Does Intelligent Manufacturing Contribute to the Enhancement of Carbon Emission Performance? Evidence from Total Factor Carbon Emission Performance

Abstract

The deep integration of intelligent technology and the manufacturing industry is a crucial driving force for promoting green and low-carbon development, which is a key strategy for achieving sustainable development. Using panel data from 30 provinces in mainland China from 2010 to 2022, this study measures the level of intelligent development and the total factor carbon emission performance (TFCEP). Additionally, a mediating effect model is constructed to explore the impact of intelligent manufacturing (IM) on carbon emission performance (CEP) and its underlying mechanisms. The findings reveal that (1) the intellectualization of the manufacturing industry significantly enhances CEP, a conclusion that remains robust under various tests; (2) the impact of IM on CEP varies by regional geographical locations, the degree of economic agglomeration (EA), and whether the province is a low-carbon pilot area; and (3) the mechanism analysis indicates that IM improves CEP by promoting EA. Given that China is the world’s largest manufacturing country and the largest carbon emitter, analyzing the impact of its IM on CEP provides valuable theoretical insights and practical experiences for China and other manufacturing countries aiming to achieve a win–win situation of sustainable economic development and environmental improvement.

1. Introduction

In the ever-evolving global industrial landscape, manufacturing remains the cornerstone of economic growth, technological innovation, and social progress. According to the “World Open Report 2022” statistical data, the manufacturing industry’s added value in the United States accounted for 10.7% of GDP and 15.3% of the global total in 2021. As the core of the European manufacturing hub, Germany’s manufacturing value added has consistently represented 20–23% of GDP since 1995 [1], constituting 4.7% of the global total in 2021. Japan and South Korea’s manufacturing sectors contributed approximately 7.8% and 3.0%, respectively, to the global manufacturing value, underscoring their critical roles in the Asian supply chain system. China, as the world’s largest manufacturing country, accounted for 29.8% of the global manufacturing value added in 2021 [2]. The current convergence of manufacturing technology with the new generation of information technology is driving a paradigm shift towards digitization and intelligence, often referred to as the Fourth Industrial Revolution [3]. All countries are seizing the opportunity to actively participate in this revolution to promote the intelligent development of the manufacturing industry. IM initiatives such as the American Smart Manufacturing Leadership Alliance promote the wide application of IM [4]; the Industry 4.0 strategy proposed by Germany aims to use advanced information technology to facilitate the transformation and upgrading of the manufacturing industry to intelligence [5]. Japan has formulated the “Society 5.0” strategy [6]; South Korea launched the “Manufacturing Innovation 3.0” plan [7]; and China has formulated the “Made in China 2025” manufacturing power strategy, which will promote IM as its main direction [8]. The deep integration of digital, networked, and intelligent technologies with the manufacturing industry is driving the new industrial revolution. As the core engine of this revolution, IM is reshaping the development trajectory, technological systems, and industrial forms of the manufacturing sector, propelling it into a new stage of development [9].

Excessive energy consumption and greenhouse gas emissions have led to rising global temperatures [10] and frequent extreme weather events, posing serious challenges to humanity and the earth’s ecosystem [11]. Reducing carbon emissions (CEs) has become a global consensus. According to the International Energy Agency, over 30% of energy-related CEs in 2021 originated from manufacturing [12], making it a top priority for emission reduction and climate change mitigation. Within the context of the Fourth Industrial Revolution, IM is viewed as a key measure for establishing competitive advantages in the manufacturing sectors of major countries [9]. The widespread application of intelligent elements will inevitably have profound impacts on the economic, social, and ecological environments, including carbon performance [13]. Therefore, studying the environmental benefits of IM, particularly CEP, is of great practical significance within the contexts of carbon neutrality and IM.

