Abstract

Translate

Against the dual backdrop of iterative AI advancement and deepening green development imperatives, AI-driven industrial intelligence (INT) has emerged as a pivotal force in driving sustainable economic growth. While the existing literature has explored the correlation between INT and green total factor productivity (GTFP), significant gaps remain in the design of multidimensional variables, analysis of environmental regulation (ER), and capture of dynamic effects. From the perspective of ER, this study utilizes provincial panel data from China (2012–2023) to construct an 11-indicator evaluation system for INT development and employs the EBM super-efficiency model to measure GTFP. Furthermore, a two-way fixed effects model combined with a moderated mediation model is established to systematically elucidate the intrinsic linkage mechanism between INT and GTFP. The key findings are as follows: First, INT has a significant positive impact on GTFP. Second, green innovation and spatio-economic synergy are crucial pathways through which INT empowers GTFP. Third, ER exhibits a substitution effect within both the direct and indirect impacts of INT on GTFP, where intensified ER significantly attenuates INT’s positive impacts. Fourth, the enhancement effect of INT on GTFP remains statistically significant with a one-year lag, and the substitution effect of ER persists. This study provides an in-depth analysis of the mechanisms of INT-driven green economic transformation, offering valuable insights for governments to implement differentiated environmental governance strategies tailored to local conditions.

Full text

Turn on search term navigation

Translate

1. Introduction

In the context of the ongoing development of global Sustainable Development Goals, promoting the transition of development models towards green and low-carbon pathways has emerged as a central agenda for the international community [1]. Statistics reveal that, after excluding the impact of the pandemic, global carbon emission intensity continued to increase during 2010–2024, falling significantly short of the 7.6% annual carbon reduction required to achieve temperature control targets [2]. As the world’s largest carbon emitter, China exhibits energy consumption per unit of GDP that is 1.5 times the global average [3], thereby necessitating the urgent optimization of environmental efficiency. Within this context, the Opinions of the Central Committee of the Communist Party of China and the State Council on Accelerating the Comprehensive Green Transformation of Economic and Social Development explicitly advocate “coordinated advancement of carbon reduction, pollution abatement, green expansion, and growth”, signifying a systemic transformation in China’s development paradigm [4]: the extensive growth model reliant on factor inputs has become unsustainable, rendering efficiency-centered high-quality development an inevitable imperative. Consequently, Green Total Factor Productivity (GTFP) has become a pivotal indicator for evaluating environmental performance and gauging economic sustainability. Thus, the effective enhancement of GTFP has become a critical issue for advancing green transformation in China and globally.

Concurrently, the ongoing technological revolution spearheaded by AI is reshaping the trajectory of global economic growth [5]. As a novel General-Purpose Technology (GPT), AI leverages machine learning algorithms and intelligent sensors to significantly enhance production efficiency [6], foster rationalized industrial structures, and facilitate low-carbon energy allocation [7]. It is widely recognized as a key enabler of sustainable development [8]. Nations have successively launched AI development strategies to enhance their competitiveness, emphasizing the instrumental role of AI-driven Industrial Intelligence in enabling green transitions [9]. Notable examples include the EU’s White Paper on Artificial Intelligence and the US’ National Artificial Intelligence Research and Development Strategic Plan. Similarly, China’s Next Generation Artificial Intelligence Development Plan prioritizes cultivating INT-driven economic growth engines and seeks more effective momentum to improve green development efficiency. Examining the mechanisms through which INT enhances GTFP thus holds significant importance for achieving high-quality economic development.

The existing literature has explored INT and GTFP from diverse perspectives, which can be broadly categorized into three thematic strands. First, research on the influencing factors of GTFP. Unlike conventional total factor productivity, GTFP incorporates undesirable outputs into its framework [10]. Consequently, investigations into its influencing factors are primarily divided into exogenous factors constraining undesirable outputs and endogenous factors promoting desirable outputs. Environmental regulation (ER) is a primary mechanism for restricting such outputs. Predominant discourse centers on the Porter Hypothesis, proposing that moderate ER enhances corporate green innovation incentives through the “innovation compensation effect,” thereby elevating GTFP [11]. However, studies also indicate that excessively stringent ER may trigger crowding-out effects due to increased corporate “compliance costs,” resulting in an inverted U-shaped relationship between ER and GTFP [12]. Regarding factors promoting desired outputs, scholars have conducted research from perspectives including finance, technological innovation, industrial structure, and factor synergy. In the financial domain, Li et al. found that green finance enhances enterprise GTFP by promoting energy conservation and emission reduction, with this effect being more pronounced among non-state-owned enterprises, large-scale enterprises, and those facing weaker financing constraints [13]. Lee & Lee demonstrated the promoting effect of green finance on GTFP by constructing a green finance index, noting that specialized green finance policies can further enhance desired outputs [14]. Furthermore, De Mariz analyzed the role of financial technology in sustainable development from a global perspective [15]. Concerning technological innovation, Zhao demonstrated that the positive impact of low-carbon technological innovation on GTFP strengthens as GTFP increases [16]. Utilizing a Spatial Durbin Model, Wang et al. analyzed the spatial spillover effects of GTFP, revealing that green technological innovation exhibits a positive influence on local GTFP but a negative effect on neighboring regions [17]. Du & Li further investigated the primary barriers to green innovation across different economies and their impact on GTFP [18]. In terms of industrial structure, focusing on the Yangtze River Economic Belt, You et al. observed that industrial structure upgrading not only positively influences industrial GTFP directly but also synergizes with carbon constraints [19]. Expanding the scope to cities across China, Tian and Zhang identified that industrial structure optimization promotes GTFP growth, noting distinct effects from its advancement and rationalization components [20]. Regarding factor synergy, based on data from Chinese prefecture-level cities, He et al. discovered that manufacturing synergy boosts GTFP through technological progress, while the co-synergy of manufacturing and producer services exhibits a significant positive effect [21]. Other studies employing methodologies such as Tobit and quantile regressions have confirmed that the digital economy can boost GTFP, with industrial structure upgrading and factor synergy serving as critical mediating pathways for output enhancement, thereby offering insights into identifying green transformation pathways [22,23,24]. While existing studies have explored GTFP from multiple dimensions, few scholars have analyzed these core driving factors from the perspective of ER. Moreover, the role of ER remains inconclusive, presenting a critical research gap for advancing GTFP.

Second, research on INT characteristics which encompasses multiple dimensions. Firstly, Innovation Incentive. INT reduces information asymmetry, thereby lowering R&D costs and fostering innovation [25]. Secondly, Economic Restructuring. INT optimizes industrial structures by displacing low-skilled labor [26] while enhancing resource allocation efficiency through sharing mechanisms [27]; Thirdly, Environmental Friendliness. Recognized as a pivotal driver of green development, INT augments environmental benefits through productivity improvements and precision pollution control [28]. However, these optimistic narratives may overlook the dual-edge nature of INT, such as its potential to exacerbate energy consumption [29] and socio-economic inequalities [6], which requires further research.

Third, direct nexus studies between INT and GTFP establish a consensus that INT significantly promotes GTFP. At the macro level, research utilizing robot deployment data validates GTFP enhancement at provincial levels with noted heterogeneity [30]. At the micro level, studies predominantly quantify INT using AI patent metrics, revealing its positive impact on corporate GTFP [31,32]. Notably, while these studies provide a certain theoretical foundation, the consensus may neglect methodological constraints, such as the reliance on patent-based proxy indicators potentially failing to adequately capture INT’s environmental performance [33], as well as the policy-making biases that may arise due to mechanisms not yet being precisely identified. These factors collectively challenge the sustainability of INT.

The review of the existing literature reveals that critical knowledge gaps still persist, including the overreliance on unidimensional INT proxies that inadequately capture its technological-industrial integration, the moderating mechanism wherein ER’s role (as enabler or inhibitor) in INT-driven GTFP growth remains a black box, and the underexplored time-lag effect inherent in INT’s diffusion pathways that necessitate rigorously designed longitudinal validation.

Accordingly, this study constructs an integrated framework incorporating INT, ER, and GTFP to systematically address the following four core questions: (1) Can INT unequivocally promote GTFP growth? (2) Through which pathways does INT empower GTFP enhancement? (3) What role does the ER play in this nexus, complementing or substituting INT’s green effects of INT? (4) Does INT’s impact on GTFP exhibit discernible time-lag effects? Methodologically, leveraging provincial panel data from China (2012–2023), we develop an 11-indicator INT evaluation system via the entropy method and measure GTFP using the super-efficiency EBM model to precisely quantify both variables. Building on this, we implement a two-way fixed effects model to examine INT’s direct influence on GTFP, rigorously validating results through five robustness tests and two instrumental variables. Furthermore, a moderated mediation model is constructed to investigate the mediating roles of green innovation and spatio-economic synergy, alongside ER’s moderating effect. Finally, temporal characteristics are analyzed by introducing INT lag terms, and substitution effects of ER across lag periods are tested through grouped regression.

The study’s marginal academic contributions manifest in three dimensions: First, methodological refinement. The 11-indicator INT evaluation system and super-efficiency-EBM-based GTFP measurement enhance variable design precision. Second, it offers theoretical advancement by uncovering mediation pathways of green innovation (accounting for 26.612%) and spatio-economic synergy (accounting for 35.889%) and revealing ER’s substitution effects within both direct and indirect INT-GTFP transmission mechanisms. Third, temporal perspective extension. Lag term (1-year delay) analysis captures INT’s delayed impact (0.250, p < 0.01) while delineating ER’s evolving influence across time horizons. Practically, this study provides policymakers with spatially differentiated governance tools and evidence-based insights for designing temporally adaptive environmental policies.

The remainder of this paper is organized as follows: Section 2 articulates the theoretical framework and hypotheses; Section 3 details the research design and methodologies; Section 4 presents the baseline regressions, robustness tests, and endogeneity treatments; Section 5 conducts mechanism validation; Section 6 further expands the analysis; and Section 7 concludes with discussions and policy implications.

2. Literature Review and Hypotheses Development

2.1. Direct Impact of INT on GTFP

The enhancement of GTFP signifies a win-win situation for economic efficiency and environmental benefits, necessitating the maximization of production efficiency while minimizing negative externalities, such as pollution. As an engine for green economic development [34], the INT’s inherent characteristics of technological generality and penetrability provide a foundational impetus for GTFP advancement [35].

Based on the existing literature [23,34] and GTFP measurement indicators (i.e., factor inputs, desired outputs, and undesired outputs), the direct impact of INT on GTFP primarily manifests in three dimensions. First, in enterprise operations and supply chain management, INT fosters circularity by leveraging big data analytics to optimize logistics, minimize carbon footprints in transportation, and enhance waste recycling efficiency via intelligent algorithms, ultimately constructing circular economic models. This systemic efficiency transformation and environmental enhancement across the entire value chain constitute the core mechanism of INT’s positive impetus on GTFP. Second, production efficiency optimization encompasses input and output dimensions. Regarding inputs, INT enables producers to dynamically analyze and optimize production parameters, including energy consumption, factor allocation, and emissions, thereby improving resource configuration precision and energy utilization efficiency. Regarding outputs, IoT and embodied intelligence establish intelligent production systems that replace low-level labor and methods, thereby reducing environmental pollution. Moreover, INT-enabled operations in high-risk environments can enhance clean production capacity. Third, governance of pollution, encompassing analysis and prevention. AI-driven industrial intelligence facilitates a paradigm shift from end-of-pipe treatment to source prevention and process control. AI-powered geographic information systems can rapidly identify pollution sources and diffusion pathways, thereby providing targeted decision support for government interventions. For prevention, AI-driven predictive models forecast contamination trends and prescribe preventive measures, reducing environmental incidents and governance costs [36], thus constraining undesirable outputs. Consequently, we propose the following first hypothesis:

Hypothesis 1.

The development of INT shows a significantly positive correlation with GTFP.

2.2. Indirect Pathways of INT’s Impact on GTFP

Beyond direct effects, INT’s influence on GTFP operates through indirect pathways, catalyzed by specific economic effects. This study focuses on two pivotal mediating mechanisms: the green innovation effect and the spatio-economic synergy effect.

First, INT enhances GTFP by improving the efficiency of green innovation. The implementation of INT markedly diminishes the barriers and costs associated with innovation, particularly in the domain of green technology [37]. On the one hand, INT accelerates the creation and propagation of green knowledge. By utilizing language models and machine learning algorithms, researchers can rapidly integrate green technology information from patents and literature, thereby shortening the R&D cycle. On the other hand, through simulation and computational modeling, INT provides low-cost, high-efficiency experimental environments and reliable market analysis for green product development, reducing resource consumption and trial-and-error costs [38]. Consequently, firms possess greater capability and motivation to engage in green innovation, yielding high-value patents and products that expand green technological frontiers. Furthermore, once these innovations are applied in production, they promote improvements in energy efficiency and reduce pollution emissions, fundamentally fostering the endogenous growth of GTFP.

Second, INT enhances GTFP by improving the level of spatio-economic synergy. According to economic agglomeration theory, the spatial concentration of industries and factors enhances regional productivity via shared infrastructure, labor market matching, and knowledge spillovers [39]. INT’s development is expanding the scope and forms of spatio-economic synergy. By amplifying the network effects of the digital economy, INT facilitates the cross-regional flow and efficient allocation of innovative factors, while concurrently attracting high-caliber talent, venture capital, and advanced productive forces, thereby fostering the formation of innovation ecosystems [40]. Such ecosystems increase per-unit output and accelerate the adoption of clean technologies, deeply integrating environmental protection technologies with various production and consumption within regions and catalyzing green industrial clusters that exhibit synergistic effects, collectively enhancing GTFP. Therefore, we propose the second hypothesis as follows:

Hypothesis 2.

INT indirectly enhances GTFP by elevating green innovation efficiency and strengthening spatio-economic synergies.

