identity":"bc14b452-862f-4f07-835e-5ebcc2c6d70e","added_by":"auto","created_at":"2025-10-26 06:52:50","extension":"jpg","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":45126,"visible":true,"origin":"","legend":"","description":"","filename":"Fig2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/60e3f500301b2b3b7b3ce44b.jpg"},{"id":94272520,"identity":"7067f1ac-5a3c-4af7-8c9b-f5365549a01d","added_by":"auto","created_at":"2025-10-26 05:23:26","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":25289,"visible":true,"origin":"","legend":"","description":"","filename":"Table.docx","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/6124b7c56886ea6fee7c40e2.docx"},{"id":94272772,"identity":"1112e7ac-04d1-4c2a-b09e-fd5e92c5ca56","added_by":"auto","created_at":"2025-10-26 06:45:51","extension":"jpg","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":48915,"visible":true,"origin":"","legend":"","description":"","filename":"Fig3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/533052baff34527e370953d5.jpg"},{"id":94272748,"identity":"4e7f47f4-38df-417a-a3ef-dee41c490648","added_by":"auto","created_at":"2025-10-26 06:44:30","extension":"jpg","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":139692,"visible":true,"origin":"","legend":"","description":"","filename":"Fig4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/5c9a9c79a3c78b07519bcf8b.jpg"},{"id":94273086,"identity":"ead754c4-9593-4494-a06f-1a3ac5f3d2cd","added_by":"auto","created_at":"2025-10-26 07:01:56","extension":"jpg","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":134979,"visible":true,"origin":"","legend":"","description":"","filename":"Fig5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/d02a87f3ef6aa6860b811d63.jpg"},{"id":94272833,"identity":"df0841c9-12b3-42c1-8445-ad81936886b3","added_by":"auto","created_at":"2025-10-26 06:49:54","extension":"json","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5585,"visible":true,"origin":"","legend":"","description":"","filename":"acdc2179c7d24000a50111a0cfca2fec.json","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/c2924a7a452f878a5ecf441a.json"},{"id":94272690,"identity":"872f6489-3bd0-4577-a83a-ce0c86c0c9ec","added_by":"auto","created_at":"2025-10-26 06:19:17","extension":"xml","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":149564,"visible":true,"origin":"","legend":"","description":"","filename":"acdc2179c7d24000a50111a0cfca2fec1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/15fb5ab159626d5be2a34601.xml"},{"id":94272605,"identity":"716ff172-3a65-45b1-8452-4ed39d5ff2ac","added_by":"auto","created_at":"2025-10-26 06:02:10","extension":"jpg","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":123036,"visible":true,"origin":"","legend":"","description":"","filename":"Fig1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/92a9d9684f50e0b9f6381a2d.jpg"},{"id":94272746,"identity":"38360f56-a20a-48f2-a984-9665722354ab","added_by":"auto","created_at":"2025-10-26 06:44:30","extension":"jpg","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":45126,"visible":true,"origin":"","legend":"","description":"","filename":"Fig2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/6ed56a0d4afd8417f8509bfc.jpg"},{"id":94272569,"identity":"046c9399-b24b-45f1-b84f-fffd1e1633ea","added_by":"auto","created_at":"2025-10-26 05:47:34","extension":"jpg","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":48915,"visible":true,"origin":"","legend":"","description":"","filename":"Fig3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/7804eb9d06282d1c293fac56.jpg"},{"id":94273075,"identity":"27762cba-78f0-45ef-99c0-b90dbaeb1b12","added_by":"auto","created_at":"2025-10-26 07:01:22","extension":"jpg","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":139692,"visible":true,"origin":"","legend":"","description":"","filename":"Fig4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/58b30b41143db54e56c42904.jpg"},{"id":94272745,"identity":"44a54978-0123-49f7-83f4-53b0ed15d629","added_by":"auto","created_at":"2025-10-26 06:44:29","extension":"jpg","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":134979,"visible":true,"origin":"","legend":"","description":"","filename":"Fig5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/5bb12d7e84dedfea3caf4e26.jpg"},{"id":94272852,"identity":"4b9f8dc8-ccca-4a60-9c42-dc83dcc763f5","added_by":"auto","created_at":"2025-10-26 06:57:24","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":27879,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig1.png","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/848704c7de91f8a7d66a63a2.png"},{"id":94272994,"identity":"a4943162-6c41-4026-bd81-943c434d3f4f","added_by":"auto","created_at":"2025-10-26 06:59:53","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6563,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig2.png","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/d570a6169fbd954ee63730d4.png"},{"id":94272696,"identity":"6353051d-fd96-4463-9982-b8f8cabaa7dd","added_by":"auto","created_at":"2025-10-26 06:27:12","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":8053,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig3.png","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/a3bdd744b839b35c31783ccb.png"},{"id":94272521,"identity":"8fa4b6d5-5a4f-4bdf-8f86-fa2f0311dde0","added_by":"auto","created_at":"2025-10-26 05:23:27","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":41339,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig4.png","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/84702f2bf9cff9b7c0766e70.png"},{"id":94273077,"identity":"5ea4c4f5-6fb7-457b-8871-1ad0bc1b5901","added_by":"auto","created_at":"2025-10-26 07:01:23","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":31568,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFig5.png","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/1208d68608c74dfeae89a4ee.png"},{"id":94273079,"identity":"43be8711-1944-4863-a46e-cd87ecd51374","added_by":"auto","created_at":"2025-10-26 07:01:37","extension":"xml","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":149092,"visible":true,"origin":"","legend":"","description":"","filename":"acdc2179c7d24000a50111a0cfca2fec1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/3222a8219abd7b9a69e6a817.xml"},{"id":94272777,"identity":"af3a7815-99cc-4bf3-b926-31c1613856f7","added_by":"auto","created_at":"2025-10-26 06:48:28","extension":"html","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":161013,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/e70c474f43e0618c1bd761d4.html"},{"id":94272842,"identity":"e6284602-7f58-4ec0-b714-3abf95cea0b5","added_by":"auto","created_at":"2025-10-26 06:56:32","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":99500,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version\u003c/p\u003e","description":"","filename":"Fig1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/df377db6abbdfdacc3888e52.jpg"},{"id":94272695,"identity":"85e9f43e-6d77-4758-af38-0ed0903f28ff","added_by":"auto","created_at":"2025-10-26 06:26:26","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":45126,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version\u003c/p\u003e","description":"","filename":"Fig2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/a3d66dc1901fd76d9742f618.jpg"},{"id":94272691,"identity":"b3b7163d-537c-49eb-a1f0-9cda9ffde0dc","added_by":"auto","created_at":"2025-10-26 06:19:17","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":48915,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version\u003c/p\u003e","description":"","filename":"Fig3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/e3f2d04c72b9b78e9405e071.jpg"},{"id":94272679,"identity":"db866d0a-0752-4eaa-9d94-ed6bbe18690b","added_by":"auto","created_at":"2025-10-26 06:07:29","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":87054,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version\u003c/p\u003e","description":"","filename":"Fig4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/1efdfef8281c9f25a94e7db7.jpg"},{"id":94272855,"identity":"19d1ff08-c437-4521-b046-a203d8060bba","added_by":"auto","created_at":"2025-10-26 06:57:26","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":76352,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version\u003c/p\u003e","description":"","filename":"Fig5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/c6f8cf1ea0d66d569b1420bd.jpg"},{"id":94320312,"identity":"8dde9279-c0ee-4d7e-a67c-80e4689a7345","added_by":"auto","created_at":"2025-10-27 12:06:03","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1630204,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7917762/v1/4f17e261-958a-4589-a1c8-036572035ac8.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"The Carbon Reduction Effect of AI Policy: Quasi-Experimental Evidence from China's National AI Innovation Pilot Zones","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe extensive consumption of fossil fuels since the Industrial Revolution has led to a rapid ascent in atmospheric CO\u003csub\u003e2\u003c/sub\u003e concentrations, pushing global warming toward the critical 1.5\u0026deg;C threshold\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. Cities, as hubs of economic activity, account for over 70% of global anthropogenic CO\u003csub\u003e2\u003c/sub\u003e emissions, making urban decarbonization a critical leverage point\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. This challenge is particularly acute in China, the world's largest emitter, where coal-dominated energy structures and carbon-intensive process industries persist despite ambitious national goals of carbon peaking by 2030 and neutrality by 2060\u003csup\u003e2\u003c/sup\u003e. While traditional policy instruments like emissions trading have achieved partial success\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e,4\u003c/sup\u003e, the transformative potential of digital technologies, particularly artificial intelligence (AI), has only recently entered the mainstream climate policy agenda\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eAI is theorized to reduce carbon emissions primarily through two interrelated channels: (i) accelerating green technology innovation that improves energy efficiency and unlocks low-carbon solutions, and (ii) enabling industrial structure upgrading that shifts value-added from carbon-intensive sectors to cleaner, high-tech industries\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e6\u003c/span\u003e,\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. However, robust empirical evidence disentangling these mechanisms at scale remains scarce. Most existing studies rely on sector-level case analyses or firm-level correlations, which struggle to address endogeneity concerns such as self-selection and reverse causality. Consequently, a fundamental question persists: can public AI policies generate measurable, scalable carbon co-benefits through green innovation and industrial upgrading, especially in emerging economies central to global emission trajectories?\u003c/p\u003e\u003cp\u003eChina's National New-Generation AI Innovation Pilot Zones (AIPZ) policy, launched in 2019, provides an ideal quasi-experimental setting to address this question\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. This policy designates specific cities as testbeds, providing them with supportive infrastructure and incentives without direct carbon mandates, thereby creating a plausibly exogenous treatment gradient. Leveraging this staggered rollout and a panel of 282 Chinese cities (2010\u0026ndash;2023), we employ a difference-in-differences (DID) strategy augmented by spatial econometrics to quantify the policy's causal effect and its geographical spillovers.