Smart Cities with Green Resilience: A Quasi-Natural Experiment Based on Artificial Intelligence

Highlights

What are the main findings?

• The CET policy can reduce the total energy consumption and promote the renewable energy consumption locally, with no significant influence on total energy consumption in surrounding areas. However, it causes a decrease in the renewable energy consumption ratio in neighboring regions.
• AI significantly reduces energy consumption and promotes renewable energy consumption in surrounding areas. Benefiting from AI-enabled smart city construction, the local region achieves a notable 8.55% reduction in total energy consumption, which exceeds the effect of implementing CET policy alone.

What is the implication of the main finding?

• The CET policy of cities exerts a catalytic effect, increasing energy consumption and carbon emission costs in local regions, promoting energy structure transformation, while avoiding the relocation of high-energy-consuming enterprises to surrounding areas. However, due to the “siphoning effect”, the policy absorbs renewable resources from neighboring regions, necessitating enhanced coordination with adjacent areas.
• AI can break down regional barriers through spatial effects, fostering cross-regional spillovers of green concepts and the application of green technological innovations, thereby counterbalancing the “siphoning effect” and facilitating the formation of a green smart city cluster. Smart city development enables the compatibility of “green resilience” and “smart functionality”.

Abstract

Amidst climate change and the energy crisis worldwide, the synergy between smart city and environmental policies has become a key path to improving the green resilience of cities. This study examines the spatial effects of carbon emission trading (CET) policy on urban energy performance under the context of artificial intelligence (AI)-empowered smart cities. Using the spatial Durbin model (SDM) and analyzing data from 262 Chinese cities covering the period 2013–2021, the results reveal that: (1) smart cities significantly benefit from the institutional support of the local CET policy, resulting in an 8.55% reduction in energy consumption in the pilot city; (2) AI advancement contributes directly to reducing energy consumption in surrounding areas by 21.84% through spatial effects, and compensates for the imbalance of regional renewable energy caused by the “siphon effect” of CET policy. This study provides empirical evidence for developing countries to build green and resilient cities. This paper proposes the need to build a national CET market, strengthen government supervision, and make reasonable use of AI technology, transforming the green and resilient model of smart cities from Chinese experience to global practice.

1. Introduction

Under the dual challenges of global climate change and accelerating urbanization, smart cities and green resilience development have become the core issues for achieving a sustainable future. How to agilely balance the tech-economic advancement and environmental resilience has become a hot issue in academic discussion. As a large manufacturer and the largest developing country, China, like many other industrialized countries, has experienced a path of high energy consumption in exchange for economic development. China realized early on that it is difficult to achieve a synergy in emissions reduction and system resilience in the traditional way, and that the process of urban construction needs to realize the integration of “green resilience” and “smartness”. Smart cities are able to reshape the logic of urban operation through AI, the Internet of Things (IoT), building a closed-loop system of “perception-analysis-decision making-optimization” to advance energy management and strengthen the infrastructure’s resilience, building green and smart cities. China clearly proposes to promote the construction of “Digital China” and “New Smart Cities” in its “14th Five-Year Plan”, aiming to build a green, low-carbon, resilient, and adaptive city system through technology advancement. In this context, smart cities, with their data-driven, technology-integrated, and dynamically adapted characteristics, are becoming an important tool for building a “low-energy, high-resilience” urban ecosystem in response to climate change and the energy crisis.

Since Holling first proposed “ecological resilience” in 1973 [1], scholars have defined the concept from diverse perspectives. In 2015, MIT Professor Vale further synthesized and critically analyzed the notion of “urban resilience”, arguing that the term resilience can serve two purposes in urban planning and design practice: (1) as a design principle to create cities that minimize damage and enable rapid comprehensive recovery, and (2) as an analytical tool to evaluate the effectiveness of plans and policies [2]. Vale distinguished two types of urban planning and policies: reactive/restorative (aimed at avoiding losses from disasters) versus proactive/preventative (focused on active prevention). Proactive/preventative resilience requires upfront investments and entails difficult trade-offs regarding which portions of the environment’s protection to fund—and consequently, which populations benefit. The “green resilience” concept in this study aligns with Vale’s definition, specifically through AI-driven smart city development that both proactively informs green urban planning decisions and facilitates rapid recovery when urban green systems face fluctuations [2].

In the era of Industry 4.0, AI is deeply integrated with various industries, subverting the traditional crude development mode, and becoming an important tool for building smart cities. AI adopts methods such as machine learning, deep learning and replicates human cognitive functions through programming and algorithms, including learning, reasoning, perception, and autonomous decision-making [3]. It can also effectively replace some of the human physical and cognitive labor aspects and provides users with agile assistive capabilities [4]. The energy sector is an important application of AI today, but the correlation between them remains mixed. Some scholars consider that AI has become an important tool for improving energy efficiency [5], capable of forecasting energy consumption, optimizing energy management systems and production processes to significantly lower the intensity of energy consumption, and promoting renewable energy use [6,7,8,9]. Badareu points out that investments have significantly increased renewable energy consumption in AI [10], and can raise the energy efficiency of companies [11]. China has been actively using AI in various fields for efficiency improvement and green transformation. Since 2016, with the development of deep learning, there have been 219 patents focusing on energy management and sustainable driving, of which more than 70% are registered in China [12]. In terms of smart city construction, Shenzhen has significantly increased the rate of renewable energy consumption by matching supply and demand in real time through smart grids; Hangzhou has significantly reduced carbon emissions by relying on its city brain to optimize traffic flow. Therefore, AI has promoted the reorganization of the energy framework and industrial upgrading. However, some scholars have pointed out that due to the high power consumption of AI computing [13,14], it may lead to intensified energy consumption. For Google, OpenAI, Microsoft, and other companies, despite setting ambitious carbon reduction targets, achieving carbon neutrality by 2030 still faces challenges [15,16]. In addition, AI technologies are mostly realized on web-based platforms, which may have regional spillover effects. Therefore, a more comprehensive understanding of AI-dependent smart city building will help improve the green resilience of cities. In recent years, China has incorporated the reduction of energy consumption into its “dual-carbon” goals—“carbon peaking” and “carbon neutrality”. Notably, the introduction of the CET policy is critical to build a market-driven green economic system. This policy accelerates the transition to renewable energy sources, making it a critical component and a vital pathway to realizing the “dual-carbon” goal [17]. To promote the green economy development, it is essential to clarify green property rights. The CET policy contributes to solving this core problem, thus boosting energy efficiency and enhancing the green resilience of the city. The rationale behind CET policy can be traced back to Coase’s theorem [18], which advocates to govern the externality by clarifying property rights and applying market mechanisms to optimize resource allocation. Specifically, the principal function of CET is to identify the scarcity of carbon emission quotas. The government allocates a limited number of emission quotas to the emitters, and these quotas generate profits or losses. As a result, emitters trade their carbon quotas freely in the carbon market. This process aligns the private marginal cost of carbon emissions with the social marginal cost, which can reduce the energy consumption and promote energy structure upgrading. Hong et al. examined 276 China cities (between 2003 and 2016) and found that CET policy helps to improve energy efficiency [19]. Tan et al. found that the CET systems have reduced energy consumption [20]. In addition, Huang et al. found that the CET policy reduces the price gap between renewable energy sources and fossil fuels, which positively affects electricity consumption and increasingly favors renewable energy [21]. Pietzcker et al. found that tightening emission targets will accelerate the transformation of different sectors within the electricity system over a period of 3 to 17 years under the European Emissions Trading System (EU-ETS) [22]. By 2030, this could result in the near-total cessation of coal use across the European Union, with renewable energy sources supplying 74% of electricity [22]. Renewable energy utilization, in turn, leads to lower CO2 emissions [23]. Therefore, the CET policy significantly decreases reliance on non-renewable energy, raising renewable energy consumption through market-based mechanisms. Theoretically, imposes additional carbon pricing costs on enterprises, thereby elevating the marginal cost of fossil fuel-based electricity generation, consequently enhancing the relative competitiveness of renewable energy [24]. Additionally, CET generates additional revenues for renewable energy projects [25], and the increased profitability of such projects will attract more investment.