Since 2010, China has held the position of possessing the world’s largest manufacturing industry by scale for 14 consecutive years [14]. In 2021, the added value of China’s manufacturing industry accounted for 27.4% of GDP [15], making it a pillar of economic development and a major driver of economic growth. However, the traditional manufacturing industry’s development has exerted significant negative impacts on the environment [16]. To balance industrial development with environmental protection, China implemented the “Made in China 2025” strategy to address environmental pollution and energy inefficiency in traditional manufacturing [17]. To further this green development policy, China formulated the “Intelligent Manufacturing Development Plan (2016–2020)”, outlining a two-step strategy and ten key tasks to promote IM [18] to reduce energy consumption and achieve green and low-carbon development. In recent years, China has continuously promoted the adjustment of the industrial structure (IS) and energy structure, vigorously developed renewable energy, and strived to take into account the simultaneous development of economic development and green transformation. It has achieved remarkable results in improving ecological environment governance and reducing pollutant emissions [17].

Compared to developed countries, developing countries, which are undergoing rapid industrialization and urbanization, often exhibit lower energy efficiency [19], and play a more crucial role in carbon emission reduction and climate change mitigation. As the world’s largest carbon emitter, accounting for about 30% of global CEs [20], China bears significant international responsibilities for carbon reduction. In response to climate change and the “Paris Agreement”, the Chinese government has set various carbon emission reduction targets. The “Made in China 2025” initiative aims to reduce energy consumption per unit of industrial added value by 34% by 2025 compared to 2015 [21]. Furthermore, at the 75th United Nations General Assembly, China committed to achieving carbon peaking by 2030 and carbon neutrality by 2060 [22,23]. In the Fourth Industrial Revolution, China has become a pioneer and leader in green modernization [24]. During COP28, China highlighted its role as a major supplier of wind power, photovoltaic equipment, and power batteries, which has significantly reduced the cost of global renewable energy deployment and assisted developing countries in accessing clean, reliable energy [25]. Through South–South cooperation and initiatives such as the green “Belt and Road”, China provides clean energy technology and equipment to developing countries and promotes intelligent transportation programs [26].

Given China’s status as the largest manufacturing country and carbon emitter, this paper examines the impact of its IM on CEP, offering significant theoretical and practical insights for achieving the dual carbon goals, addressing climate change, and fostering sustainable development globally. This study utilizes panel data from 30 provinces (including autonomous regions and municipalities directly under the central government) in mainland China from 2010 to 2022 to calculate the TFCEP, employing the entropy weight method to measure intelligence levels and EA. The analysis explores how the intelligence of the manufacturing industry promotes improvements in CEP. The innovations of this paper are as follows. Firstly, in terms of research subjects, this paper considers the agglomeration effect of intelligent development, recognizing that agglomeration inevitably affects CEP. It selects EA as an intermediary variable to explore the path mechanism of how IM influences CEP, concluding that EA plays a partial mediating role in this impact, thereby expanding the existing research. Secondly, building on previous studies, this paper uses the network DEA–Malmquist index model to calculate the TFCEP, thereby opening the production “black box” and enriching the current research.

The rest of this paper is structured as follows. Section 2 introduces the existing research related to IM, EA, and CEP. Section 3 provides a theoretical analysis and research hypotheses. Section 4 details the research design. Section 5 presents the empirical results, robustness tests, heterogeneity analysis, and impact mechanism tests. Section 6 discusses the research findings. Section 7 summarizes the main conclusions, the limitations of this study, and future research directions.

2. Literature Review

2.1. Research on IM

IM represents a significant trend in the global manufacturing industry, profoundly impacting societal and economic development and becoming a focal point for scholars worldwide [27]. The existing research on IM primarily addresses its definition, development process, and evaluation indices. IM integrates modern information technologies such as 5G, digitization, networking, and artificial intelligence (AI) into the manufacturing sector [28]. Ji (2021) suggests that IM extends and expands the concept of smart manufacturing (SM), promoting the intelligentization of manufacturing production methods [29]. The terms “SM” and “IM” are often used interchangeably, despite subtle differences. The development of the IM industry can be summarized into three stages: digital manufacturing, networked manufacturing, and IM [9]. The new generation of IM is poised to drive the future development of the manufacturing industry [30]. An accurate assessment of the current level of IM requires precise data, but there is no unified measurement standard in academia. Most scholars construct comprehensive index systems for research. Yang (2024), for instance, developed a measurement index system for IM from three dimensions: intelligent conditions, intelligent applications, and intelligent benefits [31]. Li and Liang (2023) and Liu et al. (2021) created index systems measuring intelligence across basic input, production application, and market efficiency [32,33]. Wang and Meng (2023) developed an intelligence index based on key technology, production mode, and industrial competitiveness [34]. Yue and Gu (2023) measured intelligence by the proportion of enterprises using intelligent keywords annually [35], while Lv et al. (2022) used the number of robots per 10,000 people in the industrial sector as a measure of IM [20].