\u003c/p\u003e\u003cp\u003eOur study aims to make four key contributions. First, we provide causal evidence on the efficacy of an AI-targeted industrial policy in reducing urban carbon emissions. Second, we empirically test the theoretical mediating roles of industrial structure upgrading and green technology innovation. Third, we examine the heterogeneity of effects across regions and city types. Fourth, we quantify substantial spatial spillover effects of the policy, documenting 8.6% emission reductions in neighboring cities and revealing 40% underestimation by conventional non-spatial models\u0026mdash;a dimension often overlooked in environmental policy evaluations.\u003c/p\u003e\u003cp\u003eThe remainder of this paper is structured as follows. Section 2 reviews the extant literature and develops our theoretical framework and hypotheses. Section 3 outlines the research design and data. Section 4 presents the empirical results and robustness checks. Section 5 delves into mechanism analysis, heterogeneity, and spatial spillovers. Section 6 concludes with policy implications and limitations.\u003c/p\u003e"},{"header":"Literature Review","content":"\u003cp\u003eCarbon mitigation represents a critical pillar of sustainable development. Scholarly inquiry in this domain has largely progressed along two distinct tracks: policy evaluation and the analysis of underlying drivers.\u003c/p\u003e\u003cp\u003eOn the policy front, instruments including carbon trading pilots\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, low-carbon city programs\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e, and eco-civilization zones\u003csup\u003e12\u003c/sup\u003e have been widely studied. A consensus confirms that these regulations curb emissions by optimizing resource allocation, incentivizing technological progress, and promoting cleaner energy sources\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. While this work provides a robust methodological foundation for assessing environmental regulation, it has yet to account for the specific effects of digital-technology policies centered on artificial intelligence.\u003c/p\u003e\u003cp\u003eConcurrently, research on emission drivers spans economic, demographic, and technological dimensions\u003csup\u003e\u003cspan additionalcitationids=\"CR15\" citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e–\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. For instance, economic expansion remains a primary driver, albeit moderated by energy structure and efficiency gains\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. Although this body of work illuminates the complex drivers of emissions and offers context for understanding AI's potential channels of influence, it falls short of quantifying the impact of AI as a general-purpose technology.\u003c/p\u003e\u003cp\u003eThe net environmental impact of AI is, in fact, characterized by a fundamental duality. On one hand, AI holds promise for reducing emissions through enhanced energy efficiency, structural economic optimization, and green innovation\u003csup\u003e\u003cspan additionalcitationids=\"CR20\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e–\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. On the other, the substantial computational demands of AI training and inference, coupled with the lifecycle resource consumption of its infrastructure, present significant environmental costs. To illustrate, the carbon footprint from training a single large-scale model can equal the lifetime emissions of five conventional automobiles \u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. Consequently, the net effect of AI is contingent upon model scale, energy structure, and regional context, and a clear consensus remains elusive. Critically, the current discourse is dominated by theoretical speculations and micro-level case studies, lacking systematic, macro-level assessment of targeted AI policies. It is particularly unclear how such policies might leverage channels like industrial restructuring and innovation, or how their effectiveness varies across regions. This gap underscores the pressing need for large-sample, quasi-experimental evidence.\u003c/p\u003e\u003cp\u003eGrounding our study in an integrated theoretical framework spanning environmental governance\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e, industrial economics\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e, and innovation economics\u003csup\u003e25\u003c/sup\u003e, we posit that the AIPZ policy serves as an institutional intervention. It is designed to catalyze a structural shift through a dual strategy of cultivating new, low-carbon industries while optimizing existing ones\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e. Furthermore, it can address market failures in green innovation\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e and use AI itself to de-risk and accelerate the R\u0026amp;D process\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThis leads to our core hypotheses:\u003c/p\u003e\u003cp\u003eH1: The implementation of the AIPZ policy significantly reduces urban carbon emissions.\u003c/p\u003e\u003cp\u003eH2: Industrial structure upgrading mediates the carbon reduction effect of the AIPZ policy.\u003c/p\u003e\u003cp\u003eH3: Enhanced green technology innovation mediates the carbon reduction effect of the AIPZ policy.\u003c/p\u003e\u003cp\u003eBy leveraging the AIPZ rollout as a quasi-natural experiment, this study moves beyond estimating the net treatment effect to rigorously unpack the theoretical black box through mediation analysis.\u003c/p\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003cdiv id=\"Sec4\" class=\"Section3\"\u003e\u003cdiv id=\"Sec5\" class=\"Section4\"\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"Methods","content":"\u003ch2\u003eEmpirical Strategy\u003c/h2\u003e\u003ch2\u003eBaseline Model\u003c/h2\u003e\u003cp\u003eTo identify the causal effect of the AIPZ policy on urban CO\u003csub\u003e2\u003c/sub\u003e emissions, we leverage its staggered adoption across cities as a quasi-natural experiment. Our baseline specification is a two-way fixed effects (TWFE) difference-in-differences (DID) model\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e.The geographic distribution of the treatment (pilot) and control cities is visualized in Fig.\u0026nbsp;1, which illustrates the spatial assignment of the AIPZ policy across China.\u003c/p\u003e\u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e\n$$\\:{\\text{l}\\text{n}\\text{C}\\text{O}}_{2\\text{i}\\text{t}}=\\:{\\alpha\\:}\\:+{{\\beta\\:}\\text{A}\\text{I}\\text{P}\\text{Z}}_{\\text{i}\\text{t}}\\:+\\:{{\\gamma\\:}\\text{X}}_{\\text{i}\\text{t}}\\:+{{\\mu\\:}}_{\\text{i}}\\:+\\:{{\\lambda\\:}}_{\\text{t}}\\:+\\:{{\\epsilon\\:}}_{\\text{i}\\text{t}}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\u003c/div\u003e\u003cp\u003ewhere i and t denote city and year, respectively, spanning the period 2010–2023. The dependent variable, \u003cem\u003elnCO\u003c/em\u003e\u003csub\u003e\u003cem\u003e2it\u003c/em\u003e\u003c/sub\u003e, is the natural logarithm of total CO\u003csub\u003e2\u003c/sub\u003e emissions for city i in year t. The variable of interest, \u003cem\u003eAIPZ\u003c/em\u003e\u003csub\u003e\u003cem\u003eit\u003c/em\u003e\u003c/sub\u003e, is a policy dummy that equals 1 for city i in year t and all subsequent years once it is designated as a pilot zone. Specifically, for cities approved in the first half of a year, the dummy switches to 1 in that same year; for those approved in the second half, it switches to 1 in the following year. The vector \u003cem\u003eX\u003c/em\u003e\u003csub\u003e\u003cem\u003eit\u003c/em\u003e\u003c/sub\u003e represents a set of time-varying city-level control variables. City fixed effects (\u003cem\u003eµ\u003c/em\u003e\u003csub\u003e\u003cem\u003ei\u003c/em\u003e\u003c/sub\u003e) and year fixed effects (\u003cem\u003eλ\u003c/em\u003e\u003csub\u003e\u003cem\u003et\u003c/em\u003e\u003c/sub\u003e) are included to account for time-invariant city heterogeneity and common temporal shocks, respectively. The error term, \u003cem\u003eε\u003c/em\u003e\u003csub\u003e\u003cem\u003eit\u003c/em\u003e\u003c/sub\u003e, is clustered at the city level to robustly address potential serial correlation.\u003c/p\u003e\n\u003ch3\u003eMechanism Inspection\u003c/h3\u003e\n\u003cp\u003eTo investigate the potential mediating roles of industrial structure upgrading and green technology innovation, we employ a causal steps approach\u003csup\u003e31\u003c/sup\u003e. This involves estimating the following two models:\u003cdiv id=\"Equ2\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ2\" name=\"EquationSource\"\u003e\n$$\\:{M}_{it}={\\alpha\\:}_{1}+a\\times\\:{AIPZ}_{it}+\\gamma\\:{X}_{it}+{\\mu\\:}_{i}+{\\lambda\\:}_{t}+{\\epsilon\\:}_{it}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e2\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Equ3\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ3\" name=\"EquationSource\"\u003e\n$$\\:{lnCO}_{2it}={\\alpha\\:}_{2}+c\\times\\:{AIPZ}_{it}+b\\times\\:{M}_{it}+\\gamma\\:{X}_{it}+{\\mu\\:}_{i}+{\\lambda\\:}_{t}+{\\epsilon\\:}_{it}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e3\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eHere, \u003cem\u003eM\u003c/em\u003e\u003csub\u003e\u003cem\u003eit\u003c/em\u003e\u003c/sub\u003e denotes the mediating variables. A statistically significant coefficient a in Eq.\u0026nbsp;(\u003cspan refid=\"Equ2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) indicates that the AIPZ policy significantly influences the proposed mediator\u0026mdash;a prerequisite for a mediation effect. Eq.\u0026nbsp;(\u003cspan refid=\"Equ3\" class=\"InternalRef\"\u003e3\u003c/span\u003e) then assesses the association between the mediator and CO\u003csub\u003e2\u003c/sub\u003e emissions while controlling for the policy itself. Evidence of mediation is established if both coefficients a and b are statistically significant.\u003c/p\u003e\n\u003ch3\u003eSpatial Spillovers\u003c/h3\u003e\n\u003cp\u003eTo examine whether the AIPZ policy induces extra-local impacts (i.e., spatial spillovers), we estimate a Spatial Durbin Difference-in-Differences (SDID) model, grounded in the spatial econometrics literature\u003csup\u003e\u003cspan additionalcitationids=\"CR33\" citationid=\"CR31\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e. The empirical specification is formulated as follows:\u003cdiv id=\"Equ4\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ4\" name=\"EquationSource\"\u003e\n$$\\:{lnCO}_{2it}={\\alpha\\:}_{0}+{\\rho\\:W\\bullet\\:lnCO}_{2it}+{\\varphi\\:}_{1}{AIPZ}_{it}+{\\varphi\\:}_{2}W\\bullet\\:{AIPZ}_{it}+\\theta\\:{X}_{it}+{\\mu\\:}_{i}+{\\lambda\\:}_{t}+{\\epsilon\\:}_{it}$$\u003c/div\u003e\u003cdiv class=\"EquationNumber\"\u003e4\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003ewhere \u003cem\u003eW\u003c/em\u003e denotes the row-standardized spatial weight matrix. To ensure the robustness of our findings, we employ three alternative specifications for \u003cem\u003eW\u003c/em\u003e:\u003c/p\u003e\u003cp\u003e(1) W\u003csub\u003e1\u003c/sub\u003e \u0026ndash; Rook contiguity (common border);\u003c/p\u003e\u003cp\u003e(2) W\u003csub\u003e2\u003c/sub\u003e \u0026ndash; geographic distance (inverse great-circle distance);\u003c/p\u003e\u003cp\u003e(3) W\u003csub\u003e3\u003c/sub\u003e\u0026ndash;Queen contiguity (common border or vertex)\u003c/p\u003e\u003cp\u003eThe coefficient ρ captures endogenous spatial dependence, and the coefficient φ\u003csub\u003e1\u003c/sub\u003e quantifies the magnitude of spatial spillovers from the policy.