2.3. Potential Moderating Role of Environmental Regulation

As a multifaceted external variable, ER may exhibit complex interactions with INT as a technological driver, rather than a simple linear superposition. Conventional wisdom posits that ER complements technological progress through “forced upgrading mechanisms”, compelling firms to develop eco-technology, thereby jointly advancing GTFP, consistent with the “Strong Porter Hypothesis” [12]. Under this logic, an enhanced ER would amplify INT’s positive impact on GTFP. However, evolving theoretical insights reveal that ER’s relationship with GTFP and technological innovation is not uniformly reinforcing. Instead, it follows an inverted U-shaped curve, initially rising before declining [41].

This paper thus proposes a competing hypothesis from the substitution effect perspective: While ER enhances GTFP, it may negatively moderate INT-GTFP linkages under specific conditions. The underlying logic is that INT and ER represent partially substitutable pathways for enhancing green performance [42]. Under moderate ER regimes, firms encounter limited coercive pressure, allowing INT’s efficiency-enhancing properties to maximize its “green dividend”. In other words, INT functions as a market-driven incentive for the green transition in weak regulation contexts. Conversely, stringent ER policies impose substantial compliance costs, compelling firms to allocate significant resources toward pollution abatement to meet regulatory mandates. Under such “strong-constraint” scenarios, compliance supersedes market-driven efficiency maximization as the primary corporate objective. Here, firms have predominantly attained green performance enhancements through non-INT methods. According to the law of diminishing marginal returns, INT’s potential to improve green outcomes is thus limited. Strict ER essentially “crowds out” INT’s green-enabling potential, revealing a substitution relationship between these two forces. This substitution effect may similarly manifest within INT’s indirect pathways, like green innovation and spatio-economic synergy. For instance, under a stringent ER, corporate innovation tends to prioritize compliance-driven passive innovation over efficiency-seeking proactive innovation [43]. Therefore, the following hypothesis is proposed:

Hypothesis 3.

ER exhibits a negative moderating effect on both direct and indirect impacts of INT on GTFP, namely a substitution effect.

2.4. Time-Lagged Effect of INT on GTFP

The full realization of the socioeconomic effects of any technology takes time, and INT is no exception. Its positive impact on GTFP is likely subject to a lag effect, primarily attributable to the inherent cycles of technology adoption, organizational restructuring, and industrial evolution.

First, the adoption and deployment of INT require time. For enterprises, introducing INT systems entails not only substantial capital investment but also necessitates the restructuring of existing production processes and management models, which is a protracted process itself. Moreover, a time lag exists in aligning human capital with the organizational structure. Effective implementation of INT requires a significant workforce proficient in both technology and business operations. However, requisite skill development for labor and adaptation of corporate organizational frameworks and cultures cannot be achieved overnight. During the initial implementation phase, adjustment costs may even temporarily reduce productivity, manifesting as a short-term “productivity paradox” [44]. Finally, the broader green impact of INT relies on knowledge spillovers and the development of an industrial ecosystem. The diffusion of environmental technologies from pioneering firms to industry-wide and economy-wide adoption, which generates scale effects, necessitates considerable time. Consequently, as technology, talent, and organizational structures progressively adapt over time, INT’s positive effect on GTFP is likely to become stronger. Accordingly, we propose the following last hypothesis:

Hypothesis 4.

INT’s positive impact on GTFP has a time-lagged effect.

The variable relationships in this study are shown in Figure 1.

3. Methodology and Data

3.1. Model Settings

3.1.1. Benchmark Model

To examine the facilitating effect of INT on GTFP, this study constructs the following two-way fixed effects model incorporating province and year fixed effects based on the Hausman test results.

(1) ${G T F P}_{i t} = β_{0} + β_{1} {I N T}_{i t} + β_{2} {F i n}_{i t} + β_{3} {F D I}_{i t} + β_{4} {G R P}_{i t} + β_{5} {I n d u s}_{i t} + β_{6} {G o v}_{i t} + α_{i} + γ_{t} + ϵ_{i t}$

in Equation (1), i denotes province (municipality or autonomous region), and t denotes year. The dependent variable

{G T F P}_{i t}

represents the green total factor productivity of province i in year t. The core explanatory variable

{I N T}_{i t}

measures the INT development level of province i in year t. In accordance with seminal literature on the green economy, the following control variables are employed: financial support level (

{F i n}_{i t}

), foreign direct investment (

{F D I}_{i t}

), economic development level (

{G R P}_{i t}

), industrialization level (

{I n d u s}_{i t}

), and government intervention intensity (

{G o v}_{i t}

);

α_{i}

and

γ_{t}

are employed to regulate province and year fixed effects, respectively;

ϵ_{i t}

is the stochastic error term. The coefficients

β_{1}

β_{6}

are parameters to be estimated, with primary focus placed on

β_{1}

. This study utilizes the plm function within the R (version 4.4.3) package “plm” for panel data analysis; consequently, the intercept term is not displayed [45].

3.1.2. Mediating Effect Model

To further investigate the pathways through which INT affects GTFP, this study constructs the following mediating effort models based on prior theoretical analysis and by referencing the approach proposed by Wen [46]. These models empirically assess the transmission mechanisms of green innovation (GI) and spatio-economic synergy (SES).

(2) ${G T F P}_{i t} = δ_{0} + δ_{1} {I N T}_{i t} + \sum_{k = 2}^{6} δ_{k} {C o n t r o l}_{k, i t} + α_{i} + γ_{t} + ϵ_{i t}$

(3) ${G T F P}_{i t} = μ_{0} + μ_{1} {I N T}_{i t} + μ_{2} M_{i t} + \sum_{k = 3}^{7} μ_{k} {C o n t r o l}_{k, i t} + α_{i} + γ_{t} + ϵ_{i t}$

in Equations (2)–(3),

M_{i t}

is the mediator variable. GI and SES are selected to assess the mediating mechanisms of INT on GTFP. The term

{\sum_{k = 3}^{7} C o n t r o l}_{k, i t}

denotes the complete set of control variables, while all other symbols maintain their definitions from Equation (1). Within this mediation model, the primary analytical focus rests on the coefficients

δ_{1}

and

μ_{1}

3.1.3. Moderating Effect Model

To deepen our understanding of how INT influences GTFP, this study specifies a moderating effect model incorporating an interaction term between INT and ER. Building on this foundation, the moderated mediation models are further established.

(4) $\begin{matrix} G T F P_{i t} = θ_{0} + θ_{1} I N T_{i t} + θ_{2} E R_{i t} + θ_{3} (I N T \times E R)_{i t} + \sum_{k = 4}^{8} θ_{k} {C o n t r o l}_{k, i t} + α_{i} + γ_{t} + ϵ_{i t} \end{matrix}$

(5) $\begin{matrix} M_{i t} = ρ_{0} + ρ_{1} I N T_{i t} + ρ_{2} E R_{i t} + ρ_{3} (I N T \times E R)_{i t} + \sum_{k = 4}^{8} ρ_{k} {C o n t r o l}_{k, i t} + α_{i} + γ_{t} + ϵ_{i t} \end{matrix}$

(6) $G T F P_{i t} = φ_{0} + φ_{1} I N T_{i t} + φ_{2} E R_{i t} + φ_{3} (I N T \times E R)_{i t} + φ_{4} M_{i t} + {\sum_{k = 5}^{9} φ}_{k} X_{k, i t} + α_{i} + γ_{t} + ϵ_{i t}$

in Equations (4)–(6),

E R_{i t}

denotes the moderating variable, representing environmental regulation intensity in province i and year t. The interaction term

(I N T \times E R)_{i t}

captures the joint effect of INT and ER. The coefficients that are of primary interest are

θ_{3}

ρ_{3}

and

φ_{3}

. All other variables and symbols are consistent with the definitions in Equations (1)–(3).

3.2. Variable Measures

3.2.1. Explained Variable

Green Total Factor Productivity (GTFP). Within the mainstream literature on green economy, the level of green development is predominantly characterized by either total factor productivity metrics or single-factor indicators [22,23,41,47]. While single-factor metrics provide intuitive measurements of resource utilization that align with China’s existing statistical framework and energy emission reduction targets, they exhibit inherent limitations, including a constrained informational scope, inadequate capacity for comprehensive efficiency assessment, and inability to evaluate economic development quality. In addition, such indicators are inadequate for estimating potential improvements in resource efficiency under given technological conditions, and the substitution relationship among input factors may distort the accurate measurement of productivity [48]. Consequently, this study adopts GTFP as the principal indicator for quantifying green development performance.

Regarding the specific quantification of GTFP, most scholars primarily employ parametric and non-parametric methods. As demonstrated by Gong & Sickles [49], non-parametric Data Envelopment Analysis (DEA) outperforms parametric methods such as Stochastic Frontier Analysis (SFA) in handling multi-dimensional inputs and outputs. Since Charnes et al. [50] proposed DEA, numerous scholars have employed this methodology for efficiency analysis, with further refinements and enhancements being made over time. Notably, Tone et al. [51] proposed an Epsilon-Based Measure (EBM) model that integrates radial and non-radial distance functions. This model boasts dual advantages, on the one hand considering the radial ratio between the input frontier value and the actual value, and on the other hand reflecting the non-radial relaxation variables that differentiate between the input variables. Crucially, the EBM model quantifies the potential for input changes while maintaining output stability. Moreover, it overcomes the efficiency overestimation or underestimation biases inherent in CCR and SBM models, thereby effectively improving the accuracy of the measurement results. Consequently, building upon established methodologies [47], we quantify GTFP using the following EBM specification:

$γ^{*} = \min_{θ, s^{-}} θ - ϵ \sum_{i = 1}^{m} \frac{w_{i}^{-} s_{i}^{-}}{x_{i 0}}$

$s . t . & \sum_{j = 1}^{n} λ_{j} x_{i j} = θ x_{i 0} - s_{i}^{-}, i = 1, \dots, m$

$\sum_{j = 1}^{n} λ_{j} y_{r j} \geq y_{r 0}, r = 1, \dots, s$

(7) $λ_{j} \geq 0, j = 1, \dots, n, s_{i}^{-} \geq 0, i = 1, \dots, n$

in Equation (7), γ* denotes the optimal efficiency score; λ represents the weight assigned to input i; and εₓ, a scalar between 0 and 1, serves as the core parameter capturing the proportion of changes in radial and non-radial variables. The application of the EBM model for GTFP quantification necessitates the prespecification of both εₓ and w⁻ᵢ.

Table 1 presents the GTFP indicator system. Following mainstream practices in environmental economics [10,14,17], this study measures GTFP based on three dimensions: factor input, desirable output, and undesirable output. Specifically, factor input encompasses labor, capital, and energy. Labor input is measured by year-end total employment in each region. Capital input is quantified by the fixed asset stock, calculated using the perpetual inventory method with depreciation rates, using 2011 as the benchmark period. Energy input is evaluated using the total energy consumption of each region. The rationale for selecting these three dimensions lies in the classical Cobb-Douglas production function and its extensions, where capital and labor are traditionally identified as core drivers of output. Given the increasingly critical role of energy in modern industrial production and its typically high-pollution nature, energy has also been widely incorporated into production functions [47,52], establishing a tri-factor analytical framework specifically tailored for environmental production efficiency. Notably, while land is sometimes recognized as an input with justification in certain studies, particularly those focused on agricultural GTFP, it is omitted here for several reasons. First, in broader macroeconomic GTFP analyses beyond agriculture, excluding land as a direct input is common practice [17,47,52,53], as its influence is often captured within capital inputs (e.g., factories and infrastructure) and regional fixed effects rather than land area itself. Second, land quantification is challenging. Land prices in China are significantly distorted by local government actions and market behavior, making quantification based on transaction prices or nominal rents prone to measurement errors, while using land area overlooks acquisition difficulty. Third, for non-agricultural production, land primarily serves as a spatial carrier for activities rather than acting as a continuously consumed variable input that directly affects output elasticity, unlike capital, labor, and energy.

Regarding output measures, the desired output is represented by regional real GDP. To eliminate the effects of price fluctuations, this value is deflated using provincial GDP deflators with 2011 as the base year [37]. The undesirable output is quantified using regional emissions of industrial sulfur dioxide, industrial wastewater discharge, and general industrial solid waste, following the methodology of Zhang & Dong and Zhou & Zhang [53,54]. This selection is justified not only because these are key monitoring indicators in China’s Environmental Statistical Bulletin, but more importantly, as they collectively represent the three primary categories of pollutants (gaseous, liquid, and solid) generated by production activities, thereby providing a more comprehensive reflection of the environmental negative externalities than a single pollutant type or a combination of other indicators.

3.2.2. Explanatory Variable

Industrial Intelligence (INT). Currently, no unified standard exists for measuring INT development. Existing literature predominantly relies on single indicators, primarily including the following five methods: (1) industrial robot installation density [6]; (2) AI patent applications [55]; (3) fixed-asset investment in information transmission and software industries [56]; (4) practitioner surveys [57]; and (5) INT keyword frequency or Internet search indices [58]. While these metrics benefit from data accessibility and exhibit respective rationality, they suffer from two significant limitations. First, single indicators are susceptible to substantial randomness and are not well-suited to capture the multidimensional attributes of INT as a complex system, thereby lacking precision in reflecting its pervasive and synergistic characteristics [47]. Second, singular variables risk conflating non-INT technological indicators, such as traditional automation and manufacturing, inducing measurement errors and endogeneity issues.