2.2. Research on IM and CEP

The traditional manufacturing industry, a significant part of the industrial sector, accounts for one-third of China’s total CEs [36]. The application of intelligent technology enhances energy efficiency, reduces resource waste, and supports low-carbon development [37]. CEP is a crucial factor for achieving economic growth and low-carbon development [38]. The existing literature presents two primary methods for measuring CEP: carbon emission intensity (CEI) as a reverse index or single-factor performance [39,40] and the TFCEP from the perspective of input and output [41,42,43]. Factors influencing CEP include urbanization [43], urban logistics [44], transportation infrastructure [45], digital economy [41], Internet development [42], industrial agglomeration [46], and so on. In terms of IM, most scholars focus on the impact of AI on CEP. Liang et al. (2022) assessed the production performance of China’s AI-driven low-carbon manufacturing using the interactive three-stage network DEA model [36]. Chen et al. (2022) used the Bartik method to quantify data from Chinese manufacturing enterprises and robots, demonstrating that AI significantly reduces CEI [40]. Chen and Jin (2023) further proved that AI positively impacts the CEP of manufacturing enterprises, with green innovation strengthening this effect [37].

2.3. Research on EA

Advanced manufacturing technology fosters EA through information and resource integration [47]. Regional integration enhances the role of EA in energy conservation and emission reduction [48]. EA, characterized by the concentration of economic activities in specific areas, promotes green and low-carbon development through positive externalities such as economies of scale and knowledge spillovers [49]. Various methods measure EA, including economic density [50], output density [48], employment density [51], population density [52], night lighting data [53], etc. Wang et al. (2022) think that EA is a complex and multi-dimensional agglomeration state, in which a single index has difficulty measuring the degree of agglomeration accurately, and proposed a comprehensive index system based on factor agglomeration, industrial agglomeration, and urban agglomeration [49].

Scholars have studied the relationship between EA and CE from different dimensions. Qin and Wu (2015) found that CEI initially increases and then decreases with agglomeration [54]. Yao et al. (2024) showed that a higher EA positively affects carbon productivity, with impact paths varying by the degree of agglomeration [50]. Yu et al. (2022) used spatial econometric models to examine the nonlinear effects of EA on carbon intensity across urban agglomeration development stages, concluding that excessive agglomeration increases regional carbon intensity [55]. Wang et al. (2022) argued that early agglomeration stages cause environmental pollution due to overinvestment, while high-level agglomeration reduces marginal CEs through production scale and specialization [49]. Few studies directly explore the relationship between IM and EA, with most examining industrial agglomeration as an intermediary variable [56,57], industrial intelligence and agglomeration [47], and AI and industrial agglomeration [58].

2.4. Literature Summary

The current literature on IM, CEP, and EA primarily focuses on individual aspects or their interrelationships. These studies have been categorized and summarized, as shown in

Table 1. Summary of the related studies.