China’s CET policy has been gradually implemented in pilot administrative jurisdictions since 2013. In 2021, the national online CET market officially commenced operations, initially targeting the power generation sector, thereby establishing the world’s largest carbon market. The national voluntary greenhouse gas emissions reduction trading market was then formally launched in 2024, becoming another important institutional support to achieve the “dual-carbon” goal [26]. Data show that by the end of 2024, China’s CET market quota cumulative turnover was 630 million tons, a cumulative turnover of CNY 43.033 billion. Against a general decline in the global carbon price, China’s market trading price rose steadily, with the end of the year closing price of 97.49 CNY/ton; compared with the end of 2023, it rose by 22.75% [27]. After more than a decade of development, China’s CET market has progressed from the pilot stage to the national CET market stage, and is now experiencing stability and growth.

The following gaps exist in the current research: Firstly, due to China’s relatively late development of the carbon market compared to developed countries, few studies have assessed the effects of CET, particularly at the city level. There is a notable lack of long-term studies and a shortage of city level data on energy consumption and renewable energy usage, especially at a more granular geographic scale. Secondly, extant research has primarily concentrated on the independent impact of policies or technologies, ignoring the function of smart cities as a synergistic platform of policy, technology, and resilience. Additionally, few papers discussed the mechanisms through which smart cities break down administrative boundaries and realize cross-regional green externalities. Our work seeks to address the aforementioned gaps. We primarily concentrate on exploring two questions: Do CET and AI reduce energy consumption in the region and promote energy structure transformation? How do AI-driven smart management systems enhance the green resilience of cities?

The subsequent paper is structured as follows: Section 2 is the literature review, Section 3 details the data, Section 4 describes the research methodology and results analysis from the perspectives of green resilience and smart cities, and Section 5 draws policy implications and makes recommendations for further research.

2. Literature Review

This study examines the influence of CET policy on the green resilience of cities under the scenarios of AI empowerment. This topic is related to two types of literature: the institution-based view under the CET context, and the AI-based ecosystem-specific advantages.

2.1. Institution-Based View Under the CET Context

At present, China’s CET market has two basic products, one of which is carbon emission quotas assigned by the government to firms, and the other is Certified Voluntary Emission Reductions (CCERs), which refer to the quantitative verification of the greenhouse gas emission reduction effect of renewable energy, methane utilization, and other projects. As a result, two types of transactions are derived. Firstly, in the context of quota trading, when a firm’s carbon emissions are less than the government-allocated quota, the “quota-surplus” firm can sell the remaining quota to a “quota-deficit” enterprise. This mechanism facilitates the optimal distribution of carbon allowances across various entities, fostering a more balanced and efficient market for carbon emissions reductions and controlling emissions. Second, voluntary emission reduction market trading (CCER trading). That is, emission-control enterprises purchase certified emission reductions (CERs) that can be used to offset their own carbon emissions from enterprises that implement carbon offsetting activities. Through the above two trading methods, China’s CET market realizes the division of property rights, and equips the CET market with the functions of optimal allocation of resources and price discovery, thus providing conditions for the rapid development of a green economy. Therefore, it is fundamental, forward-looking, and highly efficient for China to take the construction of the CET market as an important hand of China’s energy policy.

This paper is supported by studies related to CET energy policy and energy consumption. Summarizing previous studies, CET policy can influence energy consumption through green innovation effect and price effect, i.e., influencing corporate behavior with the institutional advantages of a CET market, reducing energy consumption, and promoting renewable energy use. Regarding the green innovation effect, the research on the externality effect points out that serious pollution is caused by the existence of an externality, i.e., the private marginal cost of pollution is different from the social marginal cost. Furthermore, CET policy raises the energy consumption cost, and promotes the convergence of the private marginal cost and the social marginal cost, which leads to the following effects on the buyers and the sellers: for the buyers in the CET market, high energy-consuming enterprises need to buy carbon emission allowances from other enterprises if they want to expand their production scale, and the larger the allowances they buy, the stronger the incentive for firms to undertake energy transformation and green innovation to improve energy efficiency; for sellers, low-energy-consuming enterprises utilize a high percentage of renewable energy and green technology to achieve a surplus of allowances, obtain surplus factor resources and trade the surplus allowances in the CET market to gain profits, and accelerate green technology R&D under the guidance of profits [19]. Prior cross-national studies utilizing cross-country panel data found that the CET policy has increased renewable energy production [28].

Therefore, the CET policy improves the competitiveness of low-energy-consuming enterprises or incentivizes green innovations to improve the energy efficiency of enterprises through market-oriented autonomous interactions between buyers and sellers, and promotes the utilization of green technologies and renewable energies, thus reducing the energy consumption level. Therefore, enhancing energy efficiency is a highly efficacious strategy for achieving long-term green growth [19]. Many current studies confirm the green innovation effect of the CET market; Lu et al. argued that the CET will render low-emission units more competitive [29]. From a long-term perspective, environmental regulations can encourage companies to engage in innovation, potentially offsetting the costs of compliance and enhancing productivity [30,31]. The second effect of CET policy affecting energy consumption is more direct price transmission. A multitude of studies have reached the conclusion that there exists a significant correlation between the energy and CET market, which is manifested in the price correlation and dynamic volatility between the two markets. Ren et al. reached the conclusion that the price volatility of the carbon market exerts a significant impact on the energy market [32], and Yousaf et al. showed that financial flows in green finance markets significantly affect the price structure of energy markets, especially crude oil and renewable energy [33]. Therefore, the CET market can directly influence firms’ various types of energy consumption and energy consumption expectations by affecting the prices and volatility of various types of energy. Accordingly, we hypothesize the following:

H1. 

CET has a negative effect on primary energy consumption.

H2. 

CET has a positive effect on encouraging renewable energy usage.

Moreover, some research has demonstrated that the carbon emissions of developing countries are among the highest worldwide and are progressively emerging as the primary contributor to global carbon emissions [34]. The scarcity of energy and environmental degradation are progressively impeding economic growth [35]. China has taken the initiative to assume the responsibility of energy conservation, and has made controlling energy consumption an important part of its national strategy; at present, the CET market has been gradually established and perfected, with steadily increasing market vitality and strong digital infrastructure safeguards. However, research specifically focusing on the relationship between CET markets and energy consumption, especially in the Chinese context, is still limited. China’s “three-step” policy system (i.e., “CET pilot market—national online CET market—national voluntary greenhouse gas emissions reduction trading market”) and quantification system based on emissions intensity rather than absolute emissions have unique characteristics and are of great interest to developing countries. This system is not only unique, but also highly relevant for developing nations. A thorough and systematic study of the impacts of this policy is crucial for the establishment and growth of CET markets in these countries. Xuan et al. demonstrated the positive impact of China’s CET market on carbon emission reduction by using provincial-level panel data [35]. Chen et al. examined CET and energy efficiency correlation, and demonstrated that the China’s CET policy improves energy efficiency through innovation and increased marketization levels [36].