To overcome the limitations identified above, this study constructs a multidimensional comprehensive indicator system by drawing upon the conceptual framework of the National Innovation Index Report 2022 and multidimensional measurement research [59,60,61]. The system encompasses three primary indicators, further refined through 11 secondary indicators, following the “technology-application-value” logic. First, the technological innovation environment serves as a dimension gauging the technical potential of INT development, determining a region’s capacity for INT technology iteration. This study selects general innovation capability, R&D funding, and AI innovation achievements as secondary indicators. General innovation capability represents the macro-level innovation climate; R&D funding constitutes a core metric for assessing innovation support intensity; and AI innovation achievements reflect technological output within the AI domain. Second, ndustrial development foundation assesses the industrial application potential of INT development, reflecting industries’ capacity to translate INT technologies into tangible productive forces. Given INT’s high dependence on data and computing power, this study employs Internet coverage and information technology infrastructure as key indicators for quantifying the digital infrastructure foundation. INT human resources represent a reserve of professionals possessing INT skills, while INT capital reflects specialized investments in INT-related software, hardware, and other resources. Together, these four elements constitute the core enablers of the deep integration of INT technology into the industry. Third, industrial output level measures the practical effectiveness of INT development, quantifying its actual economic value across the dimensions of application depth and market scale. The installation density of industrial robots is widely adopted as a proxy variable for industrial intelligence, embodying the depth of INT application within production processes. The quantity and contribution of INT enterprises, coupled with the business revenue of the INT industry, collectively characterize the market scale of INT as an emerging industry, reflecting the ultimate outcomes of INT value. This approach circumvents the inherent partiality of single-indicator measurements and better aligns with the complex systemic essence of AI-driven industrial intelligence. Detailed indicators and their corresponding weights are presented in Table 2.

This study employs the entropy method to determine the weights of the indicators at each tier to assess the INT development level. The procedure begins with data standardization, followed by the calculation of information entropy and variation coefficients, culminating in the derivation of objective weights. Through hierarchical aggregation, secondary indicators are weighted into primary indicators and subsequently synthesized into a comprehensive index. The advantages of this approach are twofold. First, objectivity. The weights are determined by the inherent degree of dispersion within the data itself, thereby mitigating the influence of expert bias inherent in subjective weighting. Second, the information content is maximized. The entropy method automatically assigns higher weights to indicators that exhibit greater variability. This mechanism ensures prominent recognition of breakthrough advancements in critical metrics, thereby aligning with the requirements of INT evaluation. The specific calculation formulas are as follows:

(1). The weight matrix (P) is calculated by measuring the distribution ratio of each sample for a certain indicator to reflect the relative importance of the data.

(8) $p_{i j} = \frac{x_{i j}}{\sum_{i = 1}^{n} x_{i j}}, j = 1,2, . . ., m$

(2). Calculate information entropy (e): Measure the degree of dispersion of indicator data. The smaller the entropy, the greater the amount of indicator information.

(9) $e_{j} = - k \sum_{i = 1}^{n} p_{i j} \ln (p_{i j}), k = \frac{1}{\ln (n)}$

(3). The difference coefficient (d) is calculated as 1—entropy value, reflecting the discriminating ability of the indicator. The larger the difference coefficient, the higher is the weight.

(10) $d_{j} = 1 - e_{j}, j = 1,2, . . ., m$

(4). Calculate weights (w): Normalize the difference coefficients to obtain the final weights for each indicator.

(11) $w_{j} = \frac{d_{j}}{\sum_{j = 1}^{m} d_{j}}, j = 1,2, . . ., m$

(5). Step-by-step calculation of the composite index (S).

(12) $S_{i} = \sum_{j = 1}^{m} w_{j} \cdot x_{i j}$

3.2.3. Mediating Variables

Green Innovation (GI). Current mainstream literature predominantly measures GI through the absolute count of green patent applications or their proportion to total patents [63]. However, these approaches have three critical limitations. First, reliance on absolute metrics, such as patent counts, neglects regional innovation investment levels and impedes an accurate assessment of GI efficiency and quality. Secondly, proportional measurements, such as the proportion of green patents, are confounded by regional development policies and industrial structures, resulting in significant estimation errors. Thirdly, there is a pervasive “patent bubble” phenomenon in the domain of GI. The presence of ESG ratings and government regulations compels innovative activities to prioritize quantity-over-quality pursuits, resulting in inflated green patent volumes coupled with declining innovation quality. Consequently, using aggregate green patent applications to measure green innovation fundamentally compromises validity [64,65]. Therefore, building upon the methodologies of Liu et al. and Shu et al. [66,67], this study adopts the ratio of green invention patent applications to R&D investment as a proxy variable for GI, which is justified by several key rationales. (1) Green invention patents typically involve longer authorization cycles and greater application complexities, which help mitigate the distortions caused by patent bubbles to a significant degree [68,69]. (2) Since R&D investment serves as a primary driver of green innovation, measuring applications per unit investment effectively captures the allocation efficiency of R&D resources, thereby avoiding biases from overemphasizing patent outputs while neglecting innovation costs and enabling a more comprehensive assessment of GI’s economic viability. (3) R&D investment preserves the accuracy of patent-based innovation output measurement, enhancing GI’s comparability across time periods and minimizing the confounding effects of regional heterogeneity on GI evaluation.

Spatio-Economic Synergy (SES). SES is commonly measured using GDP density or single-factor densities, such as employment density. Ciccone and Hall [70] proposed that GDP density, by synthesizing the dual influences of population and economic output, offers a more comprehensive measure of the integrated intensity of factor synergy within spatial units and provides a more accurate depiction of economic activity distribution than employment density. Its advantage lies in integrating the final output of multidimensional factors, including labor, capital, and technology, thereby directly reflecting the value creation capacity driven by spatial agglomeration and capturing the essence of SES more effectively than measures based on a single factor. For instance, agriculture and manufacturing may exhibit significantly divergent economic outputs despite similar employment densities, serving as evidence of spatial heterogeneity in SES. Furthermore, this study uses provincial-level samples as the research object, benefiting from their greater stability in administrative areas during the sample period relative to other spatial units. Consequently, following the approach of Yao et al. [71] and accounting for inflation, SES is measured as the ratio of real GDP (using 2011 as the base period and deflating according to the GDP deflator) to regional area.

3.2.4. Moderating Variable

Environmental Regulation (ER). Given the absence of a unified measurement methodology for ER in current academic research, this study adopts an approach to capture the comprehensive impact of diverse regulatory types. Drawing on the methodology of Zhao [72] and Dai et al. [73], ER is categorized into four dimensions: policy-oriented, tax-adjusted, public supervision, and industrial governance. The entropy weight method is employed to calculate the comprehensive index, and the specific indicator design is shown in Table 3.

First, there is the Policy-Oriented ER. Initially, environmental policy documents issued between 2012 and 2023 were manually collated from the Ministry of Ecology and the regional administrative departments. Following Python word segmentation processing, environmentally relevant terms were identified and incorporated into the ER lexicon. Subsequently, provincial government work reports were analyzed by fuzzy matching between the constructed lexicon and target documents. Accounting for variations in document length across regions, the final metric is defined as the ratio of the frequency of ER terms to the total word count in each province’s policy document. Second, tax-adjusted ER was measured as the ratio of the sum of environmental protection tax and resource tax to regional GDP, consistent with Xiong’s methodology [74]. Third, public-supervision ER employed the annual Baidu Search Index for the term “environmental pollution” as a proxy [75]. Fourth, the definition of industrial-governance ER is rooted in foundational principles articulated by the classification framework delineated in the Green Industry Guidance Catalogue and refers to Shu et al.’s approach [76], manually screening and counting the year-end stock of enterprises in the CPPGD for energy-saving and environmental protection industries, clean production industries, clean energy industries, ecological environment industries, green infrastructure upgrades, and green service industries, thus serving as a proxy variable.

3.2.5. Control Variables

To mitigate endogeneity concerns arising from omitted variables and reduce estimation bias, the following control variables were selected based on prevailing literature:

Financial support level (Fin): Measured as the ratio of regional deposits and loans to GDP. While INT adoption relies on financial resources throughout its R&D-commercialization lifecycle [77], a concurrent increase in pollution emissions may be observed in high-pollution industries due to reduced financing costs [78].

Foreign direct investment (FDI): Represented by the ratio of utilized FDI to GDP. FDI inflow has dual environmental effects. Advanced green production technologies transferred by multinational enterprises generate spillover and halo effects that improve local environmental practices [79,80]. However, FDI may concurrently facilitate pollution haven effects through relocating polluting industries.

Economic development level (GRP): Measured using the logarithm of real GRP per capita (using 2011 as the base period for deflation). Developed economies typically place greater emphasis on environmental governance and attract talent, enabling substantial investment in green infrastructure and regulation. Control is further necessitated because of the Environmental Kuznets Curve mechanism.

Industrialization level (Indus): Measured as the industrial output-to-GDP ratio. Compared to agriculture and service industries, the industrial sector is often dominated by energy-intensive industries that emit substantial quantities of carbon dioxide, thereby having a more significant impact on GTFP.

Government intervention intensity (Gov): Quantified by the ratio of general budget expenditure to GDP. On one hand, a suitable fiscal decentralization system endows governments with fiscal autonomy, which is conducive to promoting rational allocation of resources and potential for green transformation [81]; On the other hand, if governments place excessive emphasis on economic growth and ignore negative externalities on the environment, it will lead to an increase in undesirable output [82].

3.3. Data Sources and Descriptive Statistics

Considering the validity and availability of data, this study selects panel data spanning 2012 to 2023 from 30 provincial administrative regions in China (excluding Tibet Autonomous Region, Hong Kong, Macau, and Taiwan) as samples. Data sources encompass the National Bureau of Statistics of China, China Statistical Yearbook, China Statistical Yearbook on Science and Technology, China Statistical Yearbook on Environment, China Statistical Yearbook on Regional Economy, China Statistical Yearbook on High Technology Industry, China Statistical Yearbook on Population and Employment, and regional statistical yearbooks. Furthermore, to ensure the robustness and comparability of the dataset, this study addresses data quality issues through a systematic preprocessing pipeline. For missing values, we adopted a hybrid interpolation approach: linear interpolation for monotonically trending variables and moving-average imputation for volatile indicators. All variables were winsorized at the 1st and 99th percentiles. Monetary variables were deflated to 2011 constant prices using the provincial GDP deflator. Cross-source discrepancies were reconciled through backward verification using provincial statistical bulletins. Table 4 presents the descriptive statistics for the main variables.

Figure 2 illustrates the evolving trend of industrial intelligence average development in China, while Figure 3 and Figure 4 depict the regional heterogeneity of INT and GTFP, respectively. The figures reveal that, with respect to INT, the national level exhibited a consistent upward trajectory from 2012 to 2023. Spatially, a distinct pattern emerged, characterized by higher levels in the southeast and lower levels in the southwest, indicating significant regional heterogeneity. Similarly, GTFP displayed analogous spatial characteristics, being predominantly higher in the southeast and lower in the southwest. Notably, certain regions witnessed elevated levels of GTFP attributable to policy interventions.

4. Empirical Results and Analysis

4.1. Benchmark Regression

Table 5 reports the results of the benchmark regression. To verify Hypothesis 1, this study first regresses INT as the sole explanatory variable, followed by sequential incorporation of a series of control variables, including Fin, FDI, GRP, Indus, and Gov, into the fixed effects model, comprehensively examining the impact of INT on GTFP. Regression results are presented in columns (1)–(6). The coefficient of INT in column (1) is 0.356, which is significant at the 1% level. As control variables are progressively incorporated, the INT coefficient demonstrates a downward trend while remaining statistically significant. In column (6), which incorporates all controls, the INT coefficient registers at 0.305 and remains significant at the 1% level. This confirms that INT development can significantly promote GTFP, thereby validating Hypothesis 1.

It should be noted that although the adjusted R² in Column (1) is negative, its occurrence is theoretically justifiable. The calculation formula for adjusted R² incorporates a penalty term for model degrees of freedom (i.e., an adjustment for the number of explanatory variables). When a model contains only a limited set of variables with weak explanatory power for the dependent variable, the degrees of freedom penalty may result in a negative adjusted R². Model (1), as the initial specification, includes only the core explanatory variable. However, since GTFP is strongly influenced by external factors, the explanatory power of the model under this specification is limited, with a negative adjusted R². After introducing control variables, the adjusted R² significantly improves and consistently attains positive values. This demonstrates that including the control variables effectively enhances the model’s explanatory power.

According to the control variable data in column (6) of Table 5, the regression coefficient for Fin is −0.102, significant at the 1% level, indicating a negative correlation with GTFP. This phenomenon may be attributed to the current disproportionate allocation of credit resources to high-pollution industries. Such financial misallocation triggers an innovation crowding-out effect, resulting in insufficient investment in green technology R&D.

The coefficient for FDI is −6.059, with a significance level of 1%, confirming its negative association with GTFP. A plausible explanation lies in the inadequate implementation of green practices within China’s foreign trade sector, coupled with its ongoing transition from a quantity-driven to a quality-driven paradigm. Furthermore, the majority of FDI inflows are directed toward high-pollution and low-end industries relocated from developed countries [83].

The coefficient for GRP is 0.068, which is significant at the 10% level, indicating that GRP growth enhances GTFP to some extent. This positive correlation can be attributed to China’s sustained economic expansion in recent years, which has established a robust foundation for advancing green production technologies and accelerating the green transition. Concurrently, rising economic development levels have elevated living standards and shifted consumption preferences, thereby stimulating the demand for green products and services.

The coefficient for Indus is 0.341 with 1% significance, demonstrating that industrial advancement positively affects GTFP. Given China’s current high level of industrialization, carbon-intensive industries are progressively being replaced by advanced manufacturing driven by clean energy [84,85]. Consequently, the marginal benefit of industrialization on desirable outputs exceeds its marginal cost for undesirable outputs, resulting in net GTFP gains.

The coefficient for Gov is −0.060, which is insignificant. This insignificant suppression stems from two mechanisms. First, as a developing country, China still prioritizes fiscal expenditures on economic affairs. Although there has been a gradual shift toward green investments in recent years, negative environmental externalities have not been fully mitigated. Second, increased ER expenditures may induce GI bubbles and reduce innovation quality due to inefficient resource allocation [86,87].