\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eVariable Construction\u003c/h2\u003e\u003cdiv id=\"Sec9\" class=\"Section3\"\u003e\u003ch2\u003eDependent variable\u003c/h2\u003e\u003cp\u003eThe construction of the prefectural-city-level CO\u003csub\u003e2\u003c/sub\u003e emissions panel dataset (2010\u0026ndash;2023) was based on a spatial reassignment of the Emissions Database for Global Atmospheric Research (EDGAR) 2024 greenhouse gas emission inventory (at a 0.1\u0026deg; \u0026times; 0.1\u0026deg; grid resolution)\u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e. The administrative boundaries for Chinese prefectural cities, obtained from a 2019 vector map, served as the spatial framework upon which the gridded emissions data were aggregated. An area-weighting methodology was employed to sum the emissions from all grid cells located within each respective city's boundary, resulting in annual city-level emission totals. The use of a fixed-year administrative map ensures consistency and comparability across the panel.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003eCore explanatory variable\u003c/h2\u003e\u003cp\u003eThe core explanatory variable, \u003cem\u003eAIPZ\u003c/em\u003e\u003csub\u003e\u003cem\u003eit\u003c/em\u003e\u003c/sub\u003e, is a time-varying policy dummy that identifies the implementation of the AIPZ policy in a given city and year. Its construction is based on the official approval dates of the 18 pilot zones announced by the Ministry of Science and Technology from 2019 to 2021\u003csup\u003e36\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThe variable is assigned according to the following rule:\u003c/p\u003e\u003cp\u003eIt takes the value of 1 for a city starting from the year of approval if the approval was granted in the first half of the year.\u003c/p\u003e\u003cp\u003eIt switches to 1 from the year following the approval if the approval was granted in the second half of the year.\u003c/p\u003e\u003cp\u003eFor all years prior to these respective effective years, and for cities never approved, the variable is coded as 0.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eControl variables\u003c/h2\u003e\u003cp\u003eFollowing common practice in the related literature to mitigate potential reverse causality, all control variables are lagged by one period. We include the following time-varying city-level characteristics:\u003c/p\u003e\u003cp\u003ePopulation Density (\u003cem\u003ePD\u003c/em\u003e): Measured as the natural logarithm of the number of permanent residents per square kilometer of land area.\u003c/p\u003e\u003cp\u003eAffluence (\u003cem\u003eAGDP\u003c/em\u003e): Represented by the natural logarithm of real per-capita gross domestic product.\u003c/p\u003e\u003cp\u003eFinancial Depth (\u003cem\u003eFDL\u003c/em\u003e): Calculated as the ratio of the total balance of bank loans to the nominal GDP.\u003c/p\u003e\u003cp\u003eUrbanisation Level (\u003cem\u003eURB\u003c/em\u003e): Defined as the share of the non-agricultural household-registered population in the total household-registered population.\u003c/p\u003e\u003cp\u003eOpenness (\u003cem\u003eOL\u003c/em\u003e): proxied by the ratio of actually utilized foreign direct investment (FDI) to the nominal GDP.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003eMechanism Variables\u003c/h2\u003e\u003cp\u003eTo uncover the potential channels through which the AIPZ policy influences urban carbon emissions, we focus on two mechanism variables:\u003c/p\u003e\u003cp\u003eGreen Technology Innovation (\u003cem\u003eGTI\u003c/em\u003e): This variable is proxied by the number of green patent applications filed per 10,000 inhabitants. Green patents serve as a direct measure of a city's output in environmentally focused technological innovation.\u003c/p\u003e\u003cp\u003eIndustrial Structure Upgrading (\u003cem\u003eISU\u003c/em\u003e): We measure this using the ratio of the value-added of the tertiary sector (services) to that of the secondary sector (manufacturing and construction). A higher ratio indicates a more advanced economic structure that is less reliant on energy-intensive industrial production.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003eData Sources\u003c/h2\u003e\u003cp\u003eThe data for the variables described above are assembled from multiple sources. The panel covers 282 prefecture-level cities over 2010\u0026ndash;2023. The socioeconomic data for control and mechanism variables are compiled from the China City Statistical Yearbook, China Environmental Statistical Yearbook, China Regional Economic Statistical Yearbook, and China Statistical Yearbook for Regional Economy. All monetary variables are deflated to 2010 prices. Continuous variables are winsorised at the 1st and 99th percentiles to attenuate outlier influence.\u003c/p\u003e\u003cp\u003eEmpirical Results\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003eBaseline Estimates\u003c/h2\u003e\u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e reports the baseline difference-in-differences (DID) estimates of the impact of the National New-Generation Artificial Intelligence Innovation Pilot Zone (AIPZ) policy on urban CO\u003csub\u003e2\u003c/sub\u003e emissions. To examine the robustness of the estimates, we employ a strategy of progressively incorporating control variables across columns (1) to (6), with all models including both city and year fixed effects.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eBenchmark regression results.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVariables\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003elnCO\u003csub\u003e2\u003c/sub\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003elnCO\u003csub\u003e2\u003c/sub\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003elnCO\u003csub\u003e2\u003c/sub\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003elnCO\u003csub\u003e2\u003c/sub\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003elnCO\u003csub\u003e2\u003c/sub\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003elnCO\u003csub\u003e2\u003c/sub\u003e\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e(1)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e(2)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e(3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e(4)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e(5)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e(6)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAIPZ\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e\u0026minus;\u0026thinsp;0.050\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(-2.84)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-0.063\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(-3.47)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-0.061\u003csup\u003e***\u003c/sup\u003e(-3.37)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-0.061\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(-3.38)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-0.062\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(-3.46)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e-0.063\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(-3.48)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePD\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.133\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(3.12)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.121\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(2.84)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.122\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(2.85)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.110\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(2.55)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.108\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(2.51)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAGDP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.045\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(3.33)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.048\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(3.07)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.049\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(3.15)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.052\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(3.28)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFDL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.001\u003c/p\u003e\u003cp\u003e(0.29)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.002\u003c/p\u003e\u003cp\u003e(0.45)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.002\u003c/p\u003e\u003cp\u003e(0.45)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eURB\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-0.083\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(-2.12)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e-0.081\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(-2.06)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eOL\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e-1.510(-1.05)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCons\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e16.984\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(8743.89)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e16.221\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(66.26)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e15.797\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(57.34)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e15.768\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(53.57)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e15.853\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(53.39)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e15.835\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(53.24)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCity FE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTime FE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.9860\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.9860\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.9861\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.9861\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.9861\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e0.9861\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eObs\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3666\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3666\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3666\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e3666\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e3666\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c7\"\u003e\u003cp\u003e3666\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"7\"\u003eNote: The dependent variable is the natural logarithm of city-level CO\u003csub\u003e2\u003c/sub\u003e emissions (lnCO\u003csub\u003e2\u003c/sub\u003e). All models include city and year fixed effects. T-statistics based on standard errors clustered at the city level are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe results demonstrate that the estimated coefficient for the AIPZ policy is negative and statistically significant at the 1% level across all specifications. This finding indicates a consistent suppressive effect of the AI pilot zones on urban carbon emissions, which remains robust to the inclusion of various control variables. In the full model encompassing all controls (column 6), the coefficient for the AIPZ policy is -0.063. Given that the dependent variable is the natural logarithm of CO\u003csub\u003e2\u003c/sub\u003e emissions, this estimate implies that, on average, the establishment of an AI pilot zone led to a significant reduction of approximately 6.3% in CO\u003csub\u003e2\u003c/sub\u003e emissions for pilot cities compared to non-pilot cities, underscoring the substantial emission reduction potential of this policy.