Marco andrea@passaglia.it
The Bellwether

A morning brief, composed for you when the sources say something worth saying.

‹ Reference

ssrn-5854389

paper Reference Materials/AI Papers and case studies 61 KB text added 6/4/2026
Reducing Urban Energy Intensity through Artificial Intelligence: Exploring the Mediating Roles of New Energy and Green Industry in Chinese Cities Xinwei Zhanga* a College of Economics, Sichuan Agricultural University, Chengdu, 611130, China A B S T R A C T Artificial intelligence is increasingly recognized as a key driver in promoting industrial intelligence and transforming urban energy systems. This study investigates the impact of artificial intelligence development on energy intensity using panel data from 280 Chinese cities from 2011 to 2023. Empirical analyses reveal that artificial intelligence development significantly reduces urban energy intensity, and this result remains robust after a series of robustness tests and addressing endogeneity issues. Further mechanism analysis reveals that artificial intelligence primarily promotes energy intensity reduction through three emerging pathways, including supporting new energy industry development, facilitating green industry upgrading, and stimulating green technological innovation. Moreover, artificial intelligence’s effect on reducing energy intensity is stronger in cities implementing low-carbon pilot policies, possessing strong digital economy policy support, and having well-developed green infrastructure, which underscores the importance of favorable policies and infrastructure in enhancing artificial intelligence’s energy efficiency benefits. Overall, these findings highlight artificial intelligence’s vital enabling role in driving green and low-carbon urban transformation. Our analysis suggests that policymakers should further promote deep integration of artificial intelligence with urban energy systems and industrial sectors, optimize supportive policies, accelerate green infrastructure construction, and advance intelligent development in new energy and green industries to enhance energy efficiency and achieve sustainable urban development. A R T I C L E I N F O Keywords: Artificial Intelligence Energy Intensity New Energy Industry Green Industry 1. Introduction Serving as an indispensable pillar for economic and social progress, activities of energy production and consumption do not merely constitute the driving force behind modernization and urbanization advancement but also stand as the central focus of contradictions amid global climate change and environmental constraints. Energy consumption is a key contributor to carbon dioxide (CO₂) emissions. The International Energy Agency (IEA) reports that global energy-related CO₂ emissions rose 0.8% in 2024 to a record 37.8 gigatonnes (Gt). Meanwhile, rapid economic and population growth in emerging and developing economies is spurring surging energy demand and related emissions. In 2024, these economies accounted for over 80% of global energy demand growth, with their energy-related CO₂ emissions rising 1.5% to 375 million tonnes (Mt). IEA statistics show China's 2024 energy demand rose approximately 3%, leading global * Corresponding author. College of Economics, Sichuan Agricultural University, Chengdu, 611130, China ; https://orcid.org/0009-0005-7695-5801 E-mail address: zxw@sicau.edu.cn (X. Zhang). This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 1 of 19 -- absolute growth, while its energy-related carbon emissions comprised approximately 33.3% of the global total. Extreme high temperatures drove China's energy-related CO₂ emissions up by approximately 100 Mt that year (IEA, 2025). Consequently, China and the international community face dual pressures from rising energy demand and green low-carbon transition. Balancing the rising demand for economic growth with reductions in energy intensity (EI) and advancing the green transition is a pressing priority for both combating climate change and achieving the United Nations Sustainable Development Goals, particularly SDG7 (affordable and clean energy) and SDG13 (climate action). Responding to climate change and resource constraints, China launched its "dual-carbon" strategy in 2020 to achieve carbon peak and neutrality. Key measures include reducing EI per unit output and developing clean, low-carbon, secure, and efficient energy systems. However, energy consumption and economic growth demonstrate strong interdependence; for example, electricity use in urban industries acts as a key determinant and "barometer" of economic growth (Lu et al., 2024). As the world's largest energy producer and consumer, China must balance economic growth with conservation and emission reductions, yet its traditional extensive growth model and consumption patterns remain major barriers. China has positioned scientific and technological innovation as the core driver of its energy transition, emphasizing energy-technology collaboration and advancing new-quality energy production capabilities. This aims to build a safe, efficient energy system and foster a consumption model of conservation, efficiency, sustainability and inclusiveness. In 2025, the government released China's Energy Transition white paper, outlining two key measures that leverage digital technologies for energy industry transformation and accelerate smart infrastructure upgrades. Also in 2025, the "AI + Energy" High-Quality Development Guidelines designated artificial intelligence (AI) innovation as the primary pathway for high-quality energy development, targeting internationally leading energy-specific AI technologies and applications by 2030. However, emerging research reveals AI's multifaceted impacts on energy systems. AI optimizes production, enhances decision-making efficiency (Li et al., 2025c), promotes technological diffusion and innovation (Liu et al., 2025; Zhong et al., 2025), facilitates intelligent management (Zhang & Zeng, 2024), and demonstrates substantial EI reduction potential (Liu et al., 2025). Conversely, AI implementation drives increased model training scales, surging computing power demands (Vinuesa et al., 2020), and continuous data center expansion (De Vries, 2023), collectively raising net energy consumption, with GPT-3's single training run emitting CO₂ equivalent to 49 cars' annual emissions (Taddeo et al., 2021). Meanwhile, AI-driven ICT and digital sector proliferation may increase energy demand beyond efficiency gains from intelligent industrial transformation, elevating net consumption (Lin and Zhou, 2025; Lange et al., 2020). Current studies on AI's EI impact remain largely confined to enterprise or single-industry scales (Li et al., 2025c; Zhong et al., 2025). Rigorous research is needed to elucidate AI's influence pathways for accurately gauging AI's effects on energy systems and facilitating the sector's green, low-carbon transition. This study makes marginal contributions in the following aspects: First, existing research emphasizes enterprise or single-industry perspectives, lacking systematic urban-level analysis of AI's energy-saving effects. Moreover, EI measurements in literature depend on macro-level statistics with limited granularity. We employ panel data from This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 2 of 19 -- 280 Chinese cities (2011–2023), combining nighttime satellite light and macro energy statistics to systematically estimate city-level EI and examine AI development's effects. This enhances scientific validity and urban-level resolution of EI measurement while creating a robust data foundation for future studies on emerging technologies and city-level energy systems. Second, this research examines differential effects of policies and infrastructure on AI's energy-saving outcomes. Current literature explores heterogeneity through industrial or geographic lenses (Liu et al., 2025; Zhong et al., 2025). In contrast, this paper employs subgroup regressions using low-carbon city pilot status, digital economy policy attention, and green infrastructure development to assess their effects on AI's energy-saving heterogeneity, uncovering the critical role of policy environment and infrastructure in driving AI's green spillovers and supporting digital-green policy pilots. Third, this research uncovers novel mechanisms through which AI affects EI. Previous studies examine direct effects or factors like management optimization and innovation capability (Zhang & Zeng, 2024; Liu et al., 2021). This study investigates emerging pathways across new energy sector development, green industry expansion, and green innovation promotion, examining how AI empowers these domains, facilitates green patent generation, and drives sustained reductions in EI, thereby extending the theoretical framework of AI-empowered energy transformation and offering insights into AI's role in facilitating the green transition of urban energy systems. 2. Literature review and theoretical analysis 2.1. Literature review EI, defined as energy used per output (Vu & Demena, 2025), quantitatively measures integrated energy efficiency. It indicates an economy’s energy dependence (Zhang et al., 2020) and link to carbon emissions (Dunyo et al., 2024), playing a crucial role in balancing economic growth with energy and environmental constraints. EI determinants are complex and layered. Macro-level studies highlight the roles of low-carbon transition, globalization (Zaidi et al., 2024), green technology transfer (Zhang et al., 2025a), and sustainable energy use (Feng et al., 2023). In China, EI shifts are primarily driven by technological improvements (Huang et al., 2025), industrial structure optimization, and a cleaner energy mix with greater non-fossil energy shares (Dong et al., 2025). Micro-level EI changes arise from corporate governance, technological innovation, environmental pressures, and policy, with key influences including managers’ experience (Sun et al., 2024), new technology adoption (Lin and Xu, 2024; Wang et al., 2025), climate risk (Li et al., 2025a), green credit (Tan et al., 2025), and regulations (Chen et al., 2024; Liu et al., 2023). Overall, technological and emerging technology innovations are pivotal in lowering EI at both macro and micro levels (Zhang et al., 2025a; Vu & Demena, 2025). Digital technologies have transformed economic, social, and energy systems, with their EI impact gaining scholarly attention (Xiao et al., 2025; Hao et al., 2025). Digital innovations like enterprise digital transformation and industrial robot deployment optimize production and reshape energy use patterns, reducing EI (Guan et al., 2025; Matthess et al., 2023). Macro-level mechanisms reveal digital technologies expand digital economy sectors (e.g., software and IT services) and upgrade traditional industries, reducing high-energy industries' proportion and energy consumption (Huang et al., 2023; Ma & Lin, 2025). Micro-level mechanisms reveal digital applications reduce corporate EI by boosting productivity and innovation efficiency (Guan et al., 2025), enabling energy-saving financing (Xiao et al., 2025), enhancing renewable innovation This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 3 of 19 -- awareness (Wu et al., 2025), scaling energy use economies (Hao et al., 2025), and optimizing energy management (Hao et al., 2025). Nevertheless, digital technologies' EI impact demonstrates uncertainty and complexity. Superficial digitalization offers limited energy-saving potential, as generic solutions often yield only short-term optimizations without iterative, enterprise-specific capability enhancements. Moreover, the "rebound effect" (where efficiency gains stimulate heightened energy demand) may reduce energy-saving outcomes from digital technologies (Guan et al., 2025). Consequently, energy consumption growth from digital business expansion (e.g., digital equipment investment, ICT hardware production) may exceed "energy-saving effects," increasing net consumption (Lange et al., 2020; Matthess et al., 2023). Recent AI-driven technological transformation has created novel pathways for industrial digitalization and new opportunities for energy conservation and emissions reduction in socio-economic systems (Fu et al., 2024). Collectively, AI-enabled industrial innovation and intelligent transformation effectively reduce energy consumption and EI in corporate and energy-intensive sectors (Zhang & Zeng, 2024; Liu et al., 2021). Macro-level transformation shifts energy structures toward cleaner sources, improving economic system energy performance (Guo et al., 2025). AI achieves energy reduction via four mechanisms, including enhanced enterprise innovation efficiency (Liu et al., 2021), advanced intelligent management systems cutting costs (Zhang & Zeng, 2024), substituted traditional production factors with optimized processes (Li et al., 2025), and guided green technological upgrades (Liu et al., 2025). From the perspective of its mechanisms, AI-driven energy reduction operates by enhancing operational innovation systems (Liu et al., 2021), deploying intelligent management to reduce expenditures (Zhang & Zeng, 2024), substituting production factors with process optimization (Li et al., 2025), and advancing green tech upgrading (Liu et al., 2025), collectively contributing to energy conservation. Concurrently, AI facilitates technology diffusion and innovation efficiency gains (Zhong et al., 2025), enhances regional human capital and green innovation capacity (Zhu et al., 2025), and promotes renewable energy transitions (Wang, 2026), reducing energy consumption. However, uncertainty persists regarding AI's EI impact due to energy-intensive model training (Vinuesa et al., 2020), heightened computing demands from generative AI (Taddeo et al., 2021; Lin & Zhou, 2025), and accelerated digital service consumption through large-scale deployment (De Vries, 2023), constraining energy-saving potential. While research focuses on AI's conservation potential, most evidence originates from enterprise or single-industry analyses (Liu et al., 2021; Li et al., 2025; Zhong et al., 2025). Significantly, macro-scale investigations of AI's influence on EI, particularly at urban levels, remain scarce, necessitating systematic macro-level evaluation to verify reduction efficacy and advance mechanism exploration. 