4.2. Robustness Tests

To ensure the robustness of the baseline regression results, this study conducts the following six validation tests, with specific results presented in Table 6.

4.2.1. Replace the Explained Variable

To mitigate potential bias caused by nonlinear processing of slack variables in the super-efficiency EBM model, GTFP is recalculated using the non-super-efficiency EBM model. This approach aims to eliminate heterogeneity in efficiency frontier assumptions inherent to super-efficiency models and disentangle directional constraints of slack variables on evaluating INT’s impact on GTFP, thereby ensuring consistency of core conclusions across different production boundary hypotheses. As shown in column (1), INT exhibits a significant positive coefficient at the 1% level, confirming the robustness of this conclusion.

4.2.2. Replace the Explanatory Variable

Given that divergent INT metrics may yield heterogeneous regression results, we adopt mainstream measurement methods from Chun & Hwang [88] and Yin [89], replacing the explanatory variable with AI-related patent counts (AI_Tech). The advantages of this metric lie in its objectively quantifiable nature and precise targeting, which directly captures substantive outputs of AI innovation, a key driver of INT. Column (2) shows that AI_Tech is positively significant at the 1% level, indicating that INT advancement significantly enhances GTFP.

4.2.3. Exclude Special Samples

To mitigate potential bias resulting from sample heterogeneity, we exclude all autonomous regions for regression analysis following the methodologies of Liu & Liu [90] and Peng & Zhang [91]. This exclusion accounts for socioeconomic particularities of these regions, including high autonomy, unique policies, and distinct ecological environments. In such regions, underdeveloped industrial bases and nascent INT industries may distort the mechanism through which INT affects GTFP. As column (3) indicates, INT remains positively significant at the 1% level, confirming the core conclusion’s generalizability across primary samples.

4.2.4. Adjust the Sample Period

Considering that the abnormal fluctuations in economic activities caused by China’s COVID-19 containment policies may disrupt the identification of intrinsic INT-GTFP linkages [92], we exclude observations from the pandemic period (2020–2022) to eliminate the confounding effects of exogenous shocks. This tests the robustness of baseline results under conventional economic conditions. Column (4) further validates the persistence of these key findings.

4.2.5. Incorporate Additional Control Variables

To further verify robustness and address potential confounding effects caused by omitted variables, three critical control variables from extant literature [93,94,95] were integrated, including population density, resource consumption intensity, and industrial structure. By controlling for their systematic influence on GTFP, this approach strengthens the explanatory power and model adaptability of INT’s estimated impact on GTFP. Column (5) demonstrates that INT retains positive significance at the 1% level, affirming the robustness of the conclusion.

4.2.6. Employ the Driscoll-Kraay Standard Errors

Given that provincial panel data may be influenced by spatial correlations and policy transmission effects, leading to issues such as cross-sectional contemporaneous correlation, within-group autocorrelation, and between-group heteroscedasticity, which can underestimate uncertainty, this study follows Meng et al. [96] and employs Driscoll-Kraay standard errors for correction. This method can simultaneously handle all three problems and is robust to cross-sectional heterogeneity and spatial correlations. The results in column (6) demonstrate that the INT coefficient remains positively significant at the 1% level, indicating strong conclusion robustness.

4.3. Endogeneity Tests

In the process of INT development, enhancing GTFP, potential challenges of endogeneity may arise. Improvements in GTFP not only benefit environmental performance but also reciprocally foster economic upgrading, efficiency gains, and industrial transformation, thereby establishing a robust foundation for INT advancement. Concurrently, green innovation and R&D investment, driven by both policy and financing environments, may also exhibit endogeneity with INT.

To address the endogeneity issue arising from bidirectional causality, this study employs the two-stage least squares (2SLS) method with the following two instrumental variables (IVs). First, provincial mobile telephone penetration (IV1_Tel) is measured by mobile telephone exchange capacity (in 10,000 units) following Zhu [97]. This metric satisfies the relevance condition, as exchange capacity reflects regional IT infrastructure to a certain extent, which is a foundational element for INT development. Concurrently, the economic benefits of telephone penetration require long-term accumulation and have no immediate impact on GTFP enhancement, thus partially fulfilling exogeneity. Second, drawing on the practices of Ouyang [98] and Zhao [99], an exogenous policy shock (IV2_Policy×Post) is constructed using the dummy variable “Policy × Post” of the National New Generation Artificial Intelligence Innovation and Development Pilot Zones. The provincial policy variable, which the city belongs to, equals 1 if the city is a pilot city (otherwise 0); Post is a time dummy set to 1 for years when the policy takes effect and thereafter (otherwise 0). This indicates whether province i had a city designated as a pilot zone in year t. Crucially, this IV exploits exogeneity and time lag of policy shock: pilot city selection is independent of concurrent GTFP, while policy effects materialize gradually, which can effectively isolate the driving effects of policy and financing environment on GTFP.

Table 7 presents the results of the endogeneity test regression. In the first stage, both IV1_Tel and IV2_Policy×Post exhibit statistically significant positive effects at the 1% level (coefficients = 0.067 and 0.054), indicating their capacity to drive INT development. During the second stage, the positive effects of INT on GTFP are significant, confirming that INT promotes GTFP growth after improving infrastructure and policy support, which is consistent with core findings. Meanwhile, both Kleibergen-Paap LM statistics (35.353 and 57.837) and Cragg-Donald Wald F statistics (44.387 and 106.155) significantly reject the null hypothesis of weak instruments, validating the strong correlation and effectiveness of the IVs. Collectively, these results demonstrate that INT retains its positive effect on GTFP after addressing potential endogeneity bias from bidirectional causality.

4.4. Heterogeneity Tests

According to the extant literature, the impact of INT on GTFP varies significantly across regions with distinct industrial structures [100]. Consequently, this study conducts regressions using two subsample classification methods: whether a province is industry-dominated and its level of industrial structure. Table 8 reports the regression results of heterogeneity tests.

In the first classification, the sample is divided into industry-dominant regions (IndDom) and non-industry-dominant regions (Non_IndDom) based on the proportion of secondary industry in GDP for each region [101]. Columns (1) and (2) reveal that INT positively enhances GTFP at the 1% significance level in Non_IndDom, whereas its coefficient is negative and insignificant in IndDom. This divergence may stem from the current concentration of high-energy-intensive traditional manufacturing in IndDom, where production processes and technological systems exhibit strong path dependence. INT applications in these sectors likely remain in the initial automation-substitution phase, focusing on efficiency gains without yet integrating clean production technologies. Moreover, short-term increases in energy consumption may generate adverse environmental effects. Conversely, Non_IndDom features higher proportions of services or emerging industries, which are inherently characterized by lower environmental burdens. Their industrial structures are more amenable to INT-driven intelligent solutions, such as smart logistics and optimized power grids, thereby enabling INT to significantly promote GTFP.

In the second classification, the sample is divided into regions with low, medium-level, and high-level industrial structures (Low_IndStr, Med_IndStr, High_IndStr) based on the ratio of tertiary industry output value to secondary industry output value across regions. As columns (3)–(5) demonstrate, INT has a statistically significant positive impact on GTFP in both Med_IndStr (coefficient 0.303, p < 0.01) and High_IndStr (coefficient 0.423, p < 0.01). Conversely, INT exhibits an inhibitory effect in Low_IndStr (coefficient −0.628, p < 0.01), and this effect is statistically significant. This divergence can be attributed to the robust industrial foundations and conducive developmental environments in Med_IndStr and High_IndStr, which provide systematic support for INT advancement and amplify its efficacy in enhancing GTFP. In contrast, Low_IndStr faces intrinsic constraints, such as factor endowment limitations, a disproportionately high traditional industrial composition, and inadequate digital infrastructure, collectively impeding the maturation of the INT. Moreover, its short-term development priorities, which predominantly focus on energy-intensive industrial infrastructure, further suppress GTFP, thereby generating negative coefficients.

5. Mechanism Analysis

5.1. The Mediating Effect of Green Innovation and Spatio-Economic Synergy

A mechanism analysis is essential to elucidate the specific pathways through which INT enhances GTFP and to inform region-specific implementation strategies for green development. As posited in the theoretical analysis, INT elevates GTFP primarily by intensifying GI and SES. The empirical results for these pathways are presented in Table 9. Given concerns regarding the low statistical power of conventional stepwise regression for mediation testing, this study employs the Sobel test to verify the significance of mediating effects. Furthermore, to address the Sobel test’s reliance on normality assumptions, we supplement the analysis with a Bootstrap test sampling 1000 replications, thereby enhancing estimation robustness and accuracy [102].

For the green innovation effect, on the one hand, the coefficient of INT in column (1) is 0.534 and significant at the 1% level, indicating a significant positive effect of INT on GI. On the other hand, in column (2), the coefficient of INT decreases to 0.224 with its significance maintaining the 1% level (compared with the benchmark regression results), while the coefficient of GI is 0.152 and significant at the 1% level. This pattern suggests that GI plays a partial mediating role in the process through which INT enhances GTFP, thereby attenuating the direct impact of INT. The calculation indicates that the green innovation effect accounts for approximately 26.612% of the total effect. Furthermore, the Sobel test statistic is 2.912 (p < 0.01), and the conclusion remains robust upon estimation of the Bootstrap test, further substantiating that INT not only directly improves GTFP but also indirectly enhances environmental performance by boosting green innovation efficiency.

Regarding the spatio-economic synergy effect, the empirical results demonstrate that SES has a significant partial mediating role in the relationship between INT and GTFP. Column (3) reveals that the coefficient of INT’s impact on SES is 0.478 (p < 0.01) (compared with the benchmark regression results), while in Column (4), after introducing the mediating variable, the direct effect of INT on GTFP diminishes to 0.196 (p < 0.05), and SES exhibits a coefficient of 0.229 (p < 0.01). This pattern suggests that the spatio-economic synergy effect accounts for approximately 35.889% of the total effect. Furthermore, both the Sobel test (z = 3.281, p < 0.01) and the Bootstrap test (95%CI: [0.068, 0.169]) substantiate the statistical significance of the mediating effects. These findings substantiate that INT not only directly enhances GTFP but also indirectly improves green development performance by fostering regional spatio-economic synergy.

Collectively, INT elevates GTFP through two pathways: enhancing green innovation efficiency and strengthening spatio-economic synergy. This empirical evidence robustly supports Hypothesis 2.

5.2. The Moderating Effect of Environmental Regulation

To further elucidate the mechanisms underlying INT’s impact on GTFP, this study introduces an interaction term between ER and INT into both the baseline and mediating effect models to test the moderating role of ER. The regression results are presented in Table 10.

As shown in column (1), the coefficient of the interaction term INT×ER is −0.631, and is statistically significant at the 1% level. Simultaneously, the coefficients of both INT and ER are significantly positive, with the coefficient of INT (0.555) exceeding that of the benchmark regression (0.305). This indicates that ER can positively promote GTFP improvement, and the promotional effect of INT on GTFP becomes more pronounced when accounting for ER’s influence. Notably, however, the promotional effect of INT is attenuated as ER intensity increases, indicating that the ER has a substitution effect in the relationship between INT and GTFP. A plausible explanation is that ER directly constrains undesirable outputs through its guiding or coercive mechanisms, thereby enhancing GTFP, while also compelling enterprises towards green transformation, enabling them to utilize INT more efficiently for emission reduction. Conversely, an excessively high ER leads to the crowding out of innovation expenditure and the emergence of threshold effects. On the one hand, stringent ER may increase corporate compliance costs, diverting resources away from technological R&D or application, thereby weakening INT’s promotional influence [103]. On the other hand, when ER reaches a certain level of stringency, it may itself have maximized GTFP improvements through various means [104]. At this juncture, the green optimization potential of INT approaches its upper limit, the space for additional impact is constrained, and diminishing marginal returns are observed.

Columns (2)–(5) present regression results of the moderated mediation model. The findings are consistent with previous conclusions; INT exhibits significantly positive coefficients in regressions with both GI and SES. After controlling for the mediators, the INT coefficient remains significantly positive, indicating that INT still enhances GTFP through the green innovation and spatio-economic synergy effects, even when ER is accounted for. The INT×ER interaction term consistently demonstrates significantly negative coefficients, revealing that ER not only attenuates the direct impact of INT on GTFP but also negatively moderates the mediating pathways of GI and SES, thereby weakening INT’s indirect influence on GTFP.

Collectively, the ER exhibits a negative moderating effect on both the direct and indirect impacts of INT on GTFP, validating Hypothesis 3.

6. Further Analysis: Time-Lagged Effect

Considering the potential time-lag effect of INT’s green empowerment effect, this study adopted the approach of Zhou et al. [47] and determined the lag period for technology innovation application and transformation as one year. Accordingly, INT with a one-year lag (INT_lag1) was employed as the explanatory variable to test for a time-lag effect. The regression results of the time-lagged test are presented in Table 11. As shown in column (1), the coefficient of INT_lag1 is 0.250, which is statistically significant at the 1% level. This indicates a significant time lag in the promotion effect of INT on GTFP, thus validating Hypothesis 4.

Following the examination of the time-lag effect of INT on GTFP, this study incorporates the interaction term INT_lag1×ER to test the moderating role of ER in the lag period. Column (2) reveals that the coefficient of this interaction term is significantly negative, while the coefficients of both INT_lag1 and ER are significantly positive. This indicates that the substitution effect of ER remains significant within the time-lagged impact of INT on GTFP. To delineate the process of the ER substitution effect, the sample is partitioned based on the tertiles of ER for stratified regression analysis, with results presented in columns (3) to (5). The results reveal significant heterogeneity in the impact of INT_lag1 on GTFP. It exhibits a strong and significantly positive effect in the Low-ER group (coefficient = 0.721), a weakened and insignificant positive effect in the Med-ER group (coefficient = 0.152), and a significantly negative effect in the High-ER group (coefficient = −0.410). These findings demonstrate the existence of a substantial ER threshold and substitution effect concerning INT’s green empowerment effect. As ER intensity increases, this empowerment effect is progressively replaced. Overall, the influence of ER displays an inverted U-shaped transition, which aligns with theoretical expectations.