\u003c/p\u003e\u003cp\u003eWe further explore the dynamics of this effect by estimating a flexible specification that allows the treatment effect to vary by year relative to the policy adoption. The results reveal a pattern of escalating benefits: the emission reduction effect grows from 4.0% in the implementation year to 6.2%, 7.6%, and 9.3% in the first, second, and third year post-implementation, respectively. A joint test confirms the statistical significance of these dynamic post-treatment effects (F(4,281)\u0026thinsp;=\u0026thinsp;4.43, p\u0026thinsp;=\u0026thinsp;0.0017). This suggests that the carbon reduction capabilities of AI technologies strengthen over time, likely due to cumulative learning and gradual technological diffusion within the local economy.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003eParallel-Trend Validation\u003c/h2\u003e\u003cp\u003eTo validate the key parallel trends assumption underlying our difference-in-differences design, we implement a dynamic event-study model. Figure\u0026nbsp;2 plots the estimated coefficients along with their 95% confidence intervals for four years before and after the implementation of the AIPZ policy, using the period immediately prior to the policy (t = \u0026minus;\u0026thinsp;1) as the reference.\u003c/p\u003e\u003cp\u003eThe empirical patterns provide strong support for the parallel trends assumption. As illustrated, all estimated coefficients for the pre-treatment periods (pre_4 to pre_1) are statistically indistinguishable from zero, indicating no systematic divergence in emission trends between the treatment and control groups before the policy intervention. Following the implementation of the AIPZ policy, the coefficients for post-treatment periods (post_1 to post_4) turn negative and become statistically significant, demonstrating a clear causal response. This dynamic path evidences a lagged yet persistent policy effect on urban carbon emission reduction.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003ch2\u003eRobustness Checks\u003c/h2\u003e\u003cdiv id=\"Sec17\" class=\"Section3\"\u003e\u003ch2\u003eAddressing Potential Bias in Staggered DID Design\u003c/h2\u003e\u003cp\u003eRecent econometric literature has highlighted that traditional two-way fixed effects (TWFE) estimators in staggered Difference-in-Differences (DID) designs may be biased if treatment effects are heterogeneous across cohorts and over time\u003csup\u003e\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e. To ensure our baseline findings are not driven by this potential bias, we employ the Sun and Abraham (2021) estimator\u003csup\u003e38\u003c/sup\u003e. The results, presented in Column (1) of Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, yield a coefficient of \u0026minus;\u0026thinsp;0.0635 (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), which is remarkably close to our baseline estimate of \u0026minus;\u0026thinsp;0.063. This confirms that the core conclusion of a significant carbon reduction effect is robust to potential pitfalls in staggered DID estimation.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eOther robustness tests.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"6\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003e(1)\u003c/p\u003e\u003cp\u003eSun\u0026amp; braham\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e(2)\u003c/p\u003e\u003cp\u003eExcl. unicipalities\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003e(3)\u003c/p\u003e\u003cp\u003eWinsor\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003e(4)\u003c/p\u003e\u003cp\u003eAdj. window\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003e(5)\u003c/p\u003e\u003cp\u003epsm-did\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAIPZ\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-0.064*** (0.0209)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-0.067***\u003c/p\u003e\u003cp\u003e(-3.20)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-0.065***\u003c/p\u003e\u003cp\u003e(-3.62)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-0.052***\u003c/p\u003e\u003cp\u003e(-3.23)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e-0.063***\u003c/p\u003e\u003cp\u003e(-3.48)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eControls\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYES\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eYES\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eYES\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eYES\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eYES\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCity FE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTime FE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.9861\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.9854\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.9859\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.8492\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e0.8287\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3666\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3614\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3666\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e2288\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c6\"\u003e\u003cp\u003e2309\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"6\"\u003eNote:This table reports the coefficient of the AIPZ policy variable under different robustness checks. Column (1) employs the Sun \u0026amp; Abraham (2021) estimator to address potential bias in staggered difference-in-differences designs. Column (2) excludes direct-administered municipalities. Column (3) winsorizes all continuous variables at the 1st and 99th percentiles. Column (4) restricts the sample to a four-year window around the policy implementation. Column (5) uses Propensity Score Matching Difference-in-Differences (PSM-DID). All models include the full set of control variables, city fixed effects, and year fixed effects. Robust standard errors are reported in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\u003ch2\u003ePlacebo Test\u003c/h2\u003e\u003cp\u003eTo rule out spurious correlation, we conducted a placebo test by randomly assigning \"pseudo\" pilot zone status 500 times and re-estimating Eq.\u0026nbsp;(\u003cspan refid=\"Equ1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). As depicted in Fig.\u0026nbsp;3, the distribution of the resulting placebo coefficients is tightly centered around zero, and its kernel density curve closely aligns with a normal distribution reference line centered at zero. The actual estimated coefficient from our baseline model (-0.063) falls far into the left tail of this placebo distribution, demonstrating that the observed emission reduction effect is highly unlikely to be driven by chance and confirming that our baseline result is attributable to the genuine policy intervention.\u003c/p\u003e\u003cdiv id=\"Sec19\" class=\"Section3\"\u003e\u003ch2\u003eExclusion of Municipalities\u003c/h2\u003e\u003cp\u003eTo address concerns that the direct-administered municipalities (Beijing, Shanghai, Tianjin, and Chongqing) might exert undue influence due to their unique administrative status and resource advantages, we excluded them from the sample. As presented in Column (2) of Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, the re-estimated coefficient for the AIPZ policy is -0.067, which is statistically significant at the 1% level. This coefficient is marginally larger in magnitude than the baseline estimate, reinforcing that our primary finding is not driven by these atypical metropolitan entities.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec20\" class=\"Section2\"\u003e\u003ch2\u003eWinsorization\u003c/h2\u003e\u003cp\u003eTo mitigate potential bias induced by outliers, all continuous variables were winsorized at the 1st and 99th percentiles. The result of this procedure, reported in Column (3) of Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, yields an AIPZ coefficient of -0.065, virtually identical to our baseline estimate. This consistency indicates that our core results are robust to the influence of extreme values.\u003c/p\u003e\u003cdiv id=\"Sec21\" class=\"Section3\"\u003e\u003ch2\u003eShortened Event Window\u003c/h2\u003e\u003cp\u003eWe tested the sensitivity of our results to the chosen time frame by restricting the sample to a narrower event window of four years before and after the policy implementation (i.e., [-4, +\u0026thinsp;4]). The estimated coefficient within this window, shown in Column (4) of Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, is -0.052 and remains statistically significant at the 1% level. This confirms that the identified negative effect of the policy is not sensitive to the length of the study period.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec22\" class=\"Section3\"\u003e\u003ch2\u003ePropensity Score Matching with DID (PSM-DID)\u003c/h2\u003e\u003cp\u003eTo account for potential non-random selection into the treatment group, we employed a PSM-DID approach. Propensity scores were estimated using all pre-treatment covariates, and kernel matching was applied to construct a balanced control group. The DID estimate from the matched sample, presented in Column (5) of Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, is -0.063, closely aligning with the baseline result. This robustness check effectively alleviates concerns regarding selection bias.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec23\" class=\"Section3\"\u003e\u003ch2\u003eEndogeneity Treatment\u003c/h2\u003e\u003cp\u003eTo address potential endogeneity from reverse causality or omitted variables, we implement a two-stage least squares (2SLS) instrumental variable (IV) approach\u003csup\u003e\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e. Our instrument exploits the interaction between a city's historical fixed-telephone density in 1984 and the post-policy period dummy (Phone1984 \u0026times; Post). We posit that this historical communication endowment shaped a city's long-term technological trajectory, thereby influencing its contemporary suitability for AI adoption, while being plausibly exogenous to transient shocks in current CO\u003csub\u003e2\u003c/sub\u003e emissions given the substantial temporal distance. The first-stage results in Column (1) of Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e confirm a strong, statistically significant relationship between the instrument and the endogenous AIPZ variable. The first-stage F-statistic of 23.7 comfortably surpasses the Stock-Yogo critical value for weak instruments\u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e, decisively rejecting weak-instrument concerns. The second-stage results in Column (2) yield an AIPZ coefficient of \u0026minus;\u0026thinsp;0.071 (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), indicating a significant causal effect. The fact that this IV estimate is slightly larger in magnitude than its OLS counterpart further corroborates a robust negative causal impact and suggests that any attenuation bias in the OLS estimate is effectively corrected.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003e2SLS regression results.