The above studies have made useful contributions to the evaluation of the effects of China’s CET policy, and provide important references for this paper. However, as the research on the “pilot” policy itself has shown, the pilot may cause changes in the expectations of participants in the local and neighboring regions. For the construction of a CET market, this may result in the relocation of companies from the pilot cities to neighboring cities, thereby increasing the level of energy consumption in the surrounding areas. China has been practicing the CET policy for a long time, so the policy may have certain spatial spillover effects. At the same time, the non-rivalry [37], multiplier effect of data elements, and the de-spatialization of digital technology, especially production elements such as knowledge [38], which is helpful in strengthening the linkage between firms and improving the effectiveness of the system, and also has the potential to promote the CET center city’s green innovation technology spillover. Therefore, the spatial effects of CET policy are still unclear.

2.2. AI-Based Ecosystem-Specific Advantages (ESA)

Existing studies have shown that AI is beneficial in promoting energy consumption optimization, energy transition, and carbon reduction [39]. Specifically, the above effects are formed based on the ecosystem competitive advantages of AI. The core advantage of ecological competition lies in breaking the exclusivity and monopolistic nature of individual entities, thereby reorganizing resource allocation patterns. This advantage is openly accessible, as the degree to which firms gain from the ecosystem is determined by their efficiency in leveraging its strengths. Consequently, the CET policy is inherently open, but its pivotal role lies in creating universal inclusive effects that amplify the system’s positive externality mechanisms. The CET policy pulls the whole industrial chain one after another, forcing the upstream from the downstream, and the upstream innovation leads the downstream, and forming a greening and environmentally friendly way of resource allocation under the complementary roles of each subject to achieve the overall systematic effect of the optimization of resource allocation. It also breaks the traditional crude negative externality development model that relies solely on the exclusive use of a specific source of energy in to achieve economic development. Compared with ecosystems containing homogeneous or unrelated activities, system participants are more likely to bind to and invest in ecosystems because other participants and their activities make their products more valuable [40]. Full of a diversity of complementarians, AI ecosystems accommodate a large number of complementarians for efficient operation within the ecosystems, which helps to optimize the production and operation of enterprises, and the operation’s resource allocation and organization in a way that forms efficient upstream and downstream links in the industry chain.

Additionally, AI exerts its spatial spillover effect [41,42,43], breaks down spatially formed regional barriers, realizes a wider range of connecting complementarities, helps reduce the problem of information asymmetry and transaction costs faced by cross-regional enterprises [44], and promotes complementary cooperation at the individual, enterprise, industry, and city levels. Specifically, at the talent level, it promotes the exchange of knowledge among scholars, which is helpful for the research and development of green technological achievements [45], and promotes the regional spillover of green ideas.

At the enterprise level, reduced costs in deploying green technologies enable small- and medium-sized enterprises (SMEs) to realize sustainability goals previously hindered by financial limitations. Enhanced inter-regional collaboration fosters cross-provincial and municipal diffusion of eco-innovations, driving green transformation while improving the profitability of sustainable technologies. Replicable application scenarios further stimulate local and neighboring green technology markets, generating systemic positive externalities through scalable and interconnected frameworks. Research has shown that one of the major structures of ESA is the heterogeneous resources and distributed innovation contributed by ecosystem participants; the more complementary resources there are, the more valuable the ecosystem [40]. For a long time, green innovation has been characterized by high investment risks, long R&D cycles, and large R&D costs, and many companies lack enthusiasm for participating in green innovation activities [46]. The CET policy with attached AI technology realizes more extensive system heterogeneity resource connection and deployment, promotes distributed innovation of enterprises in the energy field, makes new knowledge and processes in the ecosystem diffuse, and helps to promote green technological innovation of the industrial chain and the improvement of the new quality productivity.

At the industry level, AI promotes the complementarity between the financial industry and other industries, giving rise to a series of new business models through the combination with the financial sector, such as digital finance and other financing methods [47], which effectively reduces information asymmetry, and is conducive to the access of companies to green financing support [48]. In particular, the Chinese government tends to provide financing preferential policies the financial field for green development enterprises, providing more flexible, customized services, which effectively reduces the financing constraints of enterprises, and is conducive to reducing the cost of green R&D for businesses in and improving the efficiency of energy use. Relying on data-driven decision-making at the city level, cross-regional energy demand dispatch is optimized through IoT, blockchain, and other technologies. For example, the carbon data platform tracks corporate carbon footprints in real time, enabling dynamic adjustment of quota allocation. China’s power industry has begun to implement “AI and other related technologies to enable communication between smart grids, smart meters and IoT devices”. Smart grids can balance the production and consumption of electricity, optimizing the utilization of renewable resources [49]. In logistics, AI can reduce energy consumption and improve transportation efficiency by matching closer suppliers and optimizing transportation routes and scheduling. Therefore, we hypothesize that:

H3. 

AI exhibits a green spatial effect in reducing primary energy consumption.

H4. 

AI exhibits a green spatial effect in promoting renewable energy usage.

With the rapid development of digital intelligence technology, AI greatly accelerates digital integration and industrial upgrading, and expands through technological innovation the positive externalities of system innovation. As emphasized by Lyu and Liu, the broad adoption of AI in the energy sector takes advantage of its information sharing, making it a very valuable technology [50]. AI’s wide range of energy application scenarios, including macro and energy trend assessment, demand forecasting, risk management etc. [51], have effectively improved production efficiency and information processing capabilities [11]. Therefore, AI effectively improves the information processing and forecasting capabilities of the CET system as a whole and its participants, facilitates policy makers to dynamically assess the emission reduction capabilities of enterprises, promotes the intelligence of urban construction and energy testing systems, and has become a catalyst in the process of designing and optimizing the supply and utilization of smart services in urban spaces [52]. Under the influence of AI, energy monitoring systems may undergo several revolutions [53]. Overall, AI-based ecosystem-specific advantages provide a wide range of innovative heterogeneous resources that contribute to multiplying CET’s institutional effects and thus reducing energy consumption and promoting energy structure upgrading. Therefore, the following hypotheses are proposed:

H5. 

The co-function of AI and CET has a negative effect on primary energy consumption.

H6. 

The co-function of AI and CET has a positive effect on increasing renewable energy usage.

3. Data and Statistics

3.1. Data

3.1.1. CET Policy and AI Data

In the pilot phase of CET policy, seven provinces and cities, including Beijing, Shanghai, etc., were set to become CET pilot cities. In fact, along with the seven pilot cities mentioned above, Sichuan Province is the first non-pilot area and the eighth province in the country to carry out CET, and the products traded are mainly allowances (carbon emission credits allocated by the carbon trading authority to key emission units for a set period) and CCERs (voluntary greenhouse gas emission reductions, which can be used to offset the allowances), but few studies in the past have included Sichuan in the scope of the study, and therefore it is easy to bias the estimation results. So, the actual implementation of the CET prevails, and Sichuan Province is included in the experimental group, and the time when online trading starts in the eight pilot provinces and cities above is taken as the policy shock point, and other non-implemented cities are taken as the control group.

The level of AI is measured as the logarithm of the number of AI firms in the city in that year, normalized (mmx_AI). The AI data are sourced from Qichacha and collected by https://www.macrodatas.cn/ (accessed on 8 April 2025). The main method is to use Python 3.12 crawler technology to analyze the “business scope” and “company name” columns of Qichacha. The keyword fuzzy matching query is based on “Intelligence”, “IoT”, “Machine Learning”, “Cloud”, “Data” and other contents related to AI applications, and then each is summarized by year and region to the panel data of AI enterprise cities, and the prefecture-level of AI development is measured.