7. Conclusions, Discussions, and Policy Implications

7.1. Conclusions and Discussions

Based on panel data from 30 provincial administrative regions in China from 2012 to 2023, this study constructs a three-dimensional INT comprehensive indicator system. This system provides a more reasonable measure of INT. GTFP was calculated using the super-efficiency EBM model, significantly enhancing the scientific rigor of the measurement. Furthermore, an ER moderating variable was incorporated to analyze its role within the INT-GTFP relationship, addressing a gap in the existing literature concerning ER mechanism analysis. The principal conclusions are as follows.

First, INT has a significant positive impact on GTFP. This conclusion holds robustly after multiple tests, validating the common findings of prior research. Additionally, the promoting effect of INT is more pronounced in regions with advanced industrial structures, consistent with previous studies.

Second, INT indirectly enhances GTFP by fostering green innovation and strengthening spatio-economic synergy. Specifically, INT stimulates green innovation by optimizing R&D processes, reducing trial-and-error costs, and accelerating knowledge spillover. Concurrently, its application promotes the agglomeration of green industries and talent, reinforcing scale effects and accelerating the diffusion of technologies.

Third, ER negatively moderates both the direct and indirect effects of INT on GTFP, indicating a significant substitution effect. The promoting effect of INT exhibits pronounced heterogeneity under varying ER intensities. In contrast to prior research, this study is the first to propose ER’s role as a “substitutor” within the INT-GTFP relationship. This finding represents an important supplement and revision to the Porter Hypothesis in the digital era, suggesting that the inherent technological progress and efficiency optimization within INT can partially substitute for the external pressure imposed by ER, jointly serving the ultimate goal of green development.

Fourth, the positive impact of INT on GTFP exhibits a time-lag effect, and ER’s substitution effect remains valid during the lag period. This extends the Theory of Diffusion of Innovations into the INT domain, indicating that realizing the environmental benefits of INT requires a temporal process. Crucially, ER’s substitution effect is not a transient phenomenon but a persistent influence operating throughout the entire process.

7.2. Policy Implications

Based on the above conclusions, the pivotal challenges and potential pathways for China within the INT and green development become evident. To maximize INT’s green dividends and propel high-quality, sustainable economic development, this study proposes the following policy recommendations:

First, macro strategies that synergize INT advancement with green development should be formulated, underpinned by steadfast policy consistency and a long-term perspective. The government should position the development and application of INT as a core component of the national green transition strategy, elevating its role beyond a mere industrial policy tool to a critical infrastructure for achieving “dual carbon” goals and sustainable development. This necessitates not only intensified support for INT foundational research, computing infrastructure, and data factor markets but also proactive facilitation of deep integration between INT and green scenarios like energy conservation, emission reduction, pollution control, and ecological restoration. Furthermore, recognizing that INT’s empowerment of GTFP is inherently long-term and exhibits effect lag, policymakers must demonstrate considerable strategic patience and avoid frequent policy adjustments driven by lackluster short-term outcomes. Therefore, a multi-year, scientifically rigorous policy evaluation mechanism incorporating long-term green benefits into performance assessments should be established. This will provide a stable and predictable institutional environment that is essential for INT’s profound integration into the green economic system.

Second, environmental regulation policies should be implemented that are characterized by differentiation and dynamic adjustment according to local conditions. Given the substitutive effect between ER and INT, policymakers should move beyond a one-size-fits-all approach towards more precise, flexible, and INT-tailored regulatory approaches. In regions with advanced INT development and superior industrial structures, the government should transition from high-intensity ER to incentive-based ER that stimulates innovation, thereby creating a conducive environment for INT-driven endogenous green growth. Conversely, in regions where INT is nascent and industrial structures are weaker, maintaining a relatively stringent ER as an external constraint is crucial to compel enterprises to phase out obsolete production capacities and adopt green technologies. Simultaneously, the intensity and focus of ER should undergo dynamic recalibration in response to the rapid evolution of INT [105].

Third, fostering the deep integration of INT with GI to construct a robust green innovation ecosystem underpinned by strong intellectual property rights (IPR) protection is essential. On the one hand, targeted incentives should be introduced, such as establishing dedicated INT + green technology R&D funds, providing expedited patent review channels, and tax benefits for relevant applications to provide comprehensive support for GI. On the other hand, stringent IPR laws that specifically address the GI domain must be enacted. This should be coupled with enhanced penalties for infringement and the establishment of a rapid-response mechanism for resolving IPR disputes. Collectively, these measures cultivate an external environment conducive to sustained green innovation.

Fourth, substantial investment in AI-supporting infrastructure is required, while cultivating demonstration zones and industrial clusters to foster the widespread application of AI. On one hand, governments must fully recognize the pivotal role of AI-driven industrial intelligence as a new engine for economic development, while strengthening technological integration, infrastructure development, and talent cultivation to advance AI research, application, and diffusion [105]. Constructing an AI-centric intelligent ecosystem that provides a robust underpinning for urban industrial digital transformation, networked collaboration, and intelligent upgrading. On the other hand, optimizing regional spatial planning is essential. Efforts should focus on establishing world-class integrated INT and green industry clusters centered on the AI industry. Such high-quality SES will amplify knowledge spillovers and resource allocation optimization effects, effectively transforming point-source AI technological breakthroughs into area-wide macro momentum to drive GTFP enhancement.

Fifth, the global community should deepen international cooperation in INT to collectively advance the Sustainable Development Goals. Within the context of globalization, the green application of INT should transcend national boundaries and become a focal point for international collaboration. Nations must enhance dialogue and communication to jointly formulate international norms for INT-enabled green development. Moreover, the application of INT and green technologies in developing countries should be accelerated through multilateral cooperation and targeted financial support to enable a collective response to global challenges, such as climate change. Furthermore, to balance innovation with governance, it is crucial for countries to implement prudential governance regarding INT development and deployment, ensuring that they remain consistently oriented towards enhancing human well-being. Policy experiences from OECD nations in aligning Trustworthy AI with the Sustainable Development Goals offer valuable references.

7.3. Limitations and Future Directions

Although this study systematically investigated the relationship among INT, ER, and GTFP, providing targeted policy recommendations for governments to leverage INT to enhance GTFP, the following limitations remain. First, the measurement of variables is constrained by data availability. Although the validity and rationality of the indicators are elaborated, the existing INT Indicator system primarily relies on the screening and integration of data from related industries, with insufficient dedicated INT data, potentially failing to fully capture its technological and industrial characteristics. Secondly, both GTFP and INT are composite indicators whose construction depends on underlying variables (e.g., R&D investment) susceptible to regional financing environments and policy influences, which may introduce estimation bias. While heterogeneity tests using instrumental variables and policy shocks were conducted, residual confounding factors may persist. Third, methodologically, the potential spatial spillover effects were not precisely measured using approaches like spatial econometric models. Consequently, policy transmission and technology diffusion could lead to parameter estimation bias, necessitating caution when interpreting specific coefficients. Finally, the conclusions are derived from the provincial data of China. Given the uniqueness of China’s ER, institutional context, and industrial structure, the generalizability of these findings requires careful consideration. Extrapolating to other countries, applying micro-level entities (e.g., specific industries or firms), or contexts with different types of ER may face challenges and must account for contextual dependence.

These limitations also delineate productive directions for future research. Firstly, refining measurement methodologies to develop indicators with enhanced targeted specificity and exogenous characteristics is critical for minimizing measurement errors inherent in indicator construction. Secondly, employing advanced econometric frameworks such as the Spatial Durbin Model (SDM) would allow for systematic quantification of spatial spillover effects and policy transmission mechanisms associated with INT. Thirdly, conducting comparative analyses across diverse national contexts, heterogeneous levels of entities, and varying types of ER is essential to elucidate the contextual nuances governing INT’s green-enabling effects.

Author Contributions

Conceptualization, S.X. and S.W.; methodology, S.X. and J.J.; software, S.X.; validation, S.X.; formal analysis, J.J.; investigation, J.J. and Y.Z.; resources, S.X. and J.J.; data curation, S.X. and Y.Z.; writing—original draft preparation, S.X., J.J., and Y.Z.; writing—review and editing, S.W. and S.X.; visualization, Y.Z.; supervision, S.W.; project administration, S.X. and S.W.; funding acquisition, S.X. and J.J. All authors have read and agreed to the published version of the manuscript.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The datasets used or analyzed in the current study are available from the corresponding author upon reasonable request.

Conflicts of Interest

The authors declare no conflicts of interest.

Footnotes

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Figures and Tables

Figure 1 The mechanism by which INT influences GTFP.

Figure 2 Average development trend of industrial intelligence in China from 2012 to 2023.

Figure 3 Average level of INT development in various regions from 2012 to 2023.

Figure 4 Average level of GTFP in various regions from 2012 to 2023.

Table 1

The indicator system of GTFP.

Primary Indicator	Secondary Indicator	Measurement Method	Data Source
Factor Input	Labor Input	Total number of employed persons at the end of the year in each region (10,000 persons)	CSY
	Capital Input	Fixed-asset stock with a depreciation rate calculated based on the perpetual inventory method (taking 2011 as the base period) (100 million RMB)	CSY
	Energy Input	Total energy consumption in each region (10,000 tons of standard coal)	CSYE
Desired Output	Actual Gross Domestic Product	Nominal GDP deflated using regional GDP deflators (taking 2011 as the base period) (100 million yuan)	CSY
Undesired Output	Pollution Emissions	Industrial sulfur dioxide emissions in each region (10,000 tons)	CISY
		Industrial wastewater discharge in each region (10,000 tons)	CISY
		Discharge of general industrial solid waste in each region (10,000 tons)	CISY

Note: China Statistical Yearbook (CSY), China Industrial Statistical Yearbook (CISY), and China Statistical Yearbook on Environment (CSYE). Some data were also obtained from the Statistical Yearbooks of Provinces/Municipalities.

Table 2

The indicator system of industrial intelligence.

Primary Indicator	Secondary Indicator	Measurement Method	Data Source	Weight
TechnologicalInnovation Environment	General Innovation Capacity	Number of patent applications authorized/Total population (10,000 persons)	CNIPACSY	0.31
	R&D Funding	Research and experimental development expenditures of industrial enterprises above the designated size (100 million RMB)	CSY
	AI Innovation Achievement	Number of AI patents/Total population (10,000 persons)	CNIPACSY
Industrial Development Foundation	Internet Coverage	Per capita Internet access ports	CISY	0.33
	Information Technology Infrastructure	Length of long-distance optical cable lines (10,000 km)/Urban area (10,000 square kilometers)	CISY
	INT Human Resource	Number of employees in urban units of information transmission, software, and information technology services (10,000 persons)/Total population (10,000 persons)	CPESYCSY
	INT Capital Input	Fixed-asset investment in information transmission, software, and information technology services (100 million RMB)/GDP (100 million RMB)	CSY
Industrial Output Level	Industrial Robot Installation Density	Number of industrial robots installed in various sub-sectors	IFR	0.36
	Quantity of INT Enterprise	Following Wang’s methodology [62], a Python-based (version 3.11.10) web crawler was employed to conduct fuzzy matching queries on the “business scope” field of the Tianyancha database, targeting INT-related keywords including “intelligent”, “machine learning”, “Internet of Things (IoT)”, and “big data”.	TYC
	Contribution of INT Enterprise	Total profits of computer, communication, and other electronic equipment manufacturing enterprises (100 million RMB)/GDP (100 million RMB)	CISYCSY
	INT Industry Business Revenue	Revenue of software and information technology services (100 million RMB)/Number of employees (10,000 persons)	CISYCSY

Note: China Statistical Yearbook (CSY), China Statistical Yearbook on Science and Technology (CSYST), China Population & Employment Statistical Yearbook (CPESY), China Industrial Statistical Yearbook (CISY), and Tianyancha database (TYC). Some data were also obtained from the Statistical Yearbooks of Provinces/Municipalities.

Table 3

Indicator system of ER.

Moderating Variable	Regulation Type	Measurement Method	Data Source
Environmental Regulation	Policy-Oriented	Word frequency of environmental governance in policy texts (times)/Total length of policy texts (characters)	Government website
	Tax-Adjusted	Total amount of environmental protection tax and resource tax (100 million RMB)/GDP (100 million RMB)	CSY
	Public-Supervision	Annual Baidu search index for “environmental pollution”	BAIDU
	Industrial-Governance	Year-end stock of enterprises related to green industries	CPPGD

Note: China Statistical Yearbook (CSY), Baidu database (BAIDU), China Public Policy and Green Development Database (CPPGD). Some data are obtained from the Statistical Yearbooks of Provinces/Municipalities.

Table 4

Descriptive statistics of the main variables.

Variable Type	Symbol	N	Mean	St. Dev.	Min	Max
Explained Variable	GTFP	360	0.6240	0.1528	0.2614	1.0853
Explanatory Variable	INT	360	0.1683	0.1031	0.0185	0.5645
Control Variables	Fin	360	1.5147	0.4425	0.6768	2.7592
	FDI	360	0.0027	0.0022	0.00001	0.0128
	GRP	360	10.9772	0.4435	9.8889	12.2075
	Indus	360	0.4119	0.0858	0.1449	0.5769
	Gov	360	0.2483	0.1015	0.1066	0.6430

Table 5

Benchmark regression results.