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"3\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVariables\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eFirst Stage\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSecond Stage\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e(1)\u003c/p\u003e\u003cp\u003eAIPZ\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e(2)\u003c/p\u003e\u003cp\u003elnCO\u003csub\u003e2\u003c/sub\u003e\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAIPZ\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-0.071***\u003c/p\u003e\u003cp\u003e(0.020)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eIv\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.0002***\u003c/p\u003e\u003cp\u003e(0.00004)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eControls\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCity FE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTime FE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eConstant\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-1.620**\u003c/p\u003e\u003cp\u003e(0.634)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e16.347***\u003c/p\u003e\u003cp\u003e(0.302)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.6342\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.0091\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2592\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2592\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"3\"\u003eNote: The dependent variable is lnCO\u003csub\u003e2\u003c/sub\u003e The instrument is the interaction between a city's fixed-telephone density in 1984 and the post-policy period dummy (Phone1984 \u0026times; Post). The first-stage F-statistic is 23.7, and the Kleibergen-Paap rk LM statistic is 20.4 (p\u0026thinsp;=\u0026thinsp;0.000), comfortably rejecting the null of weak instruments. Robust standard errors are in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec24\" class=\"Section2\"\u003e\u003ch2\u003eMechanism Analysis\u003c/h2\u003e\u003cp\u003eTo rigorously identify the causal mechanisms underlying the carbon reduction effects of the AIPZ policy, we employ a mediation analysis framework complemented by bootstrap testing with 5,000 replications\u003csup\u003e41\u003c/sup\u003e. The results, summarized in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, provide robust evidence for two distinct transmission channels.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eMediating effect test results.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eVariable\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e\u003cp\u003eGreen Technology Innovation\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e\u003cp\u003eIndustrial Structure Upgrading\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003e(1)\u003c/p\u003e\u003cp\u003eGTI\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e(2)\u003c/p\u003e\u003cp\u003elnCO\u003csub\u003e2\u003c/sub\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003e(3)\u003c/p\u003e\u003cp\u003eISU\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003e(4)\u003c/p\u003e\u003cp\u003elnCO\u003csub\u003e2\u003c/sub\u003e\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAIPZ\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.162\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(0.016)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-0.056\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(0.018)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.194\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(0.035)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-0.050\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(0.018)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGTI\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-0.047\u003csup\u003e**\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(0.019)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eISU\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e-0.069\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(0.009)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCons\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-0.286\u003c/p\u003e\u003cp\u003e(0.264)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e15.821\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(0.297)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e7.439\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(0.581)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e16.347\u003csup\u003e***\u003c/sup\u003e\u003c/p\u003e\u003cp\u003e(0.302)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eControls\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCity FE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTime FE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eProportion of indirect effect\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e11.83%\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e21.54%\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eR2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.687\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.986\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.880\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e0.987\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eOBS\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3666\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3666\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3666\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e3666\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"5\"\u003eNote: The dependent variable in columns (1) and (3) is the mediator (Green Technology Innovation and Industrial Structure Upgrading, respectively); in columns (2) and (4) it is lnCO₂. The significance of the indirect mediating effects is confirmed by bootstrap analysis with 5,000 replications. Robust standard errors are in parentheses. ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eGreen Technology Innovation: As shown in columns (1) and (2) of Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, the AIPZ policy significantly enhances green patent intensity (GTI), with a coefficient of 0.162 (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). The subsequent negative and statistically significant coefficient on GTI in the emission equation (β = -0.047, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05) indicates that green innovation serves as a meaningful pathway for emission abatement. The bootstrap analysis confirms the significance of this indirect effect (β = -0.00756, p\u0026thinsp;=\u0026thinsp;0.014), with a 95% confidence interval of [-0.01359, -0.00152] excluding zero, accounting for approximately 12.0% of the total policy effect.\u003c/p\u003e\u003cp\u003eIndustrial Structure Upgrading: Columns (3) and (4) demonstrate that the policy fosters a transition toward a more service-oriented economy, significantly increasing the tertiary-to-secondary output ratio (ISU) by 0.194 (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). This structural transformation is strongly associated with reduced emissions (β = -0.069, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). Bootstrap testing provides robust validation for this mediating pathway, revealing a significant indirect effect (β = -0.01335, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) with a 95% confidence interval of [-0.02043, -0.00628] excluding zero, mediating 21.2% of the total emission reduction effect.\u003c/p\u003e\u003cp\u003eThe consistent findings from both traditional mediation analysis and bootstrap testing substantiate that industrial structure upgrading and green technology innovation represent significant mechanisms through which the AIPZ policy achieves its carbon mitigation objectives.\u003c/p\u003e\u003cp\u003eFurther Discussion\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec25\" class=\"Section2\"\u003e\u003ch2\u003eHeterogeneity Analysis\u003c/h2\u003e\u003cp\u003eTo delineate the nuanced impacts of the AIPZ policy and uncover potential variations across different contexts, we conduct a heterogeneity analysis from two pivotal dimensions: geographic configuration (major urban agglomerations) and local economic structure (resource dependence). The results, synthesized in Fig.\u0026nbsp;4 and Fig.\u0026nbsp;5, reveal substantial disparities in the policy\u0026rsquo;s effectiveness, underscoring that the local context is a critical determinant of environmental outcomes.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eHeterogeneity across Major Urban Agglomerations\u003c/h3\u003e\n\u003cp\u003eGiven the extensively documented regional disparities in China's economic development, industrial composition, and environmental enforcement\u003csup\u003e\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e, we conduct a heterogeneity analysis by partitioning our sample based on the five major national-level urban agglomerations officially designated by the National Development and Reform Commission. This classification provides a robust and widely-adopted framework for analyzing regional policy impacts in China\u003csup\u003e\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e. The DID estimates for each urban agglomeration are visually summarized in Fig.\u0026nbsp;4.\u003c/p\u003e\u003cp\u003eThe results unveil a pronounced heterogeneity in the policy's impact:\u003c/p\u003e\u003cp\u003ePearl River Delta (PRD): The AIPZ policy induced a statistically significant reduction in CO\u003csub\u003e2\u003c/sub\u003e emissions by 12.7% (β = -0.127, *p* \u0026lt; 0.01). This success is likely attributable to the PRD's mature, export-oriented economy, stringent environmental regulations, and a well-functioning carbon emissions trading scheme. These conditions collectively create a conducive ecosystem for the rapid integration of AI with advanced low-carbon technologies, thereby amplifying the carbon-saving effect.\u003c/p\u003e\u003cp\u003eChengdu-Chongqing(CC): In stark contrast, the policy resulted in a significant increase in CO\u003csub\u003e2\u003c/sub\u003e emissions of approximately 10.7% (β\u0026thinsp;=\u0026thinsp;0.107, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). This adverse short-term effect likely stems from the region's well-documented heavy reliance on traditional heavy industries and its role as a manufacturing hub in the interior\u003csup\u003e44\u003c/sup\u003e. In this context, initial AI adoption appears to be oriented towards enhancing production capacity and operational efficiency within these incumbent, carbon-intensive sectors, rather than facilitating energy-saving retrofits or a green transition. The concurrent underdeployment of complementary green technologies further exacerbates this issue.\u003c/p\u003e\u003cp\u003eBeijing-Tianjin-Hebei (BTH), Yangtze River Delta (YRD), and Mid-Yangtze River (MYR): The estimated coefficients for these three agglomerations are statistically indistinguishable from zero. We propose two non-mutually exclusive explanations. First, substantial intra-cluster developmental heterogeneity\u0026mdash;encompassing both highly developed mega-cities and less developed hinterland cities\u0026mdash;may cancel out the average treatment effect when estimated for the agglomeration as a whole. Second, these more economically advanced regions might already be operating closer to the existing technological frontier in terms of emission efficiency for their current industrial mix, leaving limited scope for further immediate abatement from this specific policy intervention.\u003c/p\u003e\u003cdiv id=\"Sec27\" class=\"Section2\"\u003e\u003ch2\u003eHeterogeneity by Resource Dependence\u003c/h2\u003e\u003cp\u003eWe further probe heterogeneity by categorizing cities into resource-based (RC) and non-resource-based (NRC) types, following the seminal official classification from China's Ministry of Natural Resources\u003csup\u003e\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u003c/sup\u003e. The regression results, visually summarized in Fig.\u0026nbsp;5, reveal a striking contrast that underscores the pivotal role of a city's initial industrial structure.