2.2. Theoretical analysis and research hypotheses 2.2.1 Direct Effects: The Role of AI in Lowering EI As an emerging general-purpose technology, AI demonstrates broad applicability across multiple industrial sectors while driving technological advancements and complementary innovations within adopting industries (Teece, 2018). Given its broad integration across socioeconomic systems, AI has attracted significant attention from multiple stakeholder groups, specifically governments, industries, academia, and research institutions, as a pivotal catalyst for global energy transition (Jin et al., 2020; Lyu and Liu, 2021). Empirical evidence confirms AI's effectiveness in reducing EI and supporting energy system This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 4 of 19 -- decarbonization. AI improves energy efficiency and lowers EI by optimizing industrial and enterprise production processes. Industrial robots and intelligent control systems adjust equipment operation to match energy demands, minimizing idle or overload states and directly reducing energy use per unit output (Liu et al., 2021; Li et al., 2025c). Additionally, AI reduces human-induced downtime and errors, improves production continuity, and curbs intermittent energy waste in production (Zhang & Zeng, 2024; Liu et al., 2025). AI also enables real-time energy management optimization, further cutting EI. For instance, it optimizes renewable power systems through intelligent forecasting and dynamic grid adjustment to raise efficiency (Liu et al., 2022). Simultaneously, AI drives technological innovation and digital transformation, such as energy-saving equipment R&D and IoT monitoring, helping upgrade technology and management in energy systems, thus improving efficiency and reducing EI (Liu et al., 2025; Li et al., 2025c). Although scholars recognize that AI may increase energy use through digital technology-driven economic growth (Lange et al., 2020), this risk can be mitigated by technological innovation and policy measures (Umar et al., 2025). In summary, AI holds significant promise as an effective tool for reducing EI and consistently achieving lower EI across broad applications. H1: AI can effectively reduce EI. 2.2.2 AI Empowerment and EI Reduction: The Role of the New Energy Industry The United Nations Conference on New Energy and Renewable Energy defines new energy as technologically or developmentally innovative energy sources that support the global transition from fossil fuels to a more diversified and sustainable energy mix (United Nations, 1981). Main types include solar, geothermal, ocean, modern wind, biomass energy, and small hydropower. By supporting the move to a green and sustainable structure, new energy development greatly reduces EI. New energy primarily cuts EI by replacing high-consumption fossil fuels (Jiao et al., 2024). For instance, Rahman et al. (2022) found that a unit increase in renewable energy share lowers carbon intensity by 0.003 in 25 emerging economies and improves energy system efficiency. Additionally, promoting new energy technologies is closely tied to transforming the energy system’s structure. Raising the share of modern renewable energy in final energy use and balancing global supply and demand (Xu and Lieuea, 2025) systematically lowers EI. Furthermore, policy assessments show that encouraging new energy application effectively reduces firms’ EI (Liu et al., 2023). Thus, new energy development is a key route to reducing EI. As a transformative technology, AI accelerates new energy development and adoption. First, AI enhances renewable energy system efficiency and application scale by forecasting photovoltaic output and optimizing wind farm layouts, thus reducing unit costs (Alhasnawi et al., 2024; Bennagi et al., 2024). Second, in the long run, AI increases the renewable share in the energy mix by optimizing supply-demand balance and lessening reliance on traditional sources (Zhao et al., 2024). AI also advances new energy technology innovation and speeds up technology iteration in energy systems, promoting innovative applications (Zhang et al., 2024; Lee et al., 2025). Overall, AI supports new energy advancement and helps lower EI. H2: AI can reduce EI by facilitating the expansion of the new energy industry. 2.2.3 AI Empowerment and Reduction of EI: The Role of Green Industry The United Nations Industrial Development Organization (UNIDO) defines Green Industry as an industrial system focused on clean technologies, efficient resource use, low pollution, and reduced energy consumption to promote both economic growth and environmental protection This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 5 of 19 -- (UNIDO, 2011). Research shows that green industry development improves resource allocation, diversifies and decarbonizes energy consumption, and advances the circular economy (Pegels and Lütkenhorst, 2014; Ramalingam et al., 2025; Xie and Lin, 2025), supporting green energy transition and EI reduction. AI, as an enabling technology, plays a key role in green industry growth. In green manufacturing, AI allows for quantitative assessment of material and energy efficiency as well as emissions (Mao et al., 2019), and enhances corporate ESG performance to accelerate green transformation (Zhou et al., 2025). The Green Industrial Internet of Things (GIIoT), central to industrial digitalization, integrates AI with renewables to build smart grid ecosystems, improving efficiency (Tabaa et al., 2020). In green logistics, combining AI with knowledge graphs enables real-time route and energy optimization using transportation and environmental data (Samadhiya et al., 2025). In green supply chains, AI and IoT jointly optimize inventory and transportation, reducing lifecycle energy use and emissions (Zhang et al., 2025). In summary, AI development promotes green industry advancement, driving EI reduction. H3: AI can reduce EI by promoting the development of green industry. 2.2.4 AI Empowerment and EI Reduction: The Role of Green Innovation As a key force for sustainable development, green innovation extends beyond technological and corporate transformation to underpin macro-level green, low-carbon, and high-quality growth. First, green innovation directly improves micro-level production processes, boosting energy efficiency through process and model upgrades for both rapid and sustained EI reductions (Li et al., 2023; Wang et al., 2025). Second, it raises macro-level energy efficiency by optimizing resource allocation, reducing dependence on fossil fuels, and promoting clean energy substitution to increase overall system efficiency (Shahzad, 2025; Ullah et al., 2025). Simultaneously, AI significantly promotes green innovation. At the micro level, AI improves innovation resource allocation by easing financing constraints, accumulating human capital, enhancing human-AI collaboration, and supporting data-driven decisions, thus optimizing resource use and reducing information asymmetry to improve green innovation quality and efficiency (Zhao et al., 2025; Chen et al., 2025; Li et al., 2025b; Guo et al., 2025). At the macro level, AI acts as a digital connector, strengthening the synergy between green technology innovation and renewable energy transition (Behera et al., 2025). It also fosters cross-domain collaboration, increases R&D investment, and creates synergy effects in industrial chains, improving collaborative efficiency and green innovation outcomes (Li et al., 2025; Zhao et al., 2025; Chen et al., 2025). Consequently, AI development promotes green innovation, reducing EI. H4: AI is capable of reducing EI through fostering green innovation. Figure 1 illustrates a theoretical framework for AI's influence on EI. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 6 of 19 -- Fig. 1. Theoretical framework. 3. Method and model 3.1. Data This study examines AI’s impact on EI in Chinese cities. Based on data availability and sample representativeness, we constructed a balanced panel dataset of 280 cities from 2011 to 2023 for regression analysis. AI data were obtained from the China Big Data Platform for Social-Science (CBDPS). City-level energy consumption was estimated by combining nighttime light data with provincial energy consumption, following Wu et al. (2014) and Chen et al. (2020a). Provincial energy data were sourced from the China Energy Statistical Yearbook as well as provincial and municipal statistical yearbooks. Nighttime light data were sourced from Harvard Dataverse (Wu et al., 2021). We also collected socioeconomic data from the China City Statistical Yearbook, CNRDS, and the China Public Policy and Green Development Database (CPPGD). 3.2. Variable 3.2.1. Dependent variable As the primary metric, EI measures overall energy efficiency as energy use per unit of GDP, reflecting how effectively energy inputs generate economic output. Unlike total energy use, EI controls for economic scale, allowing for scientific comparison of energy efficiency across regions and over time. Lower EI in cities indicates greater energy efficiency and progress toward intensive green development. EI was measured as the ratio of city-level energy consumption to real GDP. City-level energy consumption was estimated by integrating provincial energy consumption data with nighttime light data. Specifically, following Wu et al. (2014) and Chen et al. (2020), we established a linear model between provincial total energy consumption and total nighttime light brightness to derive annual light-energy conversion coefficients kit for each province i and year t, representing energy consumption per unit light brightness. Subsequently, city-level energy consumption was derived by applying the conversion coefficients kit to each city's nighttime light intensity during 2011 to 2023. Fig. 2(a,b) shows the spatiotemporal distribution of EI levels. During 2011-2023, most cities declined in EI levels. In 2011, northeastern, northern, and northwestern regions exhibited higher Literature Review And Theoretical Analysis Artificial Intelligence Energy Intensity Direct Effect Energy Management System Optimization Artificial Intelligence Energy Intensity Mechanism Analysis New Energy Industry Green Industry Green Innovation Intelligent Optimization of Production Processes This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 7 of 19 -- EI than central and southeastern coastal regions. By 2023, high EI areas decreased substantially, with only northern cities maintaining high levels, while central and southeastern coastal regions consistently registered low levels. 3.2.2. Core independent variable Enterprises act as key drivers of emerging industries and are main implementers of new technology. This study uses the number of AI enterprises as an indicator for regional AI development. Fig. 2(c,d) displays the spatiotemporal patterns of AI levels. Between 2011 and 2023, most cities experienced notable growth in AI. In 2023, southeastern coastal, North China, and East China regions showed higher AI development than northwest, northeast, and southwest regions. Fig. 2. Spatial distribution of EI and AI in 2011 and 2023. 3.2.3. Control variables To rigorously assess AI’s impact on urban EI, this study includes control variables to address confounders, with definitions and descriptive statistics shown in Table 1. A region’s economic development level is likely to influence its technological innovation and energy use efficiency (Guo et al., 2025; Umar et al., 2025), and is represented by real per capita GDP. Industrial structure changes may affect EI by altering the share of high-energy sectors and resource allocation efficiency (Luan et al., 2021), thus the proportion of tertiary industry in GDP quantifies industrial structure. Foreign investment is likely to improve energy efficiency through technology transfer and managerial expertise (De et al., 2021), measured by the number of foreign-invested enterprises. Population size, assessed by the registered population, is likely to shape regional energy consumption patterns (Huang et al., 2025). Government intervention may impact technological innovation, infrastructure, and energy efficiency (Zhang et al., 2025c), measured by local public budget expenditure. Financial support may affect the adoption of clean technologies and resource allocation, thus influencing EI (Zhang et al., 2025b), with the year-end loan balance of financial institutions used as its indicator. Table 1 This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 8 of 19 -- Definition of control variables Statistical Summary Control Variable Symbol Definition Mean Std. Min Max N Economic development PGDP Real Per Capita Regional Gross Domestic Product 5.16 3.01 0.69 24.54 3640 Industrial structure TSR Proportion of Tertiary Industry Added Value to GDP 43.43 10.10 10.15 84.85 3640 Foreign investment FIE Number of Foreign-Invested Enterprises 94.06 285.51 0 3292 3640 Population size POP Registered Population 460.09 324.11 29.97 3416 3640 Government intervention GEXP Local General Public Budget Expenditure 480.07 716.96 56.69 9638.50 3640 Financial support FIN Year-end Balance of Loans from Financial Institutions 4290.70 9123.00 91.20 108600 3640 3.2.4. Intermediary variables This study selects the developmental levels of the new energy industry, green industry, and green innovation as mediating variables to explore how AI affects EI, based on its theoretical framework. The new energy industry’s development is assessed by the number of relevant enterprises, with data from CBDPS. The green industry’s development is measured by firm counts in sectors such as energy-saving equipment, advanced transportation manufacturing, carbon-reduction retrofitting, and green low-carbon industrial transitions, using CPPGD data. Green innovation is indicated by the number of granted green patents awarded per city, based on CNRDS data. Definitions and descriptive statistics for these mediators are provided in Table 2. Table 2 Definition of intermediary variables Statistical Summary Intermediary variables Symbol Definition Mean Std. Min Max N New energy industry development NEI The number of enterprises in the new energy industry 1438.21 2809.68 11 38335 3640 Green industry development GRI The number of enterprises in the green industry 950.58 2089.26 5 23794 3640 Green innovation GPA The total number of authorized green patents 483.80 1360.94 1 19373 3640 3.3. Econometric model Based on the above variables and data, this study employs a panel data regression model to investigate the impact of AI on EI and its underlying mechanisms, as specified below: (1) As specified in Equation (1), i denotes city and t denotes year. The dependent variable EIit represents EI, defined as energy consumption per unit of GDP output. AIit, the core explanatory variable, measures the level of AI development. Controlsit denotes a set of control variables, while α0 to α2 are estimated parameters. The model includes individual fixed effects ui, time fixed effects vt, and a stochastic error term εit. The coefficient α1 on AIit is the primary focus; it is hypothesized to be negative, suggesting that AI development can reduce EI. Given the limitations of the traditional three-step mediation test (Aguinis et al., 2017; Pieters, 0 1 2 it it it i t it EI AI Controls u v           This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 9 of 19 -- 2017), this study draws on the methodological framework of Chen et al. (2020b) and Li et al. (2025c) to construct the mediation model: (2) Equation (2) specifies MVit as the mediator variable set, comprising the quantity of new energy enterprises (NEI), green industry enterprises (GRI), and green patents (GPA). Parameters β0 to β2 require estimation. 