	Explained Variable
	GTFP
	(1)	(2)	(3)	(4)	(5)	(6)
INT	0.356 ***	0.376 ***	0.365 ***	0.328 ***	0.310 ***	0.305 ***
	(0.093)	(0.078)	(0.077)	(0.076)	(0.076)	(0.076)
Fin		−0.150 ***	−0.156 ***	−0.118 ***	−0.104 ***	−0.102 ***
		(0.013)	(0.013)	(0.016)	(0.017)	(0.017)
FDI			−5.427 ***	−6.724 ***	−6.296 ***	−6.059 ***
			(1.848)	(1.848)	(1.826)	(1.900)
GRP				0.102 ***	0.078 ***	0.068 *
				(0.028)	(0.029)	(0.036)
Indus					0.333 ***	0.341 ***
					(0.104)	(0.105)
Gov						−0.060
						(0.131)
Province FE	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
Observations	360	360	360	360	360	360
Adjusted R²	−0.079	0.237	0.255	0.283	0.304	0.302
F Statistic	14.706 ***	76.890 ***	55.367 ***	46.444 ***	40.312 ***	33.544 ***

Note: * p < 0.1; ** p < 0.05; *** p < 0.01. Figures in parentheses are standard errors.

Table 6

Robustness test regression results.

	Explained Variable
	NSE-EBM	GTFP
	(1)	(2)	(3)	(4)	(5)	(6)
AI_Tech		0.181 ***
		(0.037)
INT	0.287 ***		0.310 ***	0.273 ***	0.262 ***	0.305 ***
	(0.075)		(0.081)	(0.098)	(0.087)	(0.049)
Fin	−0.097 ***	−0.068 ***	−0.105 ***	−0.125 ***	−0.108 ***	−0.102 ***
	(0.017)	(0.018)	(0.018)	(0.020)	(0.018)	(0.017)
FDI	−6.028 ***	−5.354 ***	−6.008 ***	−6.267 ***	−6.901 ***	−6.059 ***
	(1.869)	(1.886)	(1.996)	(2.042)	(1.894)	(2.147)
GRP	0.071 **	0.076 **	0.046	0.108 **	0.017	0.068 **
	(0.036)	(0.036)	(0.041)	(0.044)	(0.042)	(0.033)
Indus	0.348 ***	0.383 ***	0.531 ***	0.238 *	1.160 ***	0.341 **
	(0.104)	(0.104)	(0.121)	(0.122)	(0.247)	(0.164)
Gov	−0.060	−0.143	−0.087	0.091	−0.008	−0.060
	(0.128)	(0.128)	(0.145)	(0.162)	(0.132)	(0.160)
Add Controls	NO	NO	NO	NO	YES	NO
Province FE	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
Observations	360	360	312	270	360	360
Adjusted R²	0.302	0.318	0.346	0.333	0.325	0.302
F Statistic	33.496 ***	35.570 ***	34.442 ***	29.564 ***	24.647 ***	33.544 ***

Note: * p < 0.1; ** p < 0.05; *** p < 0.01. Figures in parentheses are standard errors.

Table 7

Endogeneity test regression results.

	2SLS-Stage 1	2SLS-Stage 2	2SLS-Stage 1	2SLS-Stage 2
	(1)	(2)	(3)	(4)
IV1_Tel	0.067 ***
	(0.010)
IV2_Policy×Post			0.054 ***
			(0.005)
INT		0.799 ***		0.373 **
		(0.217)		(0.154)
Kleibergen-Paap LM	35.353 ***		57.837 ***
Cragg-Donald Wald F	44.387 ***		106.155 ***
Controls	YES	YES	YES	YES
Province FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	360	360	360	360
Adjusted R²	0.938	0.940	0.947	0.939
F Statistic	118.214 ***	124.102 ***	139.822 ***	121.032 ***

Note: * p < 0.1; ** p < 0.05; *** p < 0.01. Figures in parentheses are standard errors. The Kleibergen-Paap LM statistic is the heteroskedastic robust version, while the Cragg-Donald Wald F statistic is based on the assumption of homoscedasticity.

Table 8

Heterogeneity test regression results.

	IndDom	Non_IndDom	Low_IndStr	Med_IndStr	High_IndStr
	(1)	(2)	(3)	(4)	(5)
INT	−0.122	0.532 ***	−0.628 ***	0.303 ***	0.423 ***
	(0.153)	(0.131)	(0.145)	(0.101)	(0.158)
Controls	YES	YES	YES	YES	YES
Province FE	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES
Observations	120	240	120	132	108
Adjusted R²	0.693	0.401	0.702	0.407	0.206
F Statistic	37.152 ***	24.715 ***	51.148 ***	19.513 ***	8.784 ***

Note: * p < 0.1; ** p < 0.05; ***p < 0.01. Figures in parentheses are standard errors.

Table 9

Regression results of mediating mechanisms.

	Explained Variable
	GI	GTFP	SES	GTFP
	(1)	(2)	(3)	(4)
GI		0.152 ***
		(0.045)
SES				0.229 ***
				(0.062)
INT	0.534 ***	0.224 ***	0.478 ***	0.196 **
	(0.094)	(0.079)	(0.069)	(0.080)
Controls	YES	YES	YES	YES
Province FE	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
Observations	360	360	360	360
Adjusted R²	0.106	0.324	0.105	0.329
F Statistic	14.748 ***	31.333 ***	14.698 ***	31.907 ***

Note: * p < 0.1; ** p < 0.05; *** p < 0.01. Figures in parentheses are standard errors. Green Innovation Effect = 0.534 × 0.152/0.305 = 26.612%; Spatio-Economic Synergy Effect = 0.478 × 0.229/0.305 = 35.889%.

Table 10

Regression results of the moderating mechanism.

	Explained Variable
	GTFP	GI	GTFP	SES	GTFP
	(1)	(2)	(3)	(4)	(5)
GI			0.134 ***
			(0.044)
SES					0.193 ***
					(0.059)
INT	0.555 ***	1.043 ***	0.415 ***	0.782 ***	0.405 ***
	(0.102)	(0.131)	(0.111)	(0.097)	(0.111)
ER	0.198 ***	0.086 **	0.186 ***	0.066 **	0.185 ***
	(0.027)	(0.034)	(0.027)	(0.026)	(0.027)
INT×ER	−0.631 ***	−1.075 ***	−0.487 ***	−0.652 ***	−0.506 ***
	(0.160)	(0.204)	(0.164)	(0.151)	(0.162)
Controls	YES	YES	YES	YES	YES
Province FE	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES
Observations	360	360	360	360	360
Adjusted R²	0.403	0.178	0.419	0.151	0.421
F Statistic	36.308 ***	15.714 ***	34.187 ***	13.958 ***	34.460 ***

Note: * p < 0.1; ** p < 0.05; *** p < 0.01. Figures in parentheses are standard errors.

Table 11

Regression results of time-lagged effect test.

	Direct	Moderate	Low_ER	Med_ER	High_ER
	(1)	(2)	(3)	(4)	(5)
INT_lag1	0.250 ***	0.495 ***	0.721 ***	0.152	−0.410 ***
	(0.082)	(0.110)	(0.166)	(0.141)	(0.105)
ER		0.191 ***
		(0.028)
INT_lag1×ER		−0.590 ***
		(0.175)
Controls	YES	YES	YES	YES	YES
Province FE	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES
Observations	360	330	109	109	112
Adjusted R2	0.292	0.388	0.385	0.513	0.615
F Statistic	30.169 ***	31.957 ***	16.623 ***	24.663 ***	35.408 ***

Note: * p < 0.1; ** p < 0.05; *** p < 0.01. Figures in parentheses are standard errors.

References

1. Shinwari, R.; Wang, Y.; Gozgor, G.; Mousavi, M. Does FDI affect energy consumption in the belt and road initiative economies? The role of green technologies. Energy Econ.; 2024; 132, 107409. [DOI: https://dx.doi.org/10.1016/j.eneco.2024.107409]

2. United Nations Environment Programme, Emissions Gap Report 2019. Available online: https://digitallibrary.un.org/record/3894787?ln=en&v=pdf (accessed on 17 August 2025).

3. Tao, M.; Failler, P.; Goh, L.T.; Lau, W.Y.; Dong, H.; Xie, L. Quantify the effect of China’s emission trading scheme on low-carbon eco-efficiency: Evidence from China’s 283 cities. Mitig. Adaptat. Strat. Glob. Change; 2022; 27, 37. [DOI: https://dx.doi.org/10.1007/s11027-022-10015-8]

4. Guan, D.; Meng, J.; Reiner, D.M.; Zhang, N.; Shan, Y.; Mi, Z.; Shao, S.; Liu, Z.; Zhang, Q.; Davis, S.J. Structural decline in China’s CO₂ emissions through transitions in industry and energy systems. Nature Geosci.; 2018; 11, pp. 551-555. [DOI: https://dx.doi.org/10.1038/s41561-018-0161-1]

5. Johnson, P.C.; Laurell, C.; Ots, M.; Sandström, C. Digital innovation and the effects of artificial intelligence on firms’ research and development—Automation or augmentation, exploration or exploitation?. Technol. Forecast. Soc. Change; 2022; 179, 121136. [DOI: https://dx.doi.org/10.1016/j.techfore.2022.121636]

6. Acemoglu, D.; Restrepo, P. Robots and jobs: Evidence from US labor markets. J. Political Econ.; 2020; 128, pp. 2188-2244. [DOI: https://dx.doi.org/10.1086/705716]

7. Alsaleh, M.; Yang, Z. The evolution of information and communications technology in the fishery industry: The pathway for marine sustainability. Mar. Pollut. Bull.; 2023; 193, 115231. [DOI: https://dx.doi.org/10.1016/j.marpolbul.2023.115231]

8. Nchofoung, T.N. Asongu SA ICT for sustainable development: Global comparative evidence of globalisation thresholds. Telecommun Policy; 2022; 46, 102296. [DOI: https://dx.doi.org/10.1016/j.telpol.2021.102296]

9. Fu, T.; Qiu, Z.; Yang, X.; Li, Z. The impact of artificial intelligence on green technology cycles in China. Technol. Forecast. Soc. Change; 2024; 209, 123821. [DOI: https://dx.doi.org/10.1016/j.techfore.2024.123821]

10. Chen, S.; Golley, J. ‘Green’ productivity growth in China’s industrial economy. Energy Econ.; 2014; 44, pp. 89-98. [DOI: https://dx.doi.org/10.1016/j.eneco.2014.04.002]

11. Porter, M.E.; van der Linde, C. Toward a new conception of the environment-competitiveness relationship. J. Econ. Perspect.; 1995; 9, pp. 97-118. [DOI: https://dx.doi.org/10.1257/jep.9.4.97]

12. Lanoie, P.; Laurent-Lucchetti, J.; Johnstone, N.; Ambec, S. Environmental Policy, Innovation and Performance: New Insights on the Porter Hypothesis. J. Econ. Manag. Strategy; 2011; 20, pp. 803-842. [DOI: https://dx.doi.org/10.1111/j.1530-9134.2011.00301.x]

13. Li, H.; Chen, C.; Umair, M. Green Finance, Enterprise Energy Efficiency, and Green Total Factor Productivity: Evidence from China. Sustainability; 2023; 15, 11065. [DOI: https://dx.doi.org/10.3390/su151411065]

14. Lee, C.-C.; Lee, C.-C. How does green finance affect green total factor productivity? Evidence from China. Energy Econ.; 2022; 107, 105863. [DOI: https://dx.doi.org/10.1016/j.eneco.2022.105863]

15. De Mariz, F. Finance with a Purpose: FinTech, Development and Financial Inclusion in the Global Economy; World Scientific: Singapore, 2022; pp. 135-138. [DOI: https://dx.doi.org/10.1142/q0359]

16. Zhao, C. Is low-carbon energy technology a catalyst for driving green total factor productivity development? The case of China. J. Clean. Prod.; 2023; 428, 139507. [DOI: https://dx.doi.org/10.1016/j.jclepro.2023.139507]

17. Wang, H.; Cui, H.; Zhao, Q. Effect of green technology innovation on green total factor productivity in China: Evidence from spatial durbin model analysis. J. Clean. Prod.; 2021; 288, 125624. [DOI: https://dx.doi.org/10.1016/j.jclepro.2020.125624]

18. Du, K.; Li, J. Towards a green world: How do green technology innovations affect total-factor carbon productivity. Energy Policy; 2019; 131, pp. 240-250. [DOI: https://dx.doi.org/10.1016/j.enpol.2019.04.033]

19. You, X.; Li, Z.; Yi, Y. Carbon Constraints, Industrial Structure Upgrading, and Green Total Factor Productivity: An Empirical Study Based on the Yangtze River Economic Belt. J. Water Clim. Change; 2023; 14, pp. 3010-3026. [DOI: https://dx.doi.org/10.2166/wcc.2023.051]

20. Tian, X.; Zhang, H. Analysis of the Impact Factors of Industrial Structure Upgrading on Green Total Factor Productivity from the Perspective of Spatial Spillover Effects. Heliyon; 2024; 10, e28660. [DOI: https://dx.doi.org/10.1016/j.heliyon.2024.e28660]

21. He, Z.; Chen, Z.; Feng, X. Different types of industrial agglomeration and green total factor productivity in China: Do institutional and policy characteristics of cities make a difference?. Environ. Sci. Eur.; 2022; 34, 64. [DOI: https://dx.doi.org/10.1186/s12302-022-00645-9]

22. Liu, Y.; Yang, Y.; Li, H.; Zhong, K. Digital Economy Development, Industrial Structure Upgrading and Green Total Factor Productivity: Empirical Evidence from China’s Cities. Int. J. Environ. Res. Public Health; 2022; 19, 2414. [DOI: https://dx.doi.org/10.3390/ijerph19042414]

23. Xue, Y. Evaluation analysis on industrial green total factor productivity and energy transition policy in resource-based region. Energy Environ; 2022; 33, pp. 419-434. [DOI: https://dx.doi.org/10.1177/0958305X211005428]