3.1.2. Energy Consumption Data

Chinese city-level energy consumption data are scarce, and this paper refers to the final energy consumption data compiled and calculated by Yang et al. [54] from the National Bureau of Statistics of China (NBSC), China Electric Power Yearbook (CEPY), China Energy Statistical Yearbook (CESY). In order to sum up the various types of energy consumption, all energy consumption in each region is converted to 10,000 tons of standard coal, including final heat consumption, final other energy consumption, final energy consumption of coal total, etc.

3.1.3. Renewable Energy Data

Regional energy consumption shows the regional overall level of energy use, while the energy consumption structure further reflects the sustainability of regional energy development and the level of energy transition. Few previous papers have explicitly concentrated on final renewable energy consumption at the city level. Only Chen et al. compiled an energy consumption inventory in China from 1997–2017. Li et al. compiled an energy consumption inventory in the Central Plains. Liu et al., using a downscaling method, compiled a fossil energy consumption inventory from 2003–2019 [54,55,56,57]. To further explore whether CET promotes the regional energy consumption structure upgrading, this paper refers to Yang et al.’s [54] renewable energy consumption data to explore the CET effect on renewable energy consumption, and then to derive the transition level of regional energy. The renewable resources mainly include hydropower, solar, etc. We use the renewable energy ratio of energy consumption to the total to measure the regional energy transition.

3.1.4. Socio-Economic Data

The socio-economic data at the city level are mainly from the China City Statistical Yearbook. In this paper, we use GDP to determine the regional economic development level, the foreign investment amount actually used in the year to measure the degree of openness, the road passenger traffic volume to measure the strength of inter-city links, the number of private and self-employed workers in towns and cities to measure the level of urbanization, the total merchandise sales of the wholesale and retail trade sector above the quota to measure the level of regional consumption, the regional science expenditure to measure the emphasis on science and technology, and the share of secondary industry value-added to measure the industrial structure.

Table 1. Variables definition.

Variable Type	Title 2	Title 3	Title 4
Dependent variable	Regional energy consumption	Regional energy consumption (tons of standard coal) divide by area of administrative division (with normalization)	lnenergy	Yang et al. [54]
	Regional energy transformation level	Total energy consumption of wind power, hydropower, solar energy/total energy consumption	renewable	Yang et al. [54]
Independent variable	Carbon trading cities	1 for carbon credit-trading cities after the policy shock point and 0 for non-carbon credit-trading cities	CET	Provincial government networks
	Regional artificial intelligence levels	Logarithmic value of the number of AI firms (with normalization)	AI	Regional statistical offices
	CET cities and artificial intelligence compound role	The interaction term of CET city variables and regional AI levels	CET*AI	work out
Control variable	Regional transportation levels	Logarithmic road passenger traffic (ten thousand people)	lntrans	Regional statistical offices
	Level of regional economic development	Logarithmic value of gross regional product (ten thousand CNY)	lnGDP
	Regional openness	Logarithmic value of the amount of foreign capital actually utilized (ten thousand USD)	lnfdi
	Level of regional urbanization	Logarithmic value of the number of urban private and self-employed persons (ten thousand people)	lnurb
	Regional consumption levels	Logarithmic value of total merchandise sales of wholesale and retail trade above the limit (ten thousand CNY)	lnconsume
	Regional emphasis on science and technology	Logarithm of regional science expenditure (ten thousand CNY)	lnsciout
	Regional industrial structure	Share of secondary industry value-added in total regional value-added	secper

and

Table 2. Summary statistics.

Variable	N	Mean	Std.dev	Min	Max
CET	2358	0.201	0.401	0	1
re	2358	0.0674	0.0494	0.00673	0.264
lnenergy	2358	0.435	0.189	0	1
AI	2358	0.476	0.150	0	1
lnsciout	2358	10.66	1.429	6.624	15.53
secper	2358	0.441	0.104	0.107	0.794
lnGDPlntrans	2358	16.77	0.905	14.54	19.88
	2358	8.066	1.117	2.303	12.18
lnfdi	2358	10.01	2.064	1.099	14.94
lnurb	2358	3.852	0.979	0.0200	7.119
lnconsume	2358	15.68	1.460	5.333	21.21

present the variable definitions and descriptive statistics, respectively.

4. Research Methodology

4.1. Modeling

This paper uses spatial econometric models to analyze the influence of CET policy on regional total energy consumption and renewable energy. The data cover the period 2013–2021 and includes 262 prefecture-level cities in China. The spatial econometric model mainly uses SDM. The specific settings are as follows:

$$\begin{gathered} {\mathrm{l}\mathrm{n}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{g}\mathrm{y}}_{it}={\beta }_0+\mathsf{\rho }\mathrm{W}·{\mathrm{l}\mathrm{n}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{r}\mathrm{g}\mathrm{y}}_{it}+{\beta }_1{\mathrm{C}\mathrm{E}\mathrm{T}}_{it}+{\beta }_2{AI}_{it}+{\beta }_3{\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}}_{it}+{\theta }_1\mathrm{W}·{\mathrm{C}\mathrm{E}\mathrm{T}}_{it} \\ +{{\theta }_2\mathrm{W}·{AI}_{it}+\theta }_3{\mathrm{W}·\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}}_{it}+{\mathsf{\lambda }}_i+{\mu }_t+{\mathsf{\epsilon }}_{it} \end{gathered}$$

(1)

$$\begin{gathered} {\mathrm{l}\mathrm{n}\mathrm{r}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{w}\mathrm{a}\mathrm{b}\mathrm{l}\mathrm{e}}_{it}={\gamma }_0+\mathsf{\rho }\mathrm{W}·{\mathrm{l}\mathrm{n}\mathrm{r}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{w}\mathrm{a}\mathrm{b}\mathrm{l}\mathrm{e}}_{it}+{\gamma }_1{\mathrm{C}\mathrm{E}\mathrm{T}}_{it}+{\gamma }_2{AI}_{it}+{\gamma }_3{\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}}_{it}+{\tau }_1\mathrm{W} \\ ·{\mathrm{C}\mathrm{E}\mathrm{T}}_{it}+{{\tau }_2\mathrm{W}·{AI}_{it}+\tau }_3{\mathrm{W}·\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}}_{it}+{\mathsf{\lambda }}_i+{\mu }_t+{\mathsf{ϵ}}_{it} \end{gathered}$$

(2)

where $i,t$ represent city and year, respectively, $\beta ,\theta ,\rho ,\gamma ,\tau$ are the coefficients to be estimated, $Control$ is the control variable, and ${\mu }_t,{\mathsf{\lambda }}_i$ denote the time and individual fixed effect, respectively. ${\mathsf{ϵ}}_{it}$ denotes the random disturbance term, and W is the spatial weight. The coefficients ${\beta }_1 ,{\gamma }_1$ indicate the influence of CET cities on local energy and renewable energy usage, respectively; and the coefficients ${\theta }_1,{\tau }_1$ indicate the influence of the CET policy on adjacent cities’ energy consumption and renewable energy consumption, respectively, reflecting the spatial spillover effect.

Due to the fact that the impacts of the CET policy and AI on energy consumption and energy structure in surrounding regions involve both distance and economic factors, the economic geography matrix can organically integrate the two, providing a more accurate depiction of the comprehensiveness and complexity of spatial effects. This paper constructs the economic geography distance spatial weight matrix. To assess the spatial correlation, the global spatial autocorrelation coefficient Moran’s I is employed to quantify the degree of spatial correlation across Chinese prefectural-level cities. The specific formula is as follows:

$$Mora{n}^{\prime }s I={\displaystyle \frac{n\sum _{i=1}^n\sum _{j=1}^n{W}_{ij}\left(y_i-\overline{y}\right)\left(y_j-\overline{y}\right)}{(\sum _{i=1}^n\sum _{j=1}^n{W}_{ij})\sum _i^n{\left(y_i-\overline{y}\right)}^2}},$$

(3)

In

Table 3. Global Moran test.