4. Empirical analysis 4.1. Analysis of typical facts This study employs ridgeline plots to visualize kernel density distributions of EI and AI enterprise stock, with results in Figure 3(a) and (b). Figure 3(a) shows EI distributions from 2011-2023. Leftward curve shifts indicate sustained EI decline and significant urban energy efficiency gains in China. Increasing peaks and narrowing curves transition distributions from "low-wide" to "high-narrow," implying cross-city convergence and balanced nationwide reductions. Figure 3(b) presents AI enterprise stock distributions for 2011-2023. A log1p transformation mitigates right-skewness while axis labels display original values. Rightward curve shifts reflect accelerated AI enterprise growth. Peak morphology evolves from "high-narrow" to "low-wide," showing expanded dispersion and widening regional disparities that suggest emerging agglomeration effects. Collectively, decreasing EI and expanding AI enterprises exhibit coupled spatiotemporal evolution, providing preliminary evidence that AI development may reduce EI and supporting future empirical research. Fig.3. Density Distribution of Energy Intensity and AI Enterprise Stock 4.2. Benchmark regression results Table 3 reports baseline regression results examining AI's effect on EI. Column (1) shows a pooled regression without controls, where lnAI coefficient is -0.1486 (significant at 1%). Column (2) employs a two-way fixed effects model (city and year) without controls, yielding lnAI coefficient of -0.0646 (significant at 1%), confirming AI significantly reduces EI after controlling for temporal and regional heterogeneity. Column (3) adds controls for economic development (lnPGDP), industrial structure (lnTSR), foreign investment (lnFIE), population (lnPOP), government intervention (lnGEXP), and financial support (lnFIN) to the pooled model, where lnAI coefficient remains -0.0658 and significant at 1%, affirming a stable AI impact on EI reduction. Table 3 0 1 2 it it it i t it MV AI Controls u v           This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 10 of 19 -- The results of robustness regression (1) (2) (3) (4) lnAI -0.1486*** -0.0646*** -0.0658*** -0.0206** (-32.6326) (-6.2664) (-7.4522) (-2.4774) lnPGDP -0.5115*** -0.6449*** (-18.5063) (-27.8567) lnTSR -0.1871*** 0.1387*** (-4.2161) (4.5477) lnFIE -0.0169** -0.0110 (-2.2226) (-0.8809) lnPOP -0.2999*** -1.0009*** (-12.9643) (-15.9877) lnGEXP -0.0228 0.0441 (-0.8339) (1.5632) lnFIN 0.1778*** 0.0006 (8.8219) (0.0302) Constant 0.0711*** -0.1617*** 0.5098** 5.4482*** (4.8444) (-5.6537) (2.0631) (10.4199) City FE NO Yes NO Yes Year FE NO Yes NO Yes Observations 3,640 3,640 3,640 3,640 R-squared 0.226 0.881 0.323 0.924 Note: Figures in parentheses are t-statistics. ***p < 0.01, **p < 0.05, *p < 0.1. Column (4) results show a lnAI coefficient of -0.0206, significant at 5% level (p=0.013) and approaching 1% significance. Among control variables, lnPGDP and lnPOP exhibit negative coefficients significant at 1%, suggesting higher economic development and population size correlate with lower EI, likely due to scale and agglomeration effects. Overall, by controlling for variables and two-way fixed effects, Column (4) more precisely estimates the net AI effect. Consequently, baseline regression results indicate AI development significantly lowers EI, supporting Hypothesis 1. 4.3. Robustness checks To ensure baseline reliability, robustness checks are summarized in Table 4. Column (1) uses AI patents (lnAIP) as an alternative measure, with coefficient -0.0151 significant at 1%. Column (2) applies 1% winsorization to variables, yielding lnAI coefficient of -0.0238 significant and negative. Column (3) excludes 2020 data to eliminate COVID-19 impacts on EI, showing lnAI coefficient -0.0182 significant at 5%. Column (4) employs province-year interaction fixed effects to control for cross-provincial time trends, with lnAI coefficient -0.0415 significant at 1%. Collectively, these checks demonstrate AI significantly reduces EI across variable substitution, extreme value treatment, sample adjustment, and interactive fixed effects, confirming baseline robustness. Table 4 This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 11 of 19 -- The results of robustness regression and endogenous amendment (1) (2) (3) (4) (5) (6) lnEI lnEI lnEI lnEI lnAI lnEI lnAI -0.0238*** -0.0182** -0.0415*** -0.8554** (-2.9052) (-2.1118) (-3.8973) (-2.3625) lnAIP -0.0151*** (-5.2021) IV 0.0710** (2.4175) Controls Yes Yes Yes Yes Yes Yes City FE Yes Yes Yes NO Yes Yes Year FE Yes Yes Yes NO Yes Yes Province × Year FE Yes Constant 5.1654*** 4.5690*** 5.4168*** 3.3264*** -3.0313*** / (9.8597) (8.0778) (10.0576) (8.2654) (-2.6149) / Observations 3,640 3,640 3,360 3,575 3,640 3640 R-squared 0.925 0.919 0.921 0.724 0.962 / Note: Figures in parentheses are t-statistics. ***p < 0.01, **p < 0.05, *p < 0.1. 4.4. Endogenous Amendment While baseline regressions show AI significantly reduces EI, endogeneity remains a concern. Omitted variable bias may exist if unobserved factors influence both AI development and EI. Reverse causality is also possible, as lower EI often results from industrial upgrading and improved efficiency, which can promote high-tech industry agglomeration, including AI. To address endogeneity, this study employs an instrumental variable approach and conducts two-stage least squares (2SLS) estimation. The instrumental variable (IV) is AI policy term frequency, measured by counting AI-related policy references in city government work reports. On the one hand, AI mention frequency reflects local policy support, implying subsidies and talent programs that promote AI development, thus satisfying the IV relevance condition. First-stage results (Table 4, column 5) show the IV significantly positively impacts AI. On the other hand, AI policy term frequency reflects advocacy driven by strategy adjustments, not directly affecting EI. After controlling for macroeconomic factors, no direct causal relationship exists between policy expressions and EI, nor can it affect EI through other channels. Therefore, the IV meets the exogeneity requirement and mitigates omitted variable bias and reverse causality. Column (6) in Table 4 shows the second-stage lnAI coefficient is negative and significant at 5%, confirming AI's robust EI-reducing effect after addressing endogeneity. 4.5. Heterogeneity analysis To systematically analyze heterogeneity in AI's EI impact, this study performs subgroup analyses across three dimensions (low-carbon pilot policies, digital economy policies, and green infrastructure levels), with results detailed in Table 5. 