Field	Content		Example	Year	Contribution
Research on IM	Implication		Ji [29]	2021	The process of promoting the intelligent production mode of the manufacturing industry.
	Development process		Ji et al. [9]	2018	Digital manufacturing stage, networked manufacturing stage, IM stage.
	Evaluation index		Yang [31]	2024	It is constructed from three dimensions: intelligent conditions, intelligent applications and intelligent benefits.
			Liu et al. [33]	2021	It is constructed from three levels: basic investment, production application, and market benefit.
			Wang and Meng [34]	2023	It is constructed from three aspects: key technology, production mode, and industrial competitiveness.
			Yue and Gu [35]	2023	The industry proportion of enterprise intelligent keyword counting.
			Lv et al. [20]	2022	The use of robots per 10,000 people in the industrial field.
Research on IM and CEP	Measurement of CEP		Tan and Dong [13]	2023	Single-factor index.
			Wang et al. [41]	2023	Total-factor index.
	The impact of intelligence on CEP		Liang et al. [36]	2022	Based on the interactive three-stage network DEA model, the production performance of China’s AI-driven low-carbon manufacturing industry is evaluated.
			Chen et al. [40]	2022	It is proved that AI has a significant inhibitory effect on CEI.
			Chen et al. [37]	2023	It is proved that AI has a positive impact on the CEP of manufacturing enterprises.
Research on EA	Measurement of EA		Yao et al. [22]	2023	Single-factor index.
			Wang et al. [49]	2022	Measured from the three dimensions of feature agglomeration, industrial agglomeration, and urban agglomeration.
	EA and CEP	Agglomeration degree	Qin and Wu [54]	2015	With the increase in agglomeration degree, CEI increases first and then decreases.
		Development stage	Yu et al. [55]	2022	Excessive EA leads to an increase in regional carbon intensity.
	EA and IM		Tang and Mai [57]	2022	The intellectualization of manufacturing industry has an inverted U-shaped impact on the diversified agglomeration of producer services.
Research on IM, CEP, and EA	Explore the relationship between the three		Jin et al.	2024	The mediating effect model is used to confirm that manufacturing intelligence improves CEP by promoting EA.

. Upon reviewing the existing body of research, it has become apparent that there is a divergence in the conclusions reached by scholars. For instance, when examining the impact of EA on CEs, a subset of researchers has demonstrated that an excessive concentration of economic activities can result in heightened regional carbon intensity. Conversely, other scholars argue that a higher degree of EA can actually mitigate marginal CEs. This disparity in the findings can be attributed to two primary factors. Firstly, there is a variation in research methodologies. The index systems and empirical procedures utilized by different scholars during their analyses can lead to contrasting results. Secondly, the subjects of the research themselves vary. Given the vast number of provinces in China, the diversity of resources, and the uneven development across regions, there are inherent differences in IM and CEP levels. Consequently, the samples selected by researchers may exhibit both regional and temporal variations, which further contribute to the heterogeneity of the conclusions drawn. In light of these considerations, our study aims to provide a more nuanced perspective by accounting for these methodological and regional differences, thereby enhancing the robustness and relevance of our findings within the broader context of environmental and economic policy discussions.

While the body of research on IM is indeed substantial and well established, there remains a notable gap in the analysis of mechanisms and empirical studies focusing on the triad of IM, CEP, and EA. For instance, the interplay between IM and CEP has not been thoroughly examined from the lens of EA. To fill this gap, this paper adopts EA as a pivotal intermediary variable to delve into the effects of China’s IM on CEP. Our empirical analysis substantiates that the advancement of IM can enhance CEP through the facilitation of EA (Table 1). This study not only augments the existing theoretical discourse on the nexus between IM, CEP, and EA but also holds significant potential for advancing economic and environmental synergies, advocating for green and low-carbon developmental trajectories, attaining the ambitious dual carbon objectives, confronting climate change challenges, and furthering the cause of global sustainable development.

3. Theoretical Analysis and Hypothesis

3.1. Influence of IM on CEP

Green and low-carbon development is essential for achieving the dual carbon goals, and the integration of intelligent technology and manufacturing can help reduce costs, improve efficiency, save energy, and reduce emissions. IM enhances energy conservation and emission reduction by lowering energy consumption, increasing production efficiency, and promoting low-carbon industrial transformation. Firstly, it reduces energy consumption by encouraging enterprises to adopt energy-saving processes through innovation or technology, integrating renewable energy sources, and optimizing the energy structure [13], Real-time monitoring and data analysis enable precise control and optimal energy utilization, reducing consumption intensity and pollution emissions [57]. Secondly, it improves production efficiency by refining production management processes, optimizing input structures, and enhancing green product design and production efficiency [59]. Intelligent technology also minimizes resource waste caused by improper human operations. Finally, IM fosters low-carbon industrial transformation by improving supply chain operations, driving upstream and downstream enterprises towards a low-carbon process design and services, and creating a green supply chain [37], Thus, we propose the following hypothesis:

Hypothesis 1 (H1).