Variable/Year	Energy Consumption		Energy Transformation
	Moran’s I	p-Value	Moran’s I	p-Value
2013	0.307	0.000	−0.118	0.000
2014	0.312	0.000	−0.139	0.000
2015	0.310	0.000	−0.144	0.000
2016	0.303	0.000	−0.150	0.000
2017	0.305	0.000	−0.131	0.000
2018	0.305	0.000	−0.126	0.000
2019	0.307	0.000	−0.124	0.000
2020	0.305	0.000	−0.122	0.000
2021	0.313	0.000	−0.115	0.000

, the overall Moran’s I of energy consumption levels in prefecture-level cities across China is significantly positive, while the Moran’s I of renewable energy is negative. This demonstrated that the energy consumption levels and final energy consumption levels of renewable energy in prefecture-level cities have significant spatial dependence, so using a spatial econometric model to explore the energy consumption mechanism in prefecture-level cities across China is more accurate than a general panel model.

In this paper, Moran scatter plots (Figure 1) of the energy consumption levels of Chinese prefectural cities in 2014 and 2021 are plotted to explore the clustering patterns of the energy consumption levels of Chinese prefectural cities. In the Moran scatter plots, there are a majority of cities that are in the H-H clustering (Quadrant I) and the L-L clustering (Quadrant III), which show that there is a clustering effect between cities with higher energy consumption levels and between cities with lower energy consumption levels. The results indicate a pronounced clustering effect among cities with high energy consumption levels, while cities with low energy consumption levels exhibit a similar spatial pattern.

4.2. Selection of Spatial Measurement Models

Fixed effects models can control for unobservable heterogeneity and improve estimation accuracy. We first judge whether a spatial econometric model analyzing the impact of incorporating the co-function of AI and CET policy versus not incorporating the co-function of AI and CET policy on regional energy consumption should employ an individual, a time, or a spatio-temporal double fixed-effects model. That is, the log-likelihood values of the individual and the time spatial fixed-effects model are calculated respectively, then the log-likelihood values of the spatio-temporal dual fixed models are calculated, and the likelihood ratio statistic is further constructed, and finally the probability values are computed to judge exactly which model is chosen. Subsequently, this study conducts Likelihood Ratio (LR) tests on the aforementioned models, and all obtained p-values are 0.0000, indicating that the joint/time individual fixed effects are not significant and are rejected at the 1% level, i.e., we should use the spatio-temporal dual fixed-effects model. This means that the spatio-temporal dual fixed-effects model should be used. We further employs LR tests to determine whether the SDM with spatio-temporal double fixed effects outperforms the Spatial Lag Model (SAR) and Spatial Error Model (SEM). In models without CET-AI interaction terms, the LR statistics are 130.22 and 117.18, both significant at the 1% level, indicating the superiority of SDM over SAR and SEM, respectively. In models incorporating CET-AI interaction terms, the LR statistics increase to 145.58 and 137.63, respectively, also significant at the 1% level. Consequently, the SDM is adopted as the primary framework for analysis.

4.3. Empirical Analysis

We use spatio-temporal dual fixed-effects SDM to conduct benchmark regression (

Table 4. Spatial regression results of total energy consumption.

	(1)	(2)	(3)	(4)
Variable	Model 1	Model 2	Model 3	Model 4
CET	−0.0228 ***	0.0145 *	−0.0210 ***	0.0218 ***
	(−6.2684)	(1.7304)	(−8.7450)	(3.5348)
AI	0.0010	−0.0157	0.0166	−0.0001
	(0.0203)	(−0.3377)	(0.6768)	(−0.0023)
CET×AI		−0.0739 ***		−0.0855 ***
		(−4.5536)		(−7.6374)
lntrans	−0.0026 *	−0.0030 **	−0.0026 ***	−0.0030 ***
	(−1.7164)	(−2.0420)	(−2.6564)	(−3.1777)
lnGDP	0.0831 ***	0.0872 ***	0.0892 ***	0.0944 ***
	(6.2843)	(6.6279)	(23.6826)	(24.9388)
lnfdi	0.0001	−0.0002	−0.0000	−0.0003
	(0.0527)	(−0.1464)	(−0.0431)	(−0.4900)
lnurb	0.0018	0.0021	0.0015	0.0018
	(0.9110)	(1.0823)	(1.3393)	(1.5967)
lnconsume	0.0024 *	0.0021	0.0026 ***	0.0022 **
	(1.7212)	(1.5372)	(2.6895)	(2.2822)
lnsciout	−0.0045 **	−0.0043 **	−0.0035 ***	−0.0034 ***
	(−2.3529)	(−2.2724)	(−3.5272)	(−3.4893)
secper	0.3196 ***	0.3186 ***	0.3081 ***	0.3037 ***
	(8.8094)	(8.8568)	(24.8606)	(24.6741)
W×CET			0.0066	−0.0063
			(0.9312)	(−0.3681)
W×AI			−0.2184 ***	−0.2441 ***
			(−3.2611)	(−3.6475)
W×CET×AI				0.0411
				(1.4545)
W×lntrans			−0.0033	−0.0031
			(−1.3285)	(−1.2503)
W×lnGDP			0.0604 ***	0.0497 ***
			(6.0649)	(4.9371)
W×lnfdi			0.0016	0.0020
			(1.1342)	(1.4043)
W×lnurb			0.0056 **	0.0066 **
			(2.0527)	(2.4330)
W×lnconsume			0.0013	0.0011
			(0.5782)	(0.4854)
W×lnsciout			0.0051 *	0.0058 **
			(1.9595)	(2.2155)
W×secper			0.1331 ***	0.1497 ***
			(3.8185)	(4.3261)
cons	−1.0720 ***	−1.1245 ***
	(−5.2593)	(−5.5385)
Rho			−0.1766 ***	−0.1633 ***
			(−4.5507)	(−4.2094)
	0.9895	0.9897	0.2553	0.1727
			0.0003 ***	0.0003 ***
			(34.2487)	(34.2620)
N	2358	2358	2358	2358

Note: ***, **, and * indicate statistical significance at 1%, 5%, and 10% levels, respectively; the t-statistics are reported in parentheses.

), with Column (1) as the ordinary panel high-dimensional fixed-effects (HDFE) model and Column (3) as SDM. To further investigate whether smart cities enhance the effectiveness of green policies through self-organizing effects, this study incorporates the CET policy-AI interaction term and conducts both HDFE model regressions and SDM regressions (Columns (2) and (4)). In Columns (3)–(4), the spatial correlation coefficient (rho) for regional energy consumption levels is statistically significant, indicating a negative spatial spillover effect in energy consumption. This suggests that local energy consumption is inversely related to neighboring regions’ energy consumption, due to factors such as cross-regional energy transmission. These findings confirm the appropriateness of employing the SDM for analysis. Columns (1)–(3) show that the coefficient of CET policy is significantly negative, supporting H1. It can be assumed that the green policies of each city will play a catalytic effect to increase energy consumption cost, narrow the divergence between private and social marginal carbon costs, enhance the competitiveness of low energy-consuming enterprises through the market-oriented autonomous interaction between buyers and sellers, catalyze the enterprises to raise their energy efficiency, and ultimately reduce the level of energy consumption in the region, which is same as the theoretically derived previous results. From Column (3), it can be concluded that CET has no obvious impact on the energy consumption of the neighboring regions, did not raise the energy consumption level of the neighboring regions, did not cause the relocation of high energy-consuming companies, and China’s CET policy has a good effect.