4.5.1. Low-carbon cities policy This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 12 of 19 -- Based on China's low-carbon city pilot program list, sample cities are divided into pilot and non-pilot groups. Table 5 columns (1)-(2) show that AI significantly reduces EI in pilot cities at the 1% level, but has no significant effect in non-pilot cities. These findings suggest synergies between low-carbon policies and AI enhance energy efficiency, likely due to pilot cities' first-mover advantages in green transition frameworks and infrastructure, creating a supportive environment for AI energy management applications. 4.5.2. Digital economy policy Using CNRDS digital economy policy term frequency data, the sample is split into high- and low-attention cities. Table 5, Columns (3)-(4) show the AI coefficient is -0.0396 and significant at 1% in high-attention cities, suggesting stronger digital economy policies amplify AI's green spillover effects, reducing EI. However, it is insignificant in low-attention cities. These findings highlight synergy between local digitalization strategies and AI's green empowerment. This likely arises as digital economy policies accelerate AI integration with industries and energy systems, enhancing the link between intelligent transformation and energy efficiency. 4.5.3. Green infrastructure Using CPPGD green infrastructure enterprise data, this study performs subgroup regressions for sample cities. The data cover building energy efficiency, green buildings, green transportation, green logistics, and environmental infrastructure. Table 5 columns (5)-(6) show that in high-green-infrastructure cities, AI's EI impact is -0.0303 and significant at 1%; the effect is insignificant in low-green-infrastructure cities. This likely reflects that robust green infrastructure enables efficient AI integration in energy systems, enhancing urban energy optimization effectiveness. Table 5 Tests of heterogeneity Low-carbon cities policy Digital economy policy Green infrastructure Yes No High Low High Low lnEI (1) (2) (3) (4) (5) (6) lnAI -0.0499*** 0.0056 -0.0396*** -0.0017 -0.0303*** -0.0111 (-3.9496) (0.5129) (-3.1442) (-0.1343) (-2.9509) (-0.9297) Controls Yes Yes Yes Yes Yes Yes City FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Constant 7.6871*** 3.8474*** 6.4935*** 3.2588*** 6.7644*** 2.0750** (10.8312) (4.9725) (8.4791) (3.8230) (8.9026) (2.3719) Observations 1,586 2,054 1,765 1,850 1,816 1,815 R-squared 0.937 0.912 0.939 0.922 0.933 0.913 Note: Figures in parentheses are t-statistics. ***p < 0.01, **p < 0.05, *p < 0.1. 4.6. Mechanism of action tests This section analyzes AI's impact mechanisms on EI through new energy industry development (NEI), green industry development (GRI), and green innovation (GPA). It uses counts of new energy enterprises, green enterprises, and authorized green patents as mediators to systematically evaluate AI's role in reducing EI. 4.6.1 New energy industry development This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 13 of 19 -- Table 6 column (1) shows that the lnAI coefficient for lnNEI is significantly positive at the 1% level, indicating AI significantly promotes new energy industry growth. AI improves new energy system operations and renewable energy production efficiency, supporting the sector's expansion (Bennagi et al., 2024; Alhasnawi et al., 2024). As the new energy industry expands, its substitution for traditional energy strengthens, effectively limiting EI growth (Jiao et al., 2024; Rahman et al., 2022). These results strongly support H2, which suggests AI reduces EI by advancing new energy industry development. 4.6.2 Green industry development Column (2) of Table 6 evaluates AI's effect on green industry development. For lnGRI, the lnAI coefficient is 0.0584 and highly significant, indicating AI adoption fosters green industry growth. AI improves resource and energy use, pollution control, and recycling in green manufacturing, while enabling green logistics and supply chains (Mao et al., 2019; Zhou et al., 2025). Such AI-driven green industry progress advances regional industrial upgrades and resource optimization (Pegels & Lütkenhorst, 2014; Xie & Lin, 2025), thereby reducing EI. These results strongly support H3. 4.6.3 Green innovation effect As shown in Table 6, column (3), the AI regression coefficient for green patents is positive and highly significant, indicating that AI significantly promotes green technological innovation. By strengthening data analysis, R&D, and resource allocation, AI improves green innovation efficiency at both firm and regional scales (Li et al., 2023; Guo et al., 2025). The growth in green patents accelerates technology diffusion and application, helping firms save energy, lower consumption, and increase renewable energy use, thereby reducing EI (Li et al., 2023; Wang et al., 2025; Shahzad, 2025). These results support H4. Table 6 Mechanism results. (1) (2) (3) lnNEI lnGRI lnGPA lnAI 0.0696*** 0.0584*** 0.0598*** (6.2657) (7.1436) (2.7910) Controls Yes Yes Yes City FE Yes Yes Yes Year FE Yes Yes Yes Constant -0.2184 -1.2973*** 2.8504** (-0.2922) (-2.7085) (2.2367) Observations 3,640 3,640 3,640 R-squared 0.984 0.993 0.958 Note: Figures in parentheses are t-statistics. ***p < 0.01, **p < 0.05, *p < 0.1. 5.Conclusions and policy recommendations 5.1. Conclusions Using panel data from 280 Chinese cities from 2011 to 2023, this study empirically examines AI's impact on EI and its mechanisms through fixed-effects models, robustness tests, endogeneity controls, and heterogeneity analyses. The main findings are as follows. First, AI can significantly reduce urban EI, with results robust to checks. IV estimates confirm this after addressing endogeneity, suggesting AI's practical value for energy efficiency This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 14 of 19 -- and green transformation. Second, AI's EI impact shows heterogeneity, with stronger effects in low-carbon pilot cities, digital economy cities, and areas with green infrastructure. This suggests favorable policy environments and infrastructure enhance AI's role in energy optimization. Third, the mechanism analysis shows that AI can indirectly reduce EI through three emerging pathways. AI promotes new energy development, advancing urban energy structures toward renewables; it encourages green manufacturing and logistics, fostering greener urban industries with lower energy use; and it drives green technological progress, improving urban energy efficiency. 5.2. Policy recommendations Drawing on the research results, this study proposes four policy recommendations to guide AI for EI reduction. First, enhance AI-energy system integration. Governments should improve smart energy conditions and promote AI across all stages. Enterprises should increase AI R&D using big data and algorithms to optimize renewables and smart grids, accelerating system upgrades for green transformation and EI reduction. Second, strengthen policy-green infrastructure coordination. Promote low-carbon cities and green infrastructure for AI green empowerment, and improve green logistics, transportation, and smart grids to support AI-enabled urban energy optimization and industry upgrading. Third, accelerate new energy and green industry transformation. Encourage AI across new energy value chains and integrate AI with green manufacturing and supply chains for low-carbon, intelligent upgrades. Finally, improve green innovation support. Establish green technology innovation and IP protection mechanisms. Encourage AI-driven green patent R&D and strengthen AI-enabled green innovation diffusion through policy incentives. CRediT authorship contribution statement ***: Writing–original draft, Data curation, Project administration, Visualization, Formal analysis, Software, Writing–review & editing, Methodology, Conceptualization . Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements This research is supported by the China Postdoctoral Science Foundation (No. 2024M752299) Data availability Data will be made available on request. References Aguinis, H., Edwards, J.R., Bradley, K.J., 2017. Improving our understanding of moderation and mediation in strategic management research. Organ. Res. Methods 20, 665–685. Alhasnawi, B.N., Almutoki, S.M.M., Hussain, F.F.K., et al., 2024. A new methodology for reducing carbon emissions using multi-renewable energy systems and artificial intelligence. Sustain. Cities Soc. 114, 105721. Behera, B., Behera, P., Pata, U.K., et al., 2025. Artificial intelligence-driven green innovation for sustainable development: Empirical insights from India's renewable energy transition. J. Environ. Manage. 389, 126285. Bennagi, A., AlHousrya, O., Cotfas, D.T., et al., 2024. Comprehensive study of the artificial intelligence applied in renewable energy. Energy Strategy Rev. 54, 101446. Chen, J., Gao, M., Cheng, S., et al., 2020a. County-level CO2 emissions and sequestration in China during 1997– 2017. Sci. Data 7(1), 391. Chen, L., Li, S., She, Z., 2025. A study on the impact of artificial intelligence applications on corporate green This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 15 of 19 -- technological innovation: A mechanism analysis from multiple perspectives. Int. Rev. Econ. Financ. 104490. Chen, M., Xiao, H., Li, L., et al., 2024. How does government climate risk perception affect corporate energy consumption and intensity? Energy Sustain. Dev. 81, 101496. Chen, Y., Fan, Z., Gu, X., Zhou, L., 2020b. Arrival of young talent: The send-down movement and rural education in China. Am. Econ. Rev. 110(11), 3393–3430. De Vita, G., Li, C., Luo, Y., 2021. The inward FDI-energy intensity nexus in OECD countries: A sectoral R&D threshold analysis. J. Environ. Manage. 287, 112290. De Vries, A., 2023. The growing energy footprint of artificial intelligence. Joule 7(10), 2191–2194. Dong, Y., Chen, T., Mao, J., et al., 2025. Greening through trading: The role of policy intervention on energy intensity. Energy 314, 134193. Dunyo, S.K., Odei, S.A., Chaiwet, W., 2024. Relationship between CO2 emissions, technological innovation, and energy intensity: Moderating effects of economic and political uncertainty. J. Clean. Prod. 440, 140904. Feng, Y., Zhang, J., Geng, Y., et al., 2023. Explaining and modeling the reduction effect of low-carbon energy transition on energy intensity: Empirical evidence from global data. Energy 281, 128276. Fu, Y., Shen, Y., Malin, S., et al., 2024. Does artificial intelligence reduce corporate energy consumption? New evidence from China. Econ. Anal. Policy 83, 548–561. Guan, R., Wang, H., Zheng, R., et al., 2025. Evaluating the impact of digital technologies on energy efficiency: Evidence from Chinese publicly listed companies. Energy Policy 207, 114847. Guo, L., Pei, H., Liu, Y., 2025. Artificial intelligence and corporate green innovation: Evidence from China. Res. Int. Bus. Financ. 103039. Hao, X., Miao, E., Sun, Q., et al., 2025. When climate policy's up in the air: How digital technology impacts corporate energy intensity. Energy Econ. 144, 108311. Huang, J., Wang, Y., Luan, B., et al., 2023. The energy intensity reduction effect of developing digital economy: Theory and empirical evidence from China. Energy Econ. 128, 107193. Huang, W., Miao, Y., Ye, H., et al., 2025. Trends and determinants of energy intensity in China: A study using index decomposition and econometric analysis. Environ. Sustain. Indic. 100892. International Energy Agency. (2025). Global Energy Review 2025. https://www.iea.org/reports/global-energy-review-2025 Işık, C., Yan, J., Ongan, S., 2025. Energy intensity, supply chain digitization, technological progress bias in China's industrial sectors. Energy Econ. 145, 108442. Jiao, J., Song, J., Ding, T., 2024. The impact of synergistic development of renewable energy and digital economy on energy intensity: Evidence from 33 countries. Energy 295, 130997. Jin, D., Ocone, R., Jiao, K., et al., 2020. Energy and AI. Energy AI 1, 100002. Lange, S., Pohl, J., Santarius, T., 2020. Digitalization and energy consumption. Does ICT reduce energy demand? Ecol. Econ. 176, 106760. Lee, C.C., Li, J., Yan, J., 2025. Can artificial intelligence contribute to the new energy system? Based on the perspective of labor supply. Technol. Soc. 81, 102877. Li, A., Ma, Y., Li, B., 2025a. How do climate risks affect corporate energy intensity? Evidence from China. Energy 323, 135636. Li, C., Zhang, Y., Liu, X., et al., 2025b. Does artificial intelligence promote green technology innovation in the energy industry? Energy Econ. 144, 108402. Li, H., Lu, Z., Zhang, Z., et al., 2025c. How does artificial intelligence affect manufacturing firms' energy intensity? Energy Econ. 141, 108109. Li, M., Yuan, N., Du, W., 2025d. Artificial intelligence and firm green innovation: Empirical evidence from the This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 16 of 19 -- application of robots in China. Econ. Anal. Policy 87, 2239–2253. Li, Y., Cobbinah, J., Abban, O.J., et al., 2023. Does green manufacturing technology innovation decrease energy intensity for sustainable development? Econ. Anal. Policy 78, 1010–1025. Lin, B., Xu, C., 2024. The effects of industrial robots on firm energy intensity: From the perspective of technological innovation and electrification. Technol. Forecast. Soc. Change 203, 123373. Lin, B., Zhou, D., 2025. How does the explosive growth of AI affect China's power supply and demand: A scenario simulation based on the LEAP model. Renew. Energy 124485. Liu, L., Yang, K., Fujii, H., et al., 2021. Artificial intelligence and energy intensity in China’s industrial sector: Effect and transmission channel. Econ. Anal. Policy 70, 276–293. Liu, X., Cifuentes-Faura, J., Zhao, S., et al., 2025. Impact of artificial intelligence technology applications on corporate energy consumption intensity. Gondwana Res. 138, 89–103. Liu, X., Wang, C., Wu, H., et al., 2023. The impact of the new energy demonstration city construction on energy consumption intensity: Exploring the sustainable potential of China's firms. Energy 283, 128716. Lu, F., Ma, F., Hu, S., 2024. Does energy consumption play a key role? Re-evaluating the energy consumption-economic growth nexus from GDP growth rates forecasting. Energy Econ. 129, 107268. Luan, B., Zou, H., Chen, S., et al., 2021. The effect of industrial structure adjustment on China’s energy intensity: Evidence from linear and nonlinear analysis. Energy 218, 119517. Ma, R., Lin, B., 2025. The impact of digital technology innovation on energy-saving and emission reduction based on the urban innovation environment. J. Environ. Manage. 