\u003c/p\u003e\u003cp\u003eResource-based Cities (RC): The implementation of the AIPZ policy is associated with a significant 14.3% increase in CO\u003csub\u003e2\u003c/sub\u003e emissions (β\u0026thinsp;=\u0026thinsp;0.143, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05; Fig.\u0026nbsp;5). This finding suggests that in cities specialized in carbon-intensive foundational sectors (e.g., mining, smelting, petrochemicals), early-stage AI application is primarily geared towards optimizing extraction, refining, and processing efficiency. Without concurrent stringent carbon constraints or proactive industrial diversification, this efficiency gain effectively lowers the cost and boosts the output of fossil fuel-based production, potentially inducing a short-run \"rebound effect\" that increases overall emissions.\u003c/p\u003e\u003cp\u003eNon-Resource-based Cities (NRC): In stark contrast, non-resource cities experienced a significant 7.9% decrease in CO\u003csub\u003e2\u003c/sub\u003e emissions (β = -0.080, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01; Fig.\u0026nbsp;5). Their industrial base, characterized by lighter and more diversified manufacturing, high-tech industries, and services, is inherently more flexible and amenable to adopting AI for energy management, smart logistics, and precision emission monitoring. This structural advantage allows them to harness AI for direct carbon abatement and operational optimization, yielding more immediate and positive environmental dividends.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec28\" class=\"Section2\"\u003e\u003ch2\u003eSpatial Spillover Effects\u003c/h2\u003e\u003cp\u003eThe quasi-experimental difference-in-differences (DID) design identifies the average treatment effect on the treated (ATT) but may fail to capture spillovers to neighboring regions. Given the geographic contiguity of cities and the propensity for knowledge diffusion in technological fields like AI\u003csup\u003e46\u003c/sup\u003e, the AIPZ policy might exhibit significant spatial externalities. To account for this, we estimate a Spatial Durbin Model (SDM)\u003csup\u003e\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u003c/sup\u003e, employing three alternative spatial weight matrices: a rook contiguity matrix (W1), an inverse-geodistance matrix (W2), and a queen contiguity matrix (W3).\u003c/p\u003e\u003cdiv id=\"Sec29\" class=\"Section3\"\u003e\u003ch2\u003eModel specification and selection tests\u003c/h2\u003e\u003cp\u003eDiagnostic tests confirm the presence of spatial dependence and justify our model choice. Moran's I test\u003csup\u003e\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e\u003c/sup\u003e soundly rejects the null hypothesis of spatial independence (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). Furthermore, both the Lagrange Multiplier (LM) and robust LM tests for lag and error dependence\u003csup\u003e\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e\u003c/sup\u003e are statistically significant (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). Critically, likelihood-ratio and Wald tests\u003csup\u003e\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e reject the null that the SDM can be simplified to a spatial lag (SAR) or spatial error (SEM) model (p\u0026thinsp;\u0026le;\u0026thinsp;0.05). Consequently, we adopt the SDM specification with city and year fixed effects.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e reports the direct, indirect (spillover), and total effects derived from the Spatial Durbin Model. The AIPZ policy demonstrates robust direct effects across all specifications, reducing local carbon emissions by 5.7\u0026ndash;5.8% (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). Crucially, we identify substantial negative spatial spillovers under the rook contiguity matrix, with the policy reducing CO\u003csub\u003e2\u003c/sub\u003e emissions in geographically contiguous neighbors by 8.6% (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). This translates to a total emission abatement effect of 14.3% when accounting for both direct and indirect impacts.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eThe results of the spatial spillover effect.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e\u003cp\u003eVariable\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRook Contiguity Matrix\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eGeographic Distance Matrix\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eQueen Contiguity Matrix\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003elnCO\u003csub\u003e2\u003c/sub\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003elnCO\u003csub\u003e2\u003c/sub\u003e\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003elnCO\u003csub\u003e2\u003c/sub\u003e\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDirect\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-0.057***\u003c/p\u003e\u003cp\u003e(0.018)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-0.058***\u003c/p\u003e\u003cp\u003e(0.018)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-0.057***\u003c/p\u003e\u003cp\u003e(0.018)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eIndirect\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-0.086**\u003c/p\u003e\u003cp\u003e(0.036)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.343\u003c/p\u003e\u003cp\u003e(0.465)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-0.086**\u003c/p\u003e\u003cp\u003e(0.036)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTotal\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e-0.143***\u003c/p\u003e\u003cp\u003e(0.042)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.284\u003c/p\u003e\u003cp\u003e(0.465)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-0.143***\u003c/p\u003e\u003cp\u003e(0.042)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSpatial rho (ρ)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.117***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-1.387***\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.117***\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLog-likelihood\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e2892.20\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2875.33\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e2892.20\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eControls\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eyes\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCity FE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eTime FE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eYes\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eObs.\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3666\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e3666\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e3666\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"4\"\u003eNotes: Direct, indirect and total effects from Spatial Durbin Models. The AIPZ policy reduces local emissions by 5.7\u0026ndash;5.8% (direct effects) and generates 8.6% additional reductions in contiguous cities (spillover effects). All models include controls and two-way fixed effects. Standard errors in parentheses. *** p\u0026thinsp;\u0026lt;\u0026thinsp;0.01, ** p\u0026thinsp;\u0026lt;\u0026thinsp;0.05, * p\u0026thinsp;\u0026lt;\u0026thinsp;0.1.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe spatial spillovers, however, exhibit notable heterogeneity across weight matrices. While statistically significant under both rook and queen contiguity (both showing \u0026minus;\u0026thinsp;8.6% spillover effects), the indirect effects become insignificant when using the geographic distance matrix, and even show a positive but statistically insignificant coefficient.This pattern indicates that the carbon reduction benefits of AI policies diffuse primarily through contiguous regional linkages rather than following simple distance decay or economic similarity channels. The finding reinforces the importance of geographic proximity in technology and knowledge diffusion\u003csup\u003e\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e,52\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eIn aggregate, conventional non-spatial models would underestimate the total carbon abatement benefits of the AIPZ policy by approximately 40%. This substantial underestimation highlights the critical importance of incorporating spatial dimensions in environmental policy evaluation, particularly for technology-oriented interventions with inherent cross-border externalities.\u003c/p\u003e"},{"header":"Conclusions and Policy Implications","content":"\u003cp\u003eThis study constructs a theoretical framework linking general-purpose AI technologies to urban carbon mitigation and empirically evaluates the causal effect of China\u0026rsquo;s National New-Generation AI Innovation Pilot Zones (AIPZ) using a staggered difference-in-differences design combined with spatial econometric models. Analysis of a panel dataset covering 282 Chinese cities from 2010 to 2023 yields four principal findings.\u003c/p\u003e\u003cp\u003eFirst, the AIPZ policy significantly reduces urban carbon emissions by an average of 6.3%. This effect remains robust across multiple sensitivity checks, including recent estimators for staggered DID designs, placebo tests, propensity score matching, instrumental variable estimation, and sample adjustments, indicating that the finding is not attributable to selection bias or unobserved confounding factors.\u003c/p\u003e\u003cp\u003eSecond, mediation analysis identifies two substantive transmission channels: industrial structure upgrading and green technology innovation. The former accounts for approximately 22% of the total emission reduction effect, while the latter explains about 11.8%. Together, these pathways fully capture the policy\u0026rsquo;s abatement mechanism, with no significant residual channels detected.\u003c/p\u003e\u003cp\u003eThird, the policy\u0026rsquo;s impact exhibits considerable heterogeneity across regions and city types. Pronounced carbon reductions are observed in the Pearl River Delta (\u0026minus;\u0026thinsp;12.7%) and non-resource-based cities (\u0026minus;\u0026thinsp;7.9%). In contrast, short-term emission increases are identified in resource-dependent cities (+\u0026thinsp;14.3%) and the Chengdu-Chongqing economic zone (+\u0026thinsp;10.7%). These divergent outcomes appear closely tied to local industrial composition and the maturity of green supporting infrastructure.\u003c/p\u003e\u003cp\u003eFourth, spatial econometric estimates reveal substantial negative spillover effects, with contiguous cities under both rook and queen adjacency experiencing an additional 8.6% reduction in CO\u003csub\u003e2\u003c/sub\u003e emissions.\u003c/p\u003e\n\u003ch3\u003ePolicy Implications\u003c/h3\u003e\n\u003cp\u003eThe findings offer three targeted insights for policymakers seeking to align technological innovation with climate goals:\u003c/p\u003e\u003cdiv id=\"Sec33\" class=\"Section2\"\u003e\u003ch2\u003eMainstream Carbon Reduction in Digital Policy Frameworks\u003c/h2\u003e\u003cp\u003eThe strategic orientation of innovation policies should be expanded beyond economic productivity to explicitly incorporate carbon reduction as a measurable objective. Integrating verifiable emission targets into the performance evaluation and fiscal incentive systems for pilot zones could ensure that technological development directly supports China's \"Dual Carbon\" goals. Our results demonstrate that well-designed innovation policies can achieve total emission reductions of 14.3% when accounting for both direct and spatial effects.