At the same time, the ecological competitive advantage of AI begins to appear. AI reduces the energy consumption of the neighboring regions (Column (3)), which means that AI can break the regional barriers through spatial effects, realize a wider connectivity of complementarities, and promote the cross-regional spillover of green concepts and the cross-regional application of green technological achievements. This result verifies H3. In AI-leading regions, green technological innovations benefit from AI-driven advantages such as dematerialization and the non-rivalrous nature of data. These innovations can be deployed to SMEs that previously faced prohibitive spatial barriers and cost constraints, thereby revitalizing local green technology markets. By lowering adoption thresholds and fostering technology diffusion, AI empowers SMEs to achieve their green transition goals, ultimately catalyzing the formation of integrated green smart city clusters characterized by cross-regional sustainability synergies. For example, the promotion of intelligent equipment is realized through urban construction intelligence and energy monitoring systems, i.e., optimizing the allocation of energy through smart grids and smart logistics, balancing cross-regional energy consumption, and realizing energy conservation at the regional level. In addition, the reason that AI has a more pronounced effect on the reduction of energy consumption in the neighboring regions may be that AI itself has a higher energy consumption and AI research and development is carried out in relatively developed regions in terms of science and technology, and although the energy-saving effect brought by AI to the region offsets the energy consumption, this effect is stronger in the neighboring regions.

To further analyze the role of AI deployment in smart cities in enhancing green resilience, this study incorporates co-function between AI and CET policy. The results, presented in Columns (2) and (4), demonstrate high consistency between the HDFE model and the SDM, confirming the robustness of the findings. In Column (2), the coefficient of the AI*CET co-function is significant, demonstrating that AI-CET synergy reduces local energy consumption by 7.39%—substantially higher than the 2.28% reduction observed in Column (1) (with CET policy alone). Similarly, Column (4) reveals that the AI*CET interaction term drives an 8.55% decline in local energy consumption, outperforming the 2.10% reduction in Column (3) (CET policy only), verifying H5. This suggests that the smart city construction relying on AI is capable of the self-organization effect, with its data-driven, technology integration, and dynamic adaptation characteristics, it empowers the construction of a green, low-carbon, resilient, and adaptive urban system, and prospectively realizes the synergistic enhancement of the emission reduction efficiency and system resilience to create a “low-energy consumption, high resilience” urban ecology.

Effect decomposition enables a clear quantification of the impacts of policies or variables both locally and across regions, thereby preventing misjudgment of their actual influence. Therefore, this paper further decomposes the effects of the variables obtained from the above table, and obtains the direct effects of local factors on the local energy consumption and the indirect effects of local factors on the neighboring energy consumption, and the total effect is the sum of the direct and indirect effects (

Table 5. Decomposition of SDM spatial effects.

Variable	Direct Effect	Indirect Effect	Total Effect
CET	−0.0212 ***	0.0092	−0.0119 *
	(−8.5384)	(1.5191)	(−1.8840)
AI	0.0206	−0.1950 ***	−0.1744 ***
	(0.8639)	(−3.4324)	(−3.0510)
lntrans	−0.0024 ***	−0.0024	−0.0048 **
	(−2.5863)	(−1.0229)	(−1.9925)
lnGDP	0.0881 ***	0.0387 ***	0.1268 ***
	(23.7500)	(4.6254)	(14.1242)
lnfdi	−0.0001	0.0015	0.0015
	(−0.1047)	(1.1662)	(1.0753)
lnurb	0.0015	0.0045 *	0.0060 **
	(1.2945)	(1.9464)	(2.4783)
lnconsume	0.0026 **	0.0008	0.0033
	(2.5482)	(0.3802)	(1.6128)
lnsciout	−0.0037 ***	0.0052 **	0.0016
	(−3.7089)	(2.4298)	(0.6402)
secper	0.3074 ***	0.0670 **	0.3744 ***
	(25.3908)	(2.3306)	(12.5110)

Note: ***, **, and * indicate statistical significance at 1%, 5%, and 10% levels, respectively; the t-statistics are reported in parentheses.

). For direct effect, the coefficient of the CET policy is −0.0212, which indicates that the CET can reduce the energy consumption in the region. Regarding indirect effects, AI contributing to a 19.50% reduction in energy consumption in neighboring regions. For total effects, the CET policy leads to a 1.19% decrease in energy consumption, while AI drives a 17.44% decline.

The percentage of final energy consumption reflects the degree of regional energy transition regional renewable energy. In order to further analyze the influence on regional energy transformation and CET energy policy composite smart city construction, this paper uses the SDM to further benchmark the percentage of regional renewable energy consumption regression (

Table 6. Spatial regression results of renewable energy consumption.

	(1)	(2)	(3)	(4)
Variable	Model 5	Model 6	Model 7	Model 8
CET	0.0073 **	0.0201 ***	0.0063 ***	0.0166 ***
	(2.0801)	(3.5566)	(5.4070)	(5.4896)
AI	0.0062	0.0005	0.0133	0.0083
	(0.3344)	(0.0272)	(1.1088)	(0.6909)
CET×AI		−0.0253 *		−0.0198 ***
		(−1.6587)		(−3.6041)
lntrans	0.0004	0.0003	0.0002	0.0001
	(0.4089)	(0.2741)	(0.5096)	(0.3103)
lnGDP	0.0120 ***	0.0134 ***	0.0131 ***	0.0141 ***
	(4.1051)	(4.8894)	(7.1343)	(7.5509)
lnfdi	0.0010 *	0.0009 *	0.0007 ***	0.0007 ***
	(1.8963)	(1.8274)	(2.9265)	(2.7519)
lnurb	−0.0009	−0.0008	−0.0009 *	−0.0008
	(−0.9470)	(−0.8465)	(−1.6678)	(−1.4635)
lnconsume	−0.0020 **	−0.0021 **	−0.0020 ***	−0.0021 ***
	(−2.2309)	(−2.2739)	(−4.2631)	(−4.4920)
lnsciout	0.0011	0.0011	0.0012 **	0.0013 ***
	(1.2555)	(1.3584)	(2.4565)	(2.6096)
secper	−0.0658 ***	−0.0661 ***	−0.0637 ***	−0.0637 ***
	(−5.7538)	(−5.7591)	(−10.5567)	(−10.5377)
W×CET			−0.0101 ***	0.0000
			(−2.9316)	(0.0039)
W×AI			0.2267 ***	0.2134 ***
			(6.9422)	(6.4850)
W×CET×AI				−0.0141
				(−1.0177)
W×lntrans			−0.0008	−0.0010
			(−0.6438)	(−0.7914)
W×lnGDP			0.0164 ***	0.0154 ***
			(3.5028)	(3.2464)
W× lnfdi			−0.0029 ***	−0.0028 ***
			(−4.0633)	(−3.9822)
W×lnurb			−0.0018	−0.0014
			(−1.3466)	(−1.0857)
W×lnconsume			−0.0032 ***	−0.0034 ***
			(−2.9955)	(−3.1214)
lnsciout			0.0006	0.0011
			(0.5052)	(0.8271)
secper			−0.0108	−0.0058
			(−0.6922)	(−0.3706)
cons	−0.0988 **	−0.1168 ***
	(−2.1051)	(−2.7248)
Rho			−0.2449 ***	−0.2508 ***
			(−6.2292)	(−6.3772)
	0.9639	0.9642	0.6748	0.6783
			0.0001 ***	0.0001 ***
			(34.2099)	(34.2034)
N	2358	2358	2358	2358

Note: ***, **, and * indicate statistical significance at 1%, 5%, and 10% levels, respectively; the t-statistics are reported in parentheses.