24. Yang, Y.; Wei, X.; Wei, J.; Gao, X. Industrial Structure Upgrading, Green Total Factor Productivity and Carbon Emissions. Sustainability; 2022; 14, 1009. [DOI: https://dx.doi.org/10.3390/su14021009]

25. Braguinsky, S.; Ohyama, A.; Okazaki, T.; Syverson, C. Product Innovation, Product Diversification, and Firm Growth: Evidence from Japan’s Early Industrialization. Am. Econ. Rev.; 2022; 111, pp. 3795-3826. [DOI: https://dx.doi.org/10.1257/aer.20201656]

26. Yao, J.Q.; Zhang, K.P.; Guo, L.P.; Feng, X. How Does Artificial Intelligence Improve Enterprise Production Efficiency?—Based on the Perspective of Labor Skill Structure Adjustment. J. Manag. World; 2024; 40, pp. 101–116+133+117–122. [DOI: https://dx.doi.org/10.19744/j.cnki.11-1235/f.2024.0018]

27. Chen, J.; Liu, Y.H. Digital Intelligence Enables Operational Management Transformation: From Supply Chain to Supply Chain Ecosystem. J. Manag. World; 2021; 37, pp. 227–240+14. [DOI: https://dx.doi.org/10.19744/j.cnki.11-1235/f.2021.0180]

28. Nahar, S. Modeling the effects of artificial intelligence (AI)-based innovation on sustainable development goals (SDGs): Applying a system dynamics perspective in a cross-country setting. Technol. Forecast. Soc. Change; 2024; 201, 123203. [DOI: https://dx.doi.org/10.1016/j.techfore.2023.123203]

29. Lange, S.; Pohl, J.; Santarius, T. Digitalization and energy consumption. Does ICT reduce energy demand?. Ecol. Econ.; 2020; 176, 106760. [DOI: https://dx.doi.org/10.1016/j.ecolecon.2020.106760]

30. Tang, X.H.; Chi, Z.M. An Empirical Study on Industrial Intelligence Improving Industrial Green Development Efficiency. Economist; 2022; 2, pp. 43-52. [DOI: https://dx.doi.org/10.16158/j.cnki.51-1312/f.2022.02.006]

31. Wang, Z.; Zhang, T.; Ren, X.; Shi, Y. AI adoption rate and corporate green innovation efficiency: Evidence from Chinese energy companies. Energy Econ.; 2024; 132, 107499. [DOI: https://dx.doi.org/10.1016/j.eneco.2024.107499]

32. Li, Y.; Zhang, Y.; Pan, A.; Han, M.; Veglianti, E. Carbon emission reduction effects of industrial robot applications: Heterogeneity characteristics and influencing mechanisms. Technol. Soc.; 2022; 70, 102034. [DOI: https://dx.doi.org/10.1016/j.techsoc.2022.102034]

33. Balsalobre-Lorente, D.; Abbas, J.; He, C.; Pilař, L.; Shah, S.A.R. Tourism, urbanization and natural resources rents matter for environmental sustainability: The leading role of AI and ICT on sustainable development goals in the digital era. Resour. Policy; 2023; 82, 103445. [DOI: https://dx.doi.org/10.1016/j.resourpol.2023.103445]

34. Qian, Y.; Liu, J.; Shi, L.; Forrest, J.Y.-L.; Yang, Z. Can artificial intelligence improve green economic growth? Evidence from China. Environ. Sci. Pollut. Res.; 2023; 30, pp. 16418-16437. [DOI: https://dx.doi.org/10.1007/s11356-022-23320-1] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36184706]

35. Chen, Y.; Jin, S. Artificial Intelligence and Carbon Emissions in Manufacturing Firms: The Moderating Role of Green Innovation. Processes; 2023; 11, 2705. [DOI: https://dx.doi.org/10.3390/pr11092705]

36. Chen, P.; Gao, J.; Ji, Z.; Liang, H.; Peng, Y. Do Artificial Intelligence Applications Affect Carbon Emission Performance?—Evidence from Panel Data Analysis of Chinese Cities. Energies; 2022; 15, 5730. [DOI: https://dx.doi.org/10.3390/en15155730]

37. Vinuesa, R.; Azizpour, H.; Leite, I.; Balaam, M.; Dignum, V.; Domisch, S.; Felländer, A.; Langhans, S.D.; Tegmark, M.; Nerini, F.F. The role of artificial intelligence in achieving the Sustainable Development Goals. Nat. Commun.; 2020; 11, 233. [DOI: https://dx.doi.org/10.1038/s41467-019-14108-y]

38. Gan, V.J.L.; Lo, I.M.C.; Ma, J.; Tse, K.T.; Cheng, J.C.; Chan, C.-M. Simulation optimisation towards energy efficient green buildings: Current status and future trends. J. Clean. Prod.; 2020; 254, 120012. [DOI: https://dx.doi.org/10.1016/j.jclepro.2020.120012]

39. Krugman, P. Increasing Returns and Economic Geography. J. Political Econ.; 1990; 99, pp. 483-499. [DOI: https://dx.doi.org/10.1086/261763]

40. Autio, E.; Nambisan, S.; Thomas, L.D.W.; Wright, M. Digital affordances, spatial affordances, and the genesis of entrepreneurial ecosystems. Strateg. Entrep. J.; 2018; 12, pp. 72-95. [DOI: https://dx.doi.org/10.1002/sej.1266]

41. Yuan, B.; Xiang, Q. Environmental regulation, industrial innovation and green development of Chinese manufacturing: Based on an extended CDM model. J. Clean. Prod.; 2018; 176, pp. 895-908. [DOI: https://dx.doi.org/10.1016/j.jclepro.2017.12.034]

42. Qiu, R.; Han, L.; Xu, J.; Yin, W. Impact of environmental regulations on the green transition of China’s mariculture industry: Empirical test based on a dynamic panel model. Resour. Sci.; 2022; 44, pp. 1615-1629. [DOI: https://dx.doi.org/10.18402/resci.2022.08.07]

43. Horbach, J.; Rammer, C.; Rennings, K. Determinants of eco-innovations by type of environmental impact—The role of regulatory push/pull, technology push and market pull. Ecol. Econ.; 2012; 78, pp. 112-122. [DOI: https://dx.doi.org/10.1016/j.ecolecon.2012.04.005]

44. Acemoglu, D.; Autor, D.; Dorn, D.; Hanson, G.H.; Price, B. Return of the Solow Paradox? IT, Productivity, and Employment in US Manufacturing. Am. Econ. Rev.; 2014; 104, pp. 394-399. [DOI: https://dx.doi.org/10.1257/aer.104.5.394]

45. Zou, Z.D.; Tian, K.D.; Zhao, W.; Li, Z.Z.; Liu, Z.Q. Study on the relationship between PM2.5 concentration and urbanization in Central China. World Reg. Stud.; 2024; 33, pp. 178-188. Available online: https://sjdlyj.ecnu.edu.cn/EN/10.3969/j.issn.1004-9479.2024.01.20220112 (accessed on 17 August 2025).

46. Wen, Z.L.; Ye, B.J. Mediating Effect Analysis: Methods and Model Development. Adv. Psychol. Sci.; 2014; 22, pp. 731-745. [DOI: https://dx.doi.org/10.3724/SP.J.1042.2014.00731]

47. Zhou, J.Q.; Chen, D.; Xia, N.X. Artificial Intelligence, Industrial Structure Optimization and Green Development Efficiency—Theoretical Analysis and Empirical Evidence. Mod. Financ. Econ.-J. Tianjin Univ. Financ. Econ.; 2023; 43, pp. 96-113. [DOI: https://dx.doi.org/10.19559/j.cnki.12-1387.2023.04.006]

48. Lin, Z.H.A.O.; Xiaotong, G.A.O.; Yanxu, L.I.U.; Zenglin, H.A.N. Analysis on the Evolution Characteristics of Spatial Correlation Network Structure of Inclusive Green Efficiency in China. Econ. Geogr.; 2021; 41, pp. 69–78+90.

49. Gong, B.-H. Finite Sample Evidence on the Performance of Stochastic Frontier and Data Envelopment Models in the Estimation of Firm-Specific Technical Efficiency Using Panel Data. Ph.D. Thesis; Rice University: Houston, TX, USA, 1987.

50. Charnes, A.; Cooper, W.W.; Rhodes, E. Measuring the efficiency of decision making units. Eur. J. Oper. Res.; 1978; 2, pp. 429-444. [DOI: https://dx.doi.org/10.1016/0377-2217(78)90138-8]

51. Tone, K.; Tsutsui, M. An epsilon-based measure of efficiency in DEA—A third pole of technical efficiency. Eur. J. Oper. Res.; 2010; 207, pp. 1554-1563. [DOI: https://dx.doi.org/10.1016/j.ejor.2010.07.014]

52. Cheng, C.; Yu, X.; Hu, H.; Su, Z.; Zhang, S. Measurement of China’s Green Total Factor Productivity Introducing Human Capital Composition. Int. J. Environ. Res. Public Health; 2022; 19, 13563. [DOI: https://dx.doi.org/10.3390/ijerph192013563]

53. Zhang, H.; Dong, Y. Measurement and Spatial Correlations of Green Total Factor Productivities of Chinese Provinces. Sustainability; 2022; 14, 5071. [DOI: https://dx.doi.org/10.3390/su14095071]

54. Zhou, R.; Zhang, Y. Measurement of Urban Green Total Factor Productivity and Analysis of Its Temporal and Spatial Evolution in China. Sustainability; 2023; 15, 9435. [DOI: https://dx.doi.org/10.3390/su15129435]

55. Wang, F.; Wong, W.-K.; Ortiz, G.G.R.; Al Shraah, A.; Mabrouk, F.; Li, J.; Li, Z. Economic analysis of sustainable exports value addition through natural resource management and artificial intelligence. Resour. Policy; 2023; 82, 103541. [DOI: https://dx.doi.org/10.1016/j.resourpol.2023.103541]

56. Borland, J.; Coelli, M. Are Robots Taking Our Jobs?. Aust. Econ. Rev.; 2017; 50, pp. 377-397. [DOI: https://dx.doi.org/10.1111/1467-8462.12245]

57. Benzidia, S.; Makaoui, N.; Bentahar, O. The impact of big data analytics and artificial intelligence on green supply chain process integration and hospital environmental performance. Technol. Forecast. Soc. Change; 2021; 165, 120557. [DOI: https://dx.doi.org/10.1016/j.techfore.2020.120557]

58. Guliyev, H. Artificial intelligence and unemployment in high-tech developed countries: New insights from dynamic panel data model. Res. Glob.; 2023; 7, 100140. [DOI: https://dx.doi.org/10.1016/j.resglo.2023.100140]

59. Feng, C.; Ye, X.; Li, J.; Yang, J. How does artificial intelligence affect the transformation of China’s green economic growth? An analysis from internal-structure perspective. J. Environ. Manag.; 2024; 351, 119923. [DOI: https://dx.doi.org/10.1016/j.jenvman.2023.119923]

60. Chen, F.X. Research Progress on Measurement Methods of Artificial Intelligence Development Level. Econ. Perspect.; 2022; 02, pp. 142-158. Available online: https://kns.cnki.net/kcms2/article/abstract?v=mdHrqfOh4rGyZ_37yNEAaFyNd0Vq9huQKojUBl3gyWUHbj2I6Pl-VBIYsUpzOfb-yIARpuWTuDzyFYj92u_6hEBbsdulx1T5LRXZLHclt52XZ5fpVdboR_9LAmGuxzERyxiazQmOnTGU8UOSTajMGAPzr5m8rSz1jO_RHVRMXBVkUei9cZASrP96Fh15vnNqK9EiTXcitWw=&uniplatform=NZKPT&language=CHS (accessed on 5 May 2025).

61. Ma, G.W.; Zhong, Y.T.; Zhong, J. Construction and Empirical Measurement of China’s Artificial Intelligence Development Evaluation Index System. Sci. Technol. Manag. Res.; 2023; 43, pp. 55-61.

62. Wang, L.H.; Jiang, H.; Dong, Z.Q. Will Industrial Intelligence Reshape the Geographical Pattern of Enterprises?. China Ind. Econ.; 2022; 2, pp. 137-155. [DOI: https://dx.doi.org/10.19581/j.cnki.ciejournal.2022.02.008]

63. Song, D.Y.; Zhu, W.B.; Ding, H. Can Enterprise Digitalization Promote Green Technology Innovation? An Investigation Based on Listed Companies in Heavy Pollution Industries. J. Financ. Econ.; 2022; 48, pp. 34-48. [DOI: https://dx.doi.org/10.16538/j.cnki.jfe.20211218.304]

64. Liu, B.; Lu, J.R.; Ju, T. Formalism or Substantialism: A Study on Green Innovation under the Soft Regulation of ESG Ratings. Nankai Bus. Rev.; 2023; 26, pp. 16-28.

65. Ouyang, X.L.; Zhang, J.H.; Du, G. Environmental Regulation and Urban Green Technology Innovation: Impact Mechanism and Spatial Effect. Chin. J. Manag. Sci.; 2022; 30, pp. 141-151. [DOI: https://dx.doi.org/10.16381/j.cnki.issn1003-207x.2022.0642]

66. Liu, C.; Pan, H.F.; Li, P.; Feng, Y.X. Research on the Impact and Mechanism of Digital Transformation on Green Innovation Efficiency of Manufacturing Enterprises. China Soft Sci.; 2023; 04, pp. 121-129. Available online: https://kns.cnki.net/kcms2/article/abstract?v=QUtgg5W7F1_7J_UnuV_LseSwMYqJaE0ePinrOHNm_K-LJMzahl5BN8WJIiisudHhKTAZeArSeOErTJOYYF3AL1AKM_DhzpEsMq-YUrFOXqTfmG00Zjn7SjLw1dkhujHYLBndECMP5bpwS3NIwwBhNU55jVW5wlaxy7O3VOjIUmeZX4cxZ1SoPIusp7JOOS-J&uniplatform=NZKPT&language=CHS (accessed on 7 May 2025).