\u003c/p\u003e\u003cdiv id=\"Sec34\" class=\"Section3\"\u003e\u003ch2\u003eStrengthen Mechanism-Specific Policy Interventions\u003c/h2\u003e\u003cp\u003ePolicy instruments should be refined to actively foster the identified mediating channels. Targeted measures could include linking resource allocations to green innovation outputs, designing fiscal incentives tailored to industrial servitization and intelligentization, and establishing green finance facilities conditioned on carbon performance metrics. The documented mediation effects\u0026mdash;21.54% through industrial structure upgrading and 11.83% through green technology innovation\u0026mdash;provide clear leverage points for policy design.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec35\" class=\"Section2\"\u003e\u003ch2\u003eDevelop Differentiated and Coordinated Regional Implementation Strategies\u003c/h2\u003e\u003cp\u003ePolicy implementation must account for regional disparities and leverage spatial interconnections. In technologically advanced eastern regions, policy should encourage applications in smart energy management and carbon accounting systems. In resource-based and industrial relocation zones in central and western China, policy should couple technological transformation with just transition measures, supporting energy efficiency retrofits alongside workforce reskilling programs.\u003c/p\u003e\u003cp\u003eCritically, our spatial analysis reveals that inter-city coordination can amplify total carbon reduction benefits by approximately 40% compared to isolated implementations.* This underscores the importance of establishing regional innovation-carbon governance networks, cross-jurisdictional benefit-sharing mechanisms, and coordinated performance evaluations. Fostering open-source platforms for technological solutions and shared digital infrastructure can institutionalize the identified spatial spillovers, transforming local innovations into regional decarbonization outcomes.\u003c/p\u003e\u003cp\u003eLimitations and Future Research Directions\u003c/p\u003e\u003cp\u003eNotwithstanding its contributions, this study is subject to several limitations that warrant attention in future research.\u003c/p\u003e\u003cp\u003eA primary limitation stems from the city-level analytical scale, which cannot fully uncover the micro-level mechanisms of firm-level AI adoption and adaptation. Future research combining firm-level surveys with in-depth case studies could provide finer-grained insights into corporate decision-making and implementation processes.\u003c/p\u003e\u003cp\u003eThe use of downscaled EDGAR emission data also presents a constraint, as it does not permit precise attribution of emission changes directly to AI activities.Integrating high-resolution, real-time emission monitoring data with establishment-level AI adoption metrics could substantially improve causal attribution in future analyses.\u003c/p\u003e\u003cp\u003eFurthermore, while this study examines heterogeneity across industrial and regional dimensions, it does not systematically account for other potentially influential contextual factors such as the quality of digital infrastructure or the stringency of local environmental regulations. Subsequent studies could develop more comprehensive contingency frameworks to better understand how local conditions moderate policy effectiveness.\u003c/p\u003e\u003cp\u003eFinally, the spatial analysis, while confirming the presence of spillovers, does not distinguish between specific diffusion channels such as knowledge spillovers, industrial linkage effects, or labor mobility. Constructing specialized spatial weight matrices based on patent citation networks, inter-firm supply chain data, or skilled labor flows could help disentangle the precise mechanisms underlying the observed spatial externalities.\u003c/p\u003e\u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003eData Availability Statement\u003c/p\u003e\n\u003cp\u003eThe gridded carbon emission data from the EDGAR inventory used in this study are publicly available from the EDGAR website at https://edgar.jrc.ec.europa.eu/. The socioeconomic data for Chinese cities were compiled from publicly available statistical yearbooks, including the China City Statistical Yearbook and the China Statistical Yearbook for Regional Economy.\u003c/p\u003e\n\u003cp\u003eThe city-level panel dataset of CO\u003csub\u003e2\u0026nbsp;\u003c/sub\u003eemissions (2010-2023) generated in this study is available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003eThe full raw dataset is not publicly available due to copyright restrictions on the commercial statistical yearbooks but can be acquired by following the methodology described in the \u0026apos;Data Sources\u0026apos; section. We strongly encourage future researchers to replicate the findings using the described methodology, as the core data sources are all public.\u003c/p\u003e\n\u003cp\u003eAuthor contributions\u003c/p\u003e\n\u003cp\u003eNanxun Liu:\u0026nbsp;Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Writing - Original Draft, Writing - Review \u0026amp; Editing.\u0026nbsp;Shuqing Wang:\u0026nbsp;Data Curation, Software, Visualization, Writing - Review \u0026amp; Editing.\u0026nbsp;Yuanhong Peng:\u0026nbsp;Supervision, Project administration, Funding acquisition.\u003c/p\u003e\n\u003cp\u003eCompeting interests\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003eEthics approval\u003c/p\u003e\n\u003cp\u003eThis study used only publicly available, anonymized data at the city level and did not involve human participants or animals. Therefore, ethical approval was not required.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eIPCC. \u003cem\u003eClimate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change\u003c/em\u003e (Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 2021). https://doi.org/10.1017/9781009157896\u003c/li\u003e\n\u003cli\u003eState Council of the People\u0026apos;s Republic of China. \u003cem\u003eAction Plan for Carbon Peaking Before 2030\u003c/em\u003e (Central People\u0026apos;s Government of the People\u0026apos;s Republic of China, 2021).\u003cbr\u003e https://www.gov.cn/zhengce/content/2021-10/26/content_5644984.htm\u003c/li\u003e\n\u003cli\u003eLiu, Z. et al. Reduced carbon emission estimates from fossil fuel combustion and cement production in China. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e524\u003c/strong\u003e, 335\u0026ndash;338 (2015).\u003cbr\u003e https://doi.org/10.1038/nature14677\u003c/li\u003e\n\u003cli\u003eAuffhammer, M. \u0026amp; Carson, R. T. Forecasting the path of China\u0026apos;s CO2 emissions using province-level information. \u003cem\u003eJ. Environ. Econ. Manage.\u003c/em\u003e \u003cstrong\u003e55\u003c/strong\u003e, 229\u0026ndash;247 (2008).\u003cbr\u003e https://doi.org/10.1016/j.jeem.2007.04.003\u003c/li\u003e\n\u003cli\u003eRiahi, K. et al. The Shared Socioeconomic Pathways and their energy, land use, and greenhouse gas emissions implications: An overview. \u003cem\u003eGlob. Environ. Change\u003c/em\u003e \u003cstrong\u003e42\u003c/strong\u003e, 153\u0026ndash;168 (2017).\u003cbr\u003e https://doi.org/10.1016/j.gloenvcha.2016.05.009\u003c/li\u003e\n\u003cli\u003eRolnick, D. et al. Tackling climate change with machine learning. \u003cem\u003eACM Comput. Surv.\u003c/em\u003e \u003cstrong\u003e55\u003c/strong\u003e, 1\u0026ndash;96 (2022).https://doi.org/10.1145/3485128\u003c/li\u003e\n\u003cli\u003eVinuesa, R. et al. The role of artificial intelligence in achieving the Sustainable Development Goals. \u003cem\u003eNat. Commun.\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, 233 (2020).https://doi.org/10.1038/s41467-019-14108-y\u003c/li\u003e\n\u003cli\u003eMinistry of Science and Technology of the People\u0026apos;s Republic of China. Guidelines for the Construction of the National New Generation Artificial Intelligence Innovation and Development Pilot Zone (2019).\u003c/li\u003e\n\u003cli\u003eDewaelheyns, N., Schoubben, F., Struyfs, K. \u0026amp; Van Hulle, C. The influence of carbon risk on firm value: Evidence from the European Union emission trading scheme. \u003cem\u003eJ. Environ. Manage.\u003c/em\u003e 344, 118293 (2023).\u003cbr\u003e https://doi.org/10.1016/j.jenvman.2023.118293\u003c/li\u003e\n\u003cli\u003eLee, C. C., Feng, Y. \u0026amp; Peng, D. A green path towards sustainable development: The impact of low-carbon city pilot on energy transition. Energy Econ. 115, 106343 (2022).\u003cbr\u003e https://doi.org/10.1016/j.eneco.2022.106343\u003c/li\u003e\n\u003cli\u003eCalel, R. \u0026amp; Dechezlepr\u0026ecirc;tre, A. Environmental Policy and Directed Technological Change: Evidence from the European Carbon Market. \u003cem\u003eRev. Econ. Stat.\u003c/em\u003e \u003cstrong\u003e98\u003c/strong\u003e, 173\u0026ndash;191 (2016).\u003cbr\u003e https://doi.org/10.1162/REST_a_00470\u003c/li\u003e\n\u003cli\u003eH\u0026uuml;bler, M. \u0026amp; L\u0026ouml;schel, A. The EU decarbonisation roadmap 2050\u0026mdash;what way to travel? An informational basis for the \u0026apos;Energy Strategy 2050\u0026apos; debate. Energy Policy 55, 190\u0026ndash;207 (2013).\u003cbr\u003e https://doi.org/10.1016/j.enpol.2012.11.038\u003c/li\u003e\n\u003cli\u003eStavins, R. N. The future of US carbon-pricing policy. Energy J. 43, 1\u0026ndash;22 (2022).\u003cbr\u003e https://doi.org/10.5547/01956574.43.1.rsta\u003c/li\u003e\n\u003cli\u003eGe, X., Liu, X. \u0026amp; Zhong, M. From aging to greener homes: Understanding the link between population aging and household carbon emissions in China. \u003cem\u003eEnviron. Impact Assess. Rev.\u003c/em\u003e 106, 107459 (2024).\u003cbr\u003e https://doi.org/10.1016/j.eiar.2024.107459\u003c/li\u003e\n\u003cli\u003eShao, S., Fan, M. \u0026amp; Yang, L. Economic structure adjustment, green technological progress and China\u0026apos;s low-carbon transition development. \u003cem\u003eManag. World\u003c/em\u003e \u003cstrong\u003e38\u003c/strong\u003e, 46\u0026ndash;69 (2022).\u003cbr\u003e https://doi.org/10.19744/j.cnki.11-1235/f.2022.0056\u003c/li\u003e\n\u003cli\u003eWang, Q. \u0026amp; Su, M. Drivers of decoupling economic growth from carbon emission \u0026mdash; An empirical analysis of 192 countries using decoupling model and decomposition method. \u003cem\u003eEnviron. Impact Assess. Rev.\u003c/em\u003e \u003cstrong\u003e81\u003c/strong\u003e, 106356 (2020).\u003cbr\u003e https://doi.org/10.1016/j.eiar.2019.106356\u003c/li\u003e\n\u003cli\u003eWang, Q. \u0026amp; Wang, L. Why does China\u0026apos;s carbon intensity decline and India\u0026apos;s carbon intensity increase? A decomposition analysis on the sectors. \u003cem\u003eJ. Clean. Prod.\u003c/em\u003e 265, 121569 (2020).\u003cbr\u003e https://doi.org/10.1016/j.jclepro.2020.121569\u003c/li\u003e\n\u003cli\u003eZhong, Y., Zhao, H. \u0026amp; Yin, T. Unleashing the power of AI: How industrial intelligence impacts carbon productivity? \u003cem\u003eTechnol. Forecast. Soc. Change\u003c/em\u003e 188, 122327 (2023).\u003cbr\u003e https://doi.org/10.1016/j.techfore.2022.122327\u003c/li\u003e\n\u003cli\u003eHuang, Y., Li, J. \u0026amp; Chen, J. AI and green innovation: Evidence from Chinese firms. \u003cem\u003eTechnol. Forecast. Soc. Change\u003c/em\u003e \u003cstrong\u003e200\u003c/strong\u003e, 123181 (2024).\u003cbr\u003e https://doi.org/10.1016/j.techfore.2023.123181\u003c/li\u003e\n\u003cli\u003eLi, Z., Huang, Z. \u0026amp; Su, Y. How does artificial intelligence affect the carbon emissions of urban agglomerations? The role of green technology innovation and spatial spillover. \u003cem\u003eJ. Clean. Prod.\u003c/em\u003e 434, 140002 (2024).\u003cbr\u003e https://doi.org/10.1016/j.jclepro.2023.140002\u003c/li\u003e\n\u003cli\u003eChen, J., Gao, M., Cheng, S., Hou, W., Song, M., Liu, X., Liu, Y. \u0026amp; Shan, Y. County-level CO2 emissions and sequestration in China during 1997\u0026ndash;2017. \u003cem\u003eSci. Data\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 391 (2020).