). Column (1) is the ordinary HDFE model, Column (2) presents the HDFE model incorporating the composite effect of CET policy and AI, Column (3) is the SDM model, and Column (4) is the SDM with the addition of the compound effect of CET policy and AI. The coefficients of CET in Columns (1) and (3) are positive, suggesting that the CET policy promotes firms to accelerate the green technology research under the guidance of profit by market-oriented means, thus promoting the regional energy structure transform to renewable energy. H2 is thus supported.

However, in Column (3), the CET policy significantly reduces renewable energy consumption in neighboring cities. This phenomenon arises because CET pilot cities may generate a “siphoning effect”, absorbing renewable resources from adjacent regions. Consequently, the proportion of renewable energy usage becomes relatively higher in pilot areas while declining in neighboring regions. Therefore, the CET pilot cities should pay attention to the coordination with the neighboring areas while developing themselves. In addition, the use of AI can promote renewable energy consumption in surrounding areas, which indicates that AI will spatially compensate for the decrease of renewable energy in surrounding areas due to the siphoning effect, and due to the ecological diffusion of green technology, the surrounding areas can also benefit from the diffusion effect of advanced green technology, so as to realize the energy structure transformation. H4 is therefore verified. Finally, from Column (4), the CET*AI term will reduce renewable energy consumption locally; in order to further explore its internal reasons, this paper further conducts quantile regression.

Furthermore, this study decomposes the spatial effects, with detailed results reported in

Table 7. Decomposition of spatial effects.

Variable	Direct Effect	Indirect Effect	Total Effect
CET	0.0067 ***	−0.0096 ***	−0.0028
	(5.5367)	(−3.3721)	(−0.9740)
AI	0.0058	0.1858 ***	0.1916 ***
	(0.4943)	(6.9508)	(7.2499)
lntrans	0.0003	−0.0006	−0.0003
	(0.6882)	(−0.5968)	(−0.3008)
lnGDP	0.0126 ***	0.0108 ***	0.0234 ***
	(6.9690)	(2.8329)	(5.7930)
lnfdi	0.0008 ***	−0.0025 ***	−0.0017 ***
	(3.4077)	(−4.0980)	(−2.6353)
lnurb	−0.0008	−0.0014	−0.0022 **
	(−1.5070)	(−1.2537)	(−1.9737)
lnconsume	−0.0019 ***	−0.0023 **	−0.0042 ***
	(−3.8709)	(−2.4277)	(−4.4166)
lnsciout	0.0012 **	0.0004	0.0016
	(2.4179)	(0.3968)	(1.3729)
secper	−0.0634 ***	0.0033	−0.0601 ***
	(−10.6705)	(0.2475)	(−4.4387)

Note: *** and ** indicate statistical significance at 1%, 5%, and 10% levels, respectively; the t-statistics are reported in parentheses.

. For direct effect, CET policy has a positive effect and passes significant tests, indicating that CET implementation promotes renewable energy consumption in local regions. From the indirect effect, the CET policy exhibits a siphon effect, declining the renewable energy adoption in neighboring areas. In contrast, AI is associated with a relative increase in renewable energy consumption shares in adjacent regions, suggesting that AI development helps counteract the CET-induced spatial disparities. From the total effect, AI advancement demonstrates a net positive impact, driving an overall rise in renewable energy consumption.

In order to further explore the internal reasons for the CET*AI term to reduce the renewable energy consumption, this paper carries out quantile regression under a total of 4 quartiles of 0.25, 0.5, 0.75, and 0.9 (

Table 8. Quantile regression results.

Variable	25%	50%	75%	90%
$\mathrm{C}\mathrm{E}\mathrm{T}$$\times AI$	−0.0753 ***	−0.0451 ***	0.0093	0.0959 **
	(−6.3238)	(−2.5843)	(0.7409)	(2.5227)
$\mathrm{C}\mathrm{E}\mathrm{T}$	0.0828 ***	0.0810 ***	0.0481 ***	0.0010
	(10.1127)	(8.7841)	(6.0270)	(0.0558)
AI	0.0941 ***	0.1624 ***	0.2291 ***	0.1537 ***
	(10.1048)	(9.6107)	(10.7776)	(6.0485)
cons	0.1750 ***	0.2442 ***	0.2938 ***	0.2067 ***
	(8.0869)	(6.8395)	(6.0550)	(2.6754)
Controls	Yes	Yes	Yes	Yes
N	2358	2358	2358	2358

Note: *** and ** indicate statistical significance at 1%, 5%, and 10% levels, respectively; the t-statistics are reported in parentheses.

). The influence of each influencing factor on the energy consumption structure under different quantile points is different, that is, the influence of each influencing factor on the energy consumption structure has quantile heterogeneity. Meanwhile, at the 25% and 50% quantile points, the CET*AI term reduces renewable energy consumption, while for the 90% quantile, the CET*AI term significantly increases renewable energy consumption, partially verifying H6, which indicates that for traditional energy sources, the development of AI as well as the green policy both have a consumption-reducing effect. However, for renewable energy development, only those cities with the best green foundations can afford to exploit the advantages of intelligence empowerment.

4.4. Discussion

As demonstrated by our aforementioned empirical evidence, smart cities amplify the impact of carbon policies on energy performance and exhibit green spatial effects. In this context, an AI-empowered CET system actively provides institutional support for shaping the transformation paradigm and generating positive externalities [58], equipping smart cities with green resilience in their economic progression toward energy sustainability. In this section, we further discuss “green resilience” and elaborate on how this specific form of resilience is hierarchically constructed within China’s smart cities. Our framework contributes to understanding the co-evolution of technology and society, offering valuable insights for smart city planners in emerging markets.

At the city level, green resilience reflects the incentive compatibility [59] between economic competitiveness and energy sustainability. Building on our empirical evidence that the interplay between CET and AI enhances energy sustainability, we propose that “co-incentive effects” contribute to the greening of smart cities. On one hand, AI accelerates CET implementation through market efficiency channels. On the other hand, carbon policies provide an agile framework for AI governance. Although AI training is energy-intensive, high energy consumption does not necessarily result in high carbon emissions—the key factor is the efficiency of renewable energy utilization. CET enables AI to better support urban green development by shaping the energy mix, and thus environmental outcomes. Through institutional innovation, China’s CET integrates economic incentives and smart technologies into urban green governance, providing a “carbon reduction buffer” for enterprises facing high transformation barriers and trial-and-error costs, while generating positive innovation compensation [60] for those “green pioneers”. This internalization mechanism facilitates the unlocking of green capital and corrects resource misallocation, providing more inclusiveness for city transformation by alleviating time compression constraints [61].

At the regional level, green resilience reflects the synergy between regional advantages and resource endowments. Our previous findings indicate that while AI exhibits green spatial effects, it lacks localized advantages in enhancing energy performance, as its heavy energy consumption offsets gains in production efficiency. Expanding on this, we propose two key mechanisms: First, carbon policies create an AI-ubiquitous urban ecosystem that fosters developmental resilience. This suggests that AI’s green potential can be internalized within institutional innovation through technological embeddedness. Second, spatial co-functions primarily involve cross-regional resource redeployment. The goal of AI enablement is not to generate immediate local benefits, but to contribute to a “global optimum” of energy efficiency across a broader geographic scale. Vertical linkages among institutional innovation, urban synergy, and technological integration enhance high-dimensional locational advantages. Overall, the spatial effects of AI act as a “bond of resilience” in aligning local needs with regional resource efficiency, synergizing cities to co-intellectualize and “breathe” together.