67. Shu, L.M.; Liao, J.H.; Xie, Z. Green Credit Policy and Enterprise Green Innovation: Empirical Evidence from the Perspective of Green Industries. Financ. Econ. Res.; 2023; 38, pp. 144-160.

68. Tao, F.; Zhao, J.Y.; Zhou, H. Has Environmental Regulation Achieved “Increment and Quality Improvement” of Green Technology Innovation? Evidence from the Environmental Protection Target Responsibility System. China Ind. Econ.; 2021; 02, pp. 136-154. [DOI: https://dx.doi.org/10.19581/j.cnki.ciejournal.2021.02.016]

69. Zhang, J.; Gao, D.B.; Xia, Y.L. Can Patents Promote China’s Economic Growth? An Explanation from the Perspective of China’s Patent Subsidy Policy. China Ind. Econ.; 2016; 01, pp. 83-98. [DOI: https://dx.doi.org/10.19581/j.cnki.ciejournal.2016.01.006]

70. Ciccone, A.; Hall, R.E. Productivity and the Density of Economic Activity. Am. Econ. Rev.; 1996; 86, pp. 57-70. Available online: http://www.jstor.org/stable/2118255 (accessed on 17 August 2025).

71. Yao, S.; Zhang, X.; Zheng, W. On transportation, economic agglomeration, and CO₂ emissions in China, 2003–2017. Environ. Sci. Pollut. Res.; 2023; 30, pp. 40987-41001. [DOI: https://dx.doi.org/10.1007/s11356-022-25101-2]

72. Chen, Z.; Kahn, M.E.; Liu, Y.; Wang, Z. The consequences of spatially differentiated water pollution regulation in China. J. Environ. Econ. Manag.; 2018; 88, pp. 468-485. [DOI: https://dx.doi.org/10.1016/j.jeem.2018.01.010]

73. Dai, X.; Hua, X.Y.; Lin, J.G. Environmental Regulation, Green Total Factor Productivity and Economic Growth—Economic Logic of Harmonious Coexistence between Man and Nature. Econ. Rev. J.; 2025; 3, pp. 66-86. [DOI: https://dx.doi.org/10.16528/j.cnki.22-1054/f.202503066]

74. Xiong, L.; Yan, S.; Yang, M. Financial Development, Environmental Regulation and Industrial Green Technology Innovation—A Study from the Perspective of Biased Endogenous Growth. China Ind. Econ.; 2023; 12, pp. 99-116. [DOI: https://dx.doi.org/10.19581/j.cnki.ciejournal.2023.12.011]

75. Chen, S.Y.; Chen, D.K. Haze Pollution, Government Governance and High-Quality Economic Development. Econ. Res. J.; 2018; 53, pp. 20-34.

76. Shukai, C.; Bixia, H.; Meng, G. Research on spatial-temporal heterogeneity of driving factors of green innovation efficiency in Yangtze River Delta urban agglomeration—Empirical test based on the Geographically Weighted Regression model. Front. Energy Res.; 2024; 12, 1308494. [DOI: https://dx.doi.org/10.3389/fenrg.2024.1308494]

77. Farmanesh, P.; Solati Dehkordi, N.; Vehbi, A.; Chavali, K. Artificial Intelligence and Green Innovation in Small and Medium-Sized Enterprises and Competitive-Advantage Drive Toward Achieving Sustainable Development Goals. Sustainability; 2025; 17, 2162. [DOI: https://dx.doi.org/10.3390/su17052162]

78. Wang, B.; Wang, C. Green Finance and Technological Innovation in Heavily Polluting Enterprises: Evidence from China. Int. J. Environ. Res. Public Health; 2023; 20, 3333. [DOI: https://dx.doi.org/10.3390/ijerph20043333]

79. Li, K.; Lin, B. Economic growth model, structural transformation, and green productivity in China. Appl. Energy; 2017; 187, pp. 489-500. [DOI: https://dx.doi.org/10.1016/j.apenergy.2016.11.075]

80. Zhang, J.X.; Lu, G.Y.; Skitmore, M.; Ballesteros-Perez, P. A critical review of the current research mainstreams and the influencing factors of green total factor productivity. Environ. Sci. Pollut. Res.; 2021; 28, pp. 35392-35405. [DOI: https://dx.doi.org/10.1007/s11356-021-14467-4]

81. You, D.; Zhang, Y.; Yuan, B. Environmental regulation and firm eco-innovation: Evidence of moderating effects of fiscal decentralization and political competition from listed Chinese industrial companies. J. Clean. Prod.; 2019; 207, pp. 1072-1083. [DOI: https://dx.doi.org/10.1016/j.jclepro.2018.10.106]

82. Han, Y.; Wang, Q.; Li, Y. Does Financial Resource Misallocation Inhibit the Improvement of Green Development Efficiency? Evidence from China. Sustainability; 2023; 15, 4466. [DOI: https://dx.doi.org/10.3390/su15054466]

83. Wang, Y.; Liao, M.; Xu, L.; Malik, A. Arunima Malik, The impact of foreign direct investment on China’s carbon emissions through energy intensity and emissions trading system. Energy Econ.; 2021; 97, 105212. [DOI: https://dx.doi.org/10.1016/j.eneco.2021.105212]

84. Huang, H.P.; Zhou, G.M.; Li, G.M. Research on the Impact Mechanism of Industrial Digitalization on Green Total Factor Productivity—Also on the Threshold Effect of Environmental Regulation. China Environ. Sci.; 2025; 45, pp. 1713-1730. [DOI: https://dx.doi.org/10.19674/j.cnki.issn1000-6923.20241206.002]

85. Liu, Y.; Wang, S. Heterogeneity and Improvement Preference of Green Total Factor Productivity in Industrial Sectors Considering Undesirable Environmental Outputs. China Environ. Sci.; 2023; 43, pp. 6183-6193. [DOI: https://dx.doi.org/10.19674/j.cnki.issn1000-6923.20230926.001]

86. Wang, Z.; Li, X.; Xue, X.; Liu, Y. More government subsidies, more green innovation? The evidence from Chinese new energy vehicle enterprises. Renew. Energy; 2022; 197, pp. 11-21. [DOI: https://dx.doi.org/10.1016/j.renene.2022.07.086]

87. Liu, Y.; Nor, R.M.; Ishak, M.K.; Li, X. Spatial and Temporal Analysis of China’s Environmental Regulatory Framework: A Multidimensional Assessment. Pol. J. Environ. Stud.; 2025; 34, pp. 2297-2306. [DOI: https://dx.doi.org/10.15244/pjoes/187601]

88. Chun, Y.; Hwang, J. The Nexus of Artificial Intelligence and Green Innovation: A Cross-Density Analysis at the Country Level. J. Knowl. Econ.; 2025; 16, pp. 1688-1716. [DOI: https://dx.doi.org/10.1007/s13132-024-02076-8]

89. Yin, H.-T.; Wen, J.; Chang, C.-P. Going green with artificial intelligence: The path of technological change towards the renewable energy transition. Oeconomia Copernic.; 2023; 14, pp. 1059-1095. [DOI: https://dx.doi.org/10.24136/oc.2023.032]

90. Liu, M.F.; Cheng, S.J. Did Carbon Emission Trading Promote the Optimization and Upgrading of Regional Industrial Structure?. Manag. Rev.; 2022; 34, pp. 33-46. [DOI: https://dx.doi.org/10.14120/j.cnki.cn11-5057/f.2022.07.015]

91. Peng, D.Y.; Zhang, J. Research on the Impact of Environmental Regulation on China’s Total Factor Energy Efficiency—An Empirical Test Based on Provincial Panel Data. J. Ind. Technol. Econ.; 2019; 38, pp. 59-67.

92. Guo, S.; Zhang, Z. Green credit policy and total factor productivity: Evidence from Chinese listed companies. Energy Econ.; 2023; 128, 107115. [DOI: https://dx.doi.org/10.1016/j.eneco.2023.107115]

93. Li, J.; Huang, D.; Wu, X. The Impact of China’s Carbon Emission Trading Policy on Green Total Factor Productivity—Influence Analysis Based on Super-EBM and Multiple Mediators. Pol. J. Environ. Stud.; 2022; 31, pp. 5107-5123. [DOI: https://dx.doi.org/10.15244/pjoes/151538]

94. Lee, C.-C.; Zeng, M.; Wang, C. Environmental Regulation, Innovation Capability, and Green Total Factor Productivity: New Evidence from China. Environ. Sci. Pollut. Res.; 2022; 29, pp. 39384-39399. [DOI: https://dx.doi.org/10.1007/s11356-021-18388-0] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/35098462]

95. Zhang, J.; Li, F.; Ding, X. Will green finance promote green development: Based on the threshold effect of R&D investment. Environ. Sci. Pollut. Res.; 2022; 29, pp. 60232-60243. [DOI: https://dx.doi.org/10.1007/s11356-022-20161-w] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/35419686]

96. Meng, F.; Xu, Y.; Zhao, G. Environmental regulations, green innovation and intelligent upgrading of manufacturing enterprises: Evidence from China. Sci. Rep.; 2020; 10, 14485. [DOI: https://dx.doi.org/10.1038/s41598-020-71423-x] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/32879350]

97. Zhu, L.; Li, X.; Huang, Y.; Liu, F.; Yang, C.; Li, D.; Bai, H. Digital Technology and Green Development in Manufacturing: Evidence from China and 20 Other Asian Countries. Sustainability; 2023; 15, 12841. [DOI: https://dx.doi.org/10.3390/su151712841]

98. Ouyang, J.Q.; Wei, D.Q.; Wang, Y.M. The Impact of Artificial Intelligence on New Productivity: Based on the Policy Effect of New-Generation Artificial Intelligence Innovation and Development Pilot Zones. Soft Sci.; 2025; 39, pp. 28-36. [DOI: https://dx.doi.org/10.13956/j.ss.1001-8409.2025.03.04]

99. Zhao, C.; Wang, L. Artificial Intelligence and Enterprise Green Innovation: Evidence from a Quasi-Natural Experiment in China. Sustainability; 2025; 17, 2455. [DOI: https://dx.doi.org/10.3390/su17062455]

100. Fu, H.; Rasiah, R. Fostering Inclusive Green Growth in Chinese Cities: Investigating the Role of Artificial Intelligence. Sustainability; 2024; 16, 9809. [DOI: https://dx.doi.org/10.3390/su16229809]

101. Wang, K.; Zhao, B.; Fan, T.; Zhang, J. Economic Growth Targets and Carbon Emissions: Evidence from China. Int. J. Environ. Res. Public Health; 2022; 19, 8053. [DOI: https://dx.doi.org/10.3390/ijerph19138053]

102. Zhou, J.Q.; Chen, D.; Xia, N.X. Mechanism and Empowering Effect of Artificial Intelligence on Green Economic Growth: From the Perspective of Industrial Structure Optimization. Sci. Technol. Prog. Policy; 2023; 40, pp. 45-55. [DOI: https://dx.doi.org/10.6049/kjjbydc.2022070343]

103. Shan, C.X.; Zhong, W.Z.; Geng, Z.Z.; Zhou, M.X. Research on the Impact of Environmental Regulation and Industry Heterogeneity on Technological Innovation in Industrial Sectors. Econ. Probl.; 2019; 12, pp. 60-67. [DOI: https://dx.doi.org/10.16011/j.cnki.jjwt.2019.12.010]

104. Li, Y.; Li, S.; Zhang, W. The Influence Study on Environmental Regulation and Green Total Factor Productivity of China’s Manufacturing Industry. Discret. Dyn. Nat. Soc.; 2021; 2021, 5580414. [DOI: https://dx.doi.org/10.1155/2021/5580414]

105. Ding, W.; Hu, P. Impact of AI development on green total factor productivity. Sci. Rep.; 2025; 15, 22906. [DOI: https://dx.doi.org/10.1038/s41598-025-06461-4] [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/40594980]

© 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.

How Does Industrial Intelligence Enhance Green Total Factor Productivity in China? The Substitution Effect of Environmental Regulation

Content area

Abstract

Full text

1. Introduction

2. Literature Review and Hypotheses Development

2.1. Direct Impact of INT on GTFP

2.2. Indirect Pathways of INT’s Impact on GTFP

2.3. Potential Moderating Role of Environmental Regulation

2.4. Time-Lagged Effect of INT on GTFP

3. Methodology and Data

3.1. Model Settings

3.1.1. Benchmark Model

3.1.2. Mediating Effect Model

3.1.3. Moderating Effect Model

3.2. Variable Measures

3.2.1. Explained Variable

3.2.2. Explanatory Variable

3.2.3. Mediating Variables

3.2.4. Moderating Variable

3.2.5. Control Variables

3.3. Data Sources and Descriptive Statistics

4. Empirical Results and Analysis

4.1. Benchmark Regression

4.2. Robustness Tests

4.2.1. Replace the Explained Variable

4.2.2. Replace the Explanatory Variable

4.2.3. Exclude Special Samples

4.2.4. Adjust the Sample Period

4.2.5. Incorporate Additional Control Variables

4.2.6. Employ the Driscoll-Kraay Standard Errors

4.3. Endogeneity Tests

4.4. Heterogeneity Tests

5. Mechanism Analysis

5.1. The Mediating Effect of Green Innovation and Spatio-Economic Synergy

5.2. The Moderating Effect of Environmental Regulation

6. Further Analysis: Time-Lagged Effect

7. Conclusions, Discussions, and Policy Implications

7.1. Conclusions and Discussions

7.2. Policy Implications

7.3. Limitations and Future Directions