There is a positive effect of IM on CEP.

3.2. IM, EA, and CEP

In the era of Industry 4.0, the world is witnessing a surge in digital and intelligent transformation. The flexibility of intelligent technology facilitates EA, which, in turn, supports energy conservation and emission reduction. Firstly, IM promotes EA by concentrating economic activities such as production, distribution, exchange, and consumption within specific spatial ranges. Intelligent technology overcomes geographical limitations, enabling resource flow and reorganization, breaking information barriers in the industrial chain, and fostering knowledge and technology sharing [57], and promotes the concentration of economic activities such as production and distribution. Additionally, IM meets consumers’ personalized and intelligent product needs, enhancing enterprises’ ability to provide precise products and services [60], and promotes the concentration of economic activities such as exchange and consumption. Therefore, the IM industry promotes the formation of the EA effect. Secondly, the positive externalities of EA can minimize the degree of environmental pollution [56]. EA reduces production costs, reduces transportation and commuting distances, and improves production efficiency [48] and energy efficiency through factor sharing to reduce pollution emissions; at the same time, EA improves the utilization rate of human, capital, and other factors through spillover effects such as economies of scale, knowledge spillovers, and technology spillovers, reduces marginal CEs in the production process [49] and improves economic efficiency. Therefore, by promoting the agglomeration of economic activities in a certain space, the IM industry promotes energy conservation, emission reduction, and economic development to achieve a win–win situation, thereby improving CEP. Based on the above discussion, Hypothesis 2 is proposed:

Hypothesis 2 (H2).

EA significantly mediates the positive effect of IM on CEP.

Overall, the direct influence of IM on CEP is manifested through its ability to enhance CEP by reducing energy consumption, increasing production efficiency, and fostering the low-carbon transformation of the industrial chain. Meanwhile, the indirect effect of IM on CEP is mediated by EA. This occurs as IM facilitates the restructuring of factor flows and the precise positioning of products and services, which in turn promotes EA. EA, for its part, enhances energy efficiency and decreases marginal CEs through the promotion of factor sharing and knowledge spillover effects, thereby further improving CEP. To more effectively illustrate the complex interplay among these factors, we have depicted the mechanism relationship between IM, EA, and CEP in Figure 1. In this figure, the blue box encapsulates the direct impact effect, which corresponds to the mechanism outlined in Hypothesis 1; the green box encapsulates the indirect effect, which corresponds to the mechanism outlined in Hypothesis 2. This visual representation serves to clarify the intricate network of relationships and the dual pathways through which IM exerts its influence on CEP.

4. Research Design

4.1. Research Model

4.1.1. Fixed Effect Baseline Regression Model

The fixed effect model is appropriate for addressing time and individual correlation issues in the data, effectively resolving endogeneity problems caused by omitted variables. Li et al. (2024) analyzed the impact of the digital economy on CEP in various cities in China, employing a double fixed effect model to control for time and individual variables, ensuring the estimation results were unaffected by unobserved factors [24]. Chen and Jin (2023) studied the relationship between AI and CE in manufacturing enterprises [37], utilizing a fixed effect model to control for time variables and avoid temporal changes in regression outcomes. When analyzing the impact of IM on CEP, it is essential to control for province-specific characteristics that remain constant over time and time-specific characteristics that do not vary across provinces. This approach helps avoid the interference of inherent provincial differences and temporal factors affecting CEP, reducing measurement errors and accurately assessing the impact of IM on CEP. Based on the above analysis, we construct the