At the national level, green resilience represents a new social paradigm driven by AI and eco-friendliness. While our findings confirm the pro-sustainability effects of CET and AI, the purely green externalities of the policy are not captured. This suggests that, under a smart economy framework, cities are no longer solely dependent on policy regulations to establish exogenous cooperation models. Instead, they are spontaneously forming endogenous self-organizing mechanisms, fostering a “symbiotic” model of collaboration. Cities appear to have a clear understanding of which cooperation frameworks best enhance their green resilience in response to sustainability challenges. Accordingly, we argue that the AI-driven urban green economy, centered on energy conservation and environmental sustainability, now operates within an expanded functional boundary. This reflects the emergence of a “super-efficient” economic mechanism in smart city development. Firstly, technology iteration drives regional economic restructuring, allowing AI to amplify the multiplier effect of data [62] and establish a green economic network, where smart cities serve as key nodes, and trans-regional cooperation pathways as edges. Secondly, the synergy between policy and technology facilitates innovation clustering, transforming spillovers into new resource advantages through institutional design. Cities naturally develop closer linkages through the release and uptake of these resources. Through this process, the transformation from AI-driven operations to smart city development, from green spatial externalities to economic regionalization, and from institutional innovation to sustainable ecosystem support, can be effectively realized. The analysis framework and mechanism of smart cities with green resilience are presented in Figure 2.

5. Conclusions

This paper investigates the influence of smart city construction on green resilience based on an AI perspective. By using spatial measurement methods, this paper finds that the smart city can reduce the energy consumption by 8.55% locally, which is higher than the 2.10% of the execution of the CET policy alone. The CET policy reduces the regional energy consumption, promotes renewable energy consumption, and does not cause the relocation of high energy-consuming enterprises, and does not make the neighboring region’s energy consumption rise, but due to the “siphon effect”, it reduces renewable energy consumption in neighboring regions. AI technology has a further eco-competitive advantage for reducing energy consumption, which makes the energy consumption in the neighboring areas significantly reduced by 21.84%, and the percentage of renewable energy consumption significantly increased. In addition, for renewable energy development, only those cities with the best green foundation can afford to enjoy the empowerment of intelligence. Based on our findings, we propose to strategically focus on promoting the construction of a national CET market, strengthening governmental regulation, and rationalizing the use of AI to promote the process of energy conservation.

First of all, the object of this paper is the CET city, because its policy objective is to “provide experience for the wider carbon emissions trading through the experimental practice in some cities”, although it achieves marked reductions in localized energy demand, and does not increase the energy consumption level of the neighboring regions, but also does not reduce it in the neighboring regions. Therefore, it is crucial to build a national CET market. At present, the construction of the national carbon market is still in the initial stage, and the institutional system needs to be further improved [26]. Under the dual-carbon goal, we should continue to improve the relevant policy support system of the national CET market, enhance policy publicity, strengthen international communication and cooperation, and continuously enrich the trading varieties, subjects, and methods, so as to further enhance the market activity. Furthermore, we found that the CET policy can indeed reduce the region’s energy consumption and promote renewable consumption, but there is a reverse effect on the neighboring regions in terms of the proportion of renewable resources. Therefore, there is a need to have timely introduction of supporting policies to achieve regional-type coordination. For example, directing fiscal expenditures toward renewable energy AI infrastructure in underdeveloped regions, coupled with instituting regional renewable energy consumption quotas that mandate central cities to provide proportional financial or technical restitution when utilizing green electricity from adjacent areas, can effectively mitigate unidirectional resource appropriation. Finally, a national green market with high-quality carbon credits should be actively developed, and a scientific, reasonable, open, and transparent system with broad participation constructed.

Second, the development of AI technology is used to promote green innovation and business cooperation. This paper confirms that AI can reduce energy consumption, and helps to increase renewable energy consumption in neighboring cities, enhancing the effectiveness of the CET policy. Therefore, it should accelerate the formulation of a comprehensive AI development plan and encourage enterprises to participate in research. Let AI promote traditional industries upgrading and the transformation of small business processes. For the traditional high-carbon footprint industry, although it has a certain accumulation under the traditional development mode, in the process of accumulation, it is more necessary to realize the effective transformation of the production mode and the improvement of service effectiveness through AI. This can help traditional enterprises to find new opportunities, so that they can open up new markets on the basis of routine, and lead new business models through new technologies, thus changing the method of value creation so that there is a deviation from the traditional models and practices and embracing new, more environmentally friendly measures and policies. For small enterprises, AI provides an opportunity for small enterprises to change their organizational methods, and it is even more necessary to put AI technology into the transformation of production processes and the scientific and reasonable arrangement of production capacity and production plans. This enables enterprises to use the positive externalities of AI for optimizing resource distribution when entering the industry, and at the same time, accurately judging the direction of the future development of their own products and technologies through AI technology, thus effectively saving resources, focusing on the most efficient areas, to create the enterprise’s flagship products and advantageous growth points. In advancing AI development, policymakers should prioritize strategic rebalancing of AI infrastructure, implement tax credits for enterprises conducting cross-regional AI skills training, and establish mechanisms allowing provinces with computational surplus to offset excess AI energy consumption by funding digital literacy initiatives in underdeveloped regions. These measures collectively foster intelligent cross-regional applications in energy consumption forecasting, optimized grid scheduling, and green technology innovation. They strengthen cross-field cooperation based on the ecosystem advantages of AI, and promote energy efficiency improvement of the whole value chain with AI.

Finally, human supervision and intervention are crucial to ensure the effectiveness and value of AI applications. To ensure the sustainable, long-term development of AI within the energy sector, government policy guidance and market regulation should be strengthened, international legislative cooperation on AI should be promoted, and irregular and non-long-term rational behaviors of components of the AI-based ecosystem should be reasonably prevented and controlled. We can improve enterprises’ understanding of the relationship between AI and the environment and combine market mechanisms and legal regulatory tools to incentivize the greening and upgrading of AI technology and reduce energy consumption. For instance, the green smart zero-carbon port constructed by China’s Dongfang Electronics Group has significantly improved port clearance efficiency. The company’s development of the country’s first urban-level virtual power plant has notably reduced urban carbon emissions, fostering the advancement of smart cities and green low-carbon development. Furthermore, Shanghai’s adoption of AI agents for waste classification, utilizing automatic recognition and optimized disposal solutions, has increased garbage recycling rates by 30%. These cases demonstrate the immense potential of strategically leveraging digital technologies to build green resilient smart cities. Overall, as the artificial intelligence industry remains an emerging field, there is currently a relative scarcity of precise city-level AI data. While the AI measurement indicators selected in this study capture the core factors reflecting regional AI development levels, they still exhibit certain limitations in granularity. A promising avenue for future research lies in refining the methodology for measuring city-level AI development and designing more comprehensive AI evaluation metrics. Concurrently, with the rapid advancement of AI, ecosystem competitive advantages are poised to become a crucial dimension in analyzing smart city development and urban green resilience. However, there remains a notable absence of standardized metrics for assessing these ecological competitive advantages, making it challenging to effectively characterize them through empirical research. Consequently, investigating the interrelationships among components within AI ecosystems and developing multidimensional, synergistic AI ecosystem frameworks with corresponding measurement indicators will emerge as a novel and potentially impactful research direction. By addressing these issues, we help to achieve energy sustainability.