\u003cbr\u003e https://doi.org/10.1038/s41597-020-00736-3\u003c/li\u003e\n\u003cli\u003eStrubell, E., Ganesh, A. \u0026amp; McCallum, A. Energy and policy considerations for deep learning in NLP. in \u003cem\u003eProceedings of the 58th Annual Meeting of the Association for Computational Linguistics\u003c/em\u003e 3645\u0026ndash;3653 (2020).\u003cbr\u003e https://doi.org/10.18653/v1/2020.acl-main.334\u003c/li\u003e\n\u003cli\u003eCheng, Y. \u0026amp; Yao, X. Carbon intensity reduction assessment of renewable energy technology innovation in China: A panel data model with cross-section dependence and slope heterogeneity. \u003cem\u003eRenew. Sust. Energ. R\u003c/em\u003eev. 135, 110157 (2021).\u003cbr\u003e https://doi.org/10.1016/j.rser.2021.112052\u003c/li\u003e\n\u003cli\u003eZhou, X., Zhang, J. \u0026amp; Li, J. Industrial structural transformation and carbon dioxide emissions in China. \u003cem\u003eEnergy Policy\u003c/em\u003e \u003cstrong\u003e57\u003c/strong\u003e, 43\u0026ndash;51 (2013).\u003cbr\u003e https://doi.org/10.1016/j.enpol.2012.07.017\u003c/li\u003e\n\u003cli\u003eBresnahan, T. F. \u0026amp; Trajtenberg, M. General purpose technologies: \u0026apos;Engines of growth\u0026apos;? \u003cem\u003eJ. Econometrics\u003c/em\u003e \u003cstrong\u003e65\u003c/strong\u003e, 83\u0026ndash;108 (1995).\u003cbr\u003e https://doi.org/10.1016/0304-4076(94)01598-T\u003c/li\u003e\n\u003cli\u003eBrevini, B. \u003cem\u003eIs AI Good for the Planet?\u003c/em\u003e (Polity Press, 2021).\u003c/li\u003e\n\u003cli\u003eGeels, F. W. From sectoral systems of innovation to socio-technical systems: Insights about dynamics and change from sociology and institutional theory. \u003cem\u003eRes. Policy\u003c/em\u003e 33, 897\u0026ndash;920 (2004).\u003cbr\u003e https://doi.org/10.1016/j.respol.2004.01.015\u003c/li\u003e\n\u003cli\u003eAcemoglu, D., Aghion, P., Bursztyn, L. \u0026amp; Hemous, D. The environment and directed technical change. \u003cem\u003eAm. Econ. Rev.\u003c/em\u003e \u003cstrong\u003e102\u003c/strong\u003e, 131\u0026ndash;166 (2012).\u003cbr\u003e https://doi.org/10.1257/aer.102.1.131\u003c/li\u003e\n\u003cli\u003eChen, J. \u0026amp; Lin, S. How does artificial intelligence drive the transformation and upgrading of industrial structure? \u003cem\u003eEcon. Model.\u003c/em\u003e \u003cstrong\u003e133\u003c/strong\u003e, 106666 (2024).\u003cbr\u003e https://doi.org/10.1016/j.econmod.2024.106666\u003c/li\u003e\n\u003cli\u003eDechezlepr\u0026ecirc;tre, A. \u0026amp; Sato, M. The impacts of environmental regulations on competitiveness. \u003cem\u003eRev. Environ. Econ. Policy\u003c/em\u003e 11, 183\u0026ndash;206 (2017).\u003cbr\u003e https://doi.org/10.1093/reep/rex013\u003c/li\u003e\n\u003cli\u003eBarbieri, N., Marzucchi, A. \u0026amp; Rizzo, U. Knowledge sources and impacts on the eco-innovation: Evidence from European firms. \u003cem\u003eRes. Policy\u003c/em\u003e \u003cstrong\u003e49\u003c/strong\u003e, 104070 (2020).\u003cbr\u003e https://doi.org/10.1016/j.respol.2020.104070\u003c/li\u003e\n\u003cli\u003eRennings, K. Redefining innovation\u0026mdash;eco-innovation research and the contribution from ecological economics. \u003cem\u003eEcol. Econ.\u003c/em\u003e \u003cstrong\u003e32\u003c/strong\u003e, 319\u0026ndash;332 (2000).\u003cbr\u003e https://doi.org/10.1016/S0921-8009(99)00112-3\u003c/li\u003e\n\u003cli\u003eLi, G. \u0026amp; Wang, X. Can artificial intelligence enhance green innovation efficiency? Evidence from China\u0026apos;s manufacturing sector. \u003cem\u003eJ. Environ. Manage.\u003c/em\u003e 351, 119658 (2024).\u003cbr\u003e https://doi.org/10.1016/j.jenvman.2023.119658\u003c/li\u003e\n\u003cli\u003eFerraris, A. et al. The role of AI in reducing uncertainty in innovation processes: A resource orchestration perspective. \u003cem\u003eTechnovation\u003c/em\u003e 132, 102981 (2024).\u003cbr\u003e https://doi.org/10.1016/j.technovation.2024.102981\u003c/li\u003e\n\u003cli\u003eCrippa, M. et al. CO2 emissions of all world countries 2024 Report. \u003cem\u003ePublications Office of the European Union\u003c/em\u003e (2024).\u003cbr\u003e https://edgar.jrc.ec.europa.eu/report_2024\u003c/li\u003e\n\u003cli\u003eMinistry of Science and Technology of the People\u0026apos;s Republic of China. Notice of the Ministry of Science and Technology on the Issuance of the \u0026apos;Guidelines for the Construction of National New Generation Artificial Intelligence Innovation and Development Pilot Zones\u0026apos;. \u003cem\u003eMoST\u003c/em\u003e (2019).\u003c/li\u003e\n\u003cli\u003ede Chaisemartin, C. \u0026amp; D\u0026apos;Haultf\u0026oelig;uille, X. Two-way fixed effects estimators with heterogeneous treatment effects. \u003cem\u003eAm. Econ. Rev.\u003c/em\u003e \u003cstrong\u003e110\u003c/strong\u003e, 2964\u0026ndash;2996 (2020).\u003cbr\u003e https://doi.org/10.1257/aer.20181169\u003c/li\u003e\n\u003cli\u003eSun, L. \u0026amp; Abraham, S. Estimating dynamic treatment effects in event studies with heterogeneous treatment effects. \u003cem\u003eJ. Econom.\u003c/em\u003e 225, 175\u0026ndash;199 (2021).\u003cbr\u003e https://doi.org/10.1016/j.jeconom.2020.09.006\u003c/li\u003e\n\u003cli\u003eAnselin, L. \u003cem\u003eSpatial Econometrics: Methods and Models\u003c/em\u003e (Kluwer Academic Publishers, 1988).\u003c/li\u003e\n\u003cli\u003eLeSage, J. P. \u0026amp; Pace, R. K. \u003cem\u003eIntroduction to Spatial Econometrics\u003c/em\u003e (CRC Press, 2009).\u003c/li\u003e\n\u003cli\u003ePreacher, K. J. \u0026amp; Hayes, A. F. Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. \u003cem\u003eBehav. Res. Methods\u003c/em\u003e 40, 879\u0026ndash;891 (2008).\u003cbr\u003e https://doi.org/10.3758/BRM.40.3.879\u003c/li\u003e\n\u003cli\u003eCrippa, M. et al. GHG emissions of all world countries - 2023 Report. \u003cem\u003ePublications Office of the European Union\u003c/em\u003e (2023).\u003c/li\u003e\n\u003cli\u003eFang, C. \u0026amp; Yu, D. Urban agglomeration: An evolving concept of an emerging phenomenon. \u003cem\u003eLandsc. Urban Plan.\u003c/em\u003e \u003cstrong\u003e162\u003c/strong\u003e, 126\u0026ndash;136 (2017).\u003cbr\u003e https://doi.org/10.1016/j.landurbplan.2017.02.014\u003c/li\u003e\n\u003cli\u003eLiu, J., Zhao, M. \u0026amp; Wang, Y. Impact of digital economy on urban green innovation: Evidence from China. \u003cem\u003eJ. Clean. Prod.\u003c/em\u003e \u003cstrong\u003e434\u003c/strong\u003e, 139996 (2024).https://doi.org/10.1016/j.jclepro.2024.139996\u003c/li\u003e\n\u003cli\u003eAghion, P. et al. Carbon taxes, path dependency, and directed technical change: Evidence from the auto industry. \u003cem\u003eJ. Polit. Econ.\u003c/em\u003e \u003cstrong\u003e124\u003c/strong\u003e, 1\u0026ndash;51 (2016).https://doi.org/10.1086/684581\u003c/li\u003e\n\u003cli\u003eAngrist, J. D. \u0026amp; Pischke, J.-S. \u003cem\u003eMostly Harmless Econometrics: An Empiricist\u0026apos;s Companion\u003c/em\u003e (Princeton University Press, 2009).\u003c/li\u003e\n\u003cli\u003eDuranton, G. \u0026amp; Turner, M. A. Urban growth and transportation. \u003cem\u003eRev. Econ. Stud.\u003c/em\u003e \u003cstrong\u003e79\u003c/strong\u003e, 1407\u0026ndash;1440 (2012).https://doi.org/10.1093/restud/rdr042\u003c/li\u003e\n\u003cli\u003eStock, J. H. \u0026amp; Yogo, M. Testing for weak instruments in linear IV regression. in \u003cem\u003eIdentification and Inference for Econometric Models: Essays in Honor of Thomas Rothenberg\u003c/em\u003e (eds Andrews, D. W. K. \u0026amp; Stock, J. H.) 80\u0026ndash;108 (Cambridge University Press, 2005).\u003cbr\u003e https://doi.org/10.1017/CBO9780511614491.006\u003c/li\u003e\n\u003cli\u003eJiang, Z. \u0026amp; Ding, P. Causal mediation analysis with latent mediators. \u003cem\u003ePsychometrika\u003c/em\u003e 88, 1093\u0026ndash;1115 (2023).https://doi.org/10.1007/s11336-023-09924-7\u003c/li\u003e\n\u003cli\u003eConley, T. G., Hansen, C. B. \u0026amp; Rossi, P. E. Plausibly exogenous. \u003cem\u003eRev. Econ. Stat.\u003c/em\u003e 94, 260\u0026ndash;272 (2012).https://doi.org/10.1162/REST_a_00139\u003c/li\u003e\n\u003cli\u003eJaffe, A. B., Trajtenberg, M. \u0026amp; Henderson, R. Geographic localization of knowledge spillovers as evidenced by patent citations. \u003cem\u003eQ. J. Econ.\u003c/em\u003e \u003cstrong\u003e108\u003c/strong\u003e, 577\u0026ndash;598 (1993).https://doi.org/10.2307/2118401\u003c/li\u003e\n\u003cli\u003eAudretsch, D. B. \u0026amp; Feldman, M. P. R\u0026amp;D spillovers and the geography of innovation and production. \u003cem\u003eAm. Econ. Rev.\u003c/em\u003e \u003cstrong\u003e86\u003c/strong\u003e, 630\u0026ndash;640 (1996).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"artificial intelligence, carbon emissions, pilot zone, difference-in-differences, spatial spillover, urban China","lastPublishedDoi":"10.21203/rs.3.rs-7917762/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7917762/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThe global low-carbon transition necessitates innovative policy interventions. Using staggered difference-in-differences estimation on a panel of 282 Chinese cities (2010\u0026ndash;2023), this study provides causal evidence that China's National AI Innovation Pilot Zones (AIPZ) policy significantly reduces urban carbon emissions by 6.3% on average. Spatial econometric models reveal substantial negative spillovers, inducing an additional 8.6% reduction in contiguous cities, leading to a total abatement effect of 14.3%. Mechanism and heterogeneity analyses show that industrial upgrading and green innovation are key channels, with effects pronounced in the Pearl River Delta and non-resource-based cities, but short-run rebound effects occur in resource-dependent areas. This study demonstrates demonstrate that AI policies generate carbon co-benefits, yet their efficacy depends critically on local industrial context and spatial linkages, underscoring the importance of regional coordination in climate governance. Our findings underscore the importance of integrating AI policies into regional climate strategies to maximize carbon co-benefits.\u003c/p\u003e","manuscriptTitle":"The Carbon Reduction Effect of AI Policy: Quasi-Experimental Evidence from China's National AI Innovation Pilot Zones","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-24 09:04:36","doi":"10.21203/rs.3.rs-7917762/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-01-06T17:27:04+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-01-04T06:42:38+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-12-15T05:33:54+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"274438303668372931180047208098078125761","date":"2025-12-04T01:06:27+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"256472492748868786727340020757680144889","date":"2025-11-18T11:41:16+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-11-18T00:12:40+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-10-31T20:02:53+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-10-27T02:30:07+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-10-27T02:29:53+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-10-21T14:36:35+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"fa7db8de-7c30-45ed-b793-d5b59c0d31ff","owner":[],"postedDate":"October 24th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":56778529,"name":"Earth and environmental sciences/Environmental social sciences"},{"id":56778530,"name":"Scientific community and society/Geography"},{"id":56778531,"name":"Social science/Geography"}],"tags":[],"updatedAt":"2026-05-18T14:24:50+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-24 09:04:36","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7917762","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7917762","identity":"rs-7917762","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}