375, 124176. Mao, S., Wang, B., Tang, Y., et al., 2019. Opportunities and challenges of artificial intelligence for green manufacturing in the process industry. Eng. 5(6), 995–1002. Matthess, M., Kunkel, S., Dachrodt, M.F., et al., 2023. The impact of digitalization on energy intensity in manufacturing sectors–A panel data analysis for Europe. J. Clean. Prod. 397, 136598. Pegels, A., Lütkenhorst, W., 2014. Is Germany’s energy transition a case of successful green industrial policy? Contrasting wind and solar PV. Energy Policy 74, 522–534. Pieters, R., 2017. Meaningful mediation analysis: Plausible causal inference and informative communication. J. Consum. Res. 44, 692–716. Rahman, M.M., Sultana, N., Velayutham, E., 2022. Renewable energy, energy intensity and carbon reduction: Experience of large emerging economies. Renew. Energy 184, 252–265. Ramalingam, S., Hassan, W.H., Subramanian, M., et al., 2025. Synergies for sustainability: Renewable energy, urban planning, and green industry in carbon emission reduction. Sustain. Futures 10, 101222. Samadhiya, A., Kumar, A., Luthra, S., 2025. Integrating generative artificial intelligence into green logistics: A systematic review and policy-oriented research agenda. J. Clean. Prod. 519, 146018. Shahzad, U., 2025. Achieving clean energy transitions: How green innovation and financial development shape energy usage. Energy Econ. 108851. Sun, H., Han, F., Liu, Z., 2024. Does executive environmental protection experience reduce enterprise energy consumption intensity? Econ. Anal. Policy 83, 652–666. Tabaa, M., Monteiro, F., Bensag, H., et al., 2020. Green industrial internet of things from a smart industry perspectives. Energy Rep. 6, 430–446. Taddeo, M., Tsamados, A., Cowls, J., et al., 2021. Artificial intelligence and the climate emergency: Opportunities, challenges, and recommendations. One Earth 4(6), 776–779. Tan, W., Huang, Y., Chen, X.H., et al., 2025. Green credit policy and energy consumption intensity in China: Firm-level evidence and insights. J. Environ. Manage. 373, 123664. Teece, D.J., 2018. Profiting from innovation in the digital economy: Enabling technologies, standards, and This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 17 of 19 -- licensing models in the wireless world. Res. Policy 47(8), 1367–1387. Ullah, U., Shaheen, W.A., Abdalkrim, G.M., et al., 2025. Past and present of energy: The role of green finance, technological innovation, and financial risk in sustainability indicators. Environ. Sustain. Indic. 100936. Umar, M., Horobet, A., Negreanu, C.C., et al., 2025. Fintech adoption, AI development, and energy transition dynamics. Energy Econ. 108905. United Nations Industrial Development Organization. (2011). Web Policies for Green Industry. https://www.unido.org/sites/default/files/2011-05/web_policies_green_industry_0.pdf United Nations. (1981). Report of the United Nations Conference on New and Renewable Sources of Energy, Nairobi, 10 to 21 August 1981. https://digitallibrary.un.org/record/25034/?v=pdf Vinuesa, R., Azizpour, H., Leite, I., et al., 2020. The role of artificial intelligence in achieving the Sustainable Development Goals. Nat. Commun. 11(1), 233. Vu, T.T.T., Demena, B.A., 2025. How does technology adoption affect energy intensity? Evidence from a meta-analysis. Appl. Energy 398, 126439. Wang, J., Duan, K., Zheng, Y., 2025. Green supply chain management, green technology innovation and firms' energy consumption intensity. Energy Econ. 141, 108133. Wang, Q., Zhang, S., Li, R., 2026. Artificial intelligence in the renewable energy transition: The critical role of financial development. Renew. Sustain. Energy Rev. 226, 116280. Wu, J.S., Niu, Y., Peng, J., et al., 2014. Research on energy consumption dynamic among prefecture-level cities in China based on DMSP/OLS nighttime light. Geogr. Res. 33(4), 625–634. Wu, L., Shi, J., 2025. From peer influence to green cognition: How digital transformation fosters renewable energy innovation in manufacturing. Energy Econ. 108691. Wu, Y., Shi, K., Chen, Z., et al., 2021. An improved time-series DMSP-OLS-like data (1992–2023) in China by integrating DMSP-OLS and SNPP-VIIRS. Harvard Dataverse, V6. https://doi.org/10.7910/DVN/GIYGJU. Xiao, Y., Duan, Y., Zhou, H., et al., 2025. Has digital technology innovation improved urban total factor energy efficiency?—Evidence from 282 prefecture-level cities in China. J. Environ. Manage. 378, 124784. Xie, J., Lin, B., 2025. Evaluating the role of green industrial policy in enhancing energy efficiency: A quasi-natural experiment based on the Five-Year Plans in China. Process Saf. Environ. Prot. 107902. Xu, X., Lieuea, B., 2025. Contribution of modern renewables to final energy consumption, energy intensity, and environmental quality: The role of world energy balances, economic performance, and shadow economies. Energy 137191. Zaidi, S.A.H., Ashraf, R.U., Hassan, T., 2024. Effects of globalization and financial inclusion on energy intensity: The case of emerging economies. Energy 306, 132380. Zhang, B., Tian, X., He, B., et al., 2025a. The impact of China's green technology transfer on energy intensity in countries along the Belt and Road. J. Environ. Manage. 373, 123691. Zhang, C., Liu, J., Abedin, M.Z., et al., 2025b. Navigating sustainable finance: Examining the impact of sustainable credit policy on energy consumption intensity. Res. Int. Bus. Financ. 73, 102594. Zhang, C., Su, B., Zhou, K., et al., 2020. A multi-dimensional analysis on microeconomic factors of China's industrial energy intensity (2000–2017). Energy Policy 147, 111836. Zhang, H., Lv, Y., Zhang, J.Z., et al., 2025c. Government science and technology spending and energy efficiency: An inverted U-shaped analysis. Energy Econ. 108818. Zhang, L., Innab, N., Shuhidan, S.M., et al., 2025d. Artificial intelligence-driven internet of things-based green supply chain for carbon reduction in sustainable manufacturing. J. Environ. Manage. 389, 126170. Zhang, W., Zeng, M., 2024. Is artificial intelligence a curse or a blessing for enterprise energy intensity? Evidence from China. Energy Econ. 134, 107561. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 18 of 19 -- Zhang, X., Khan, K., Shao, X., et al., 2024. The rising role of artificial intelligence in renewable energy development in China. Energy Econ. 132, 107489. Zhao, M., Fu, X., Sun, J., et al., 2025. Optimal strategy of artificial intelligence on low-carbon energy transformation: Perspective from enterprise green technology innovation efficiency. Energy 319, 135035. Zhao, Q., Wang, L., Stan, S.E., et al., 2024. Can artificial intelligence help accelerate the transition to renewable energy? Energy Econ. 134, 107584. Zhong, C., Cai, H., Fang, S., et al., 2025. Does artificial intelligence reduce energy intensity in manufacturing? Evidence from country-level data. Energy Econ. 108784. Zhou, C., Zhang, H., Ying, J., et al., 2025. Artificial intelligence and green transformation of manufacturing enterprises. Int. Rev. Financ. Anal. 1043 Zhu, Q., Che, J., Liu, S., et al., 2025. How can artificial intelligence technology applications accelerate energy innovation in China? Evidence from provincial regional data. Econ. Anal. Policy. This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=5854389 Preprint not peer reviewed -- 19 of 19 --
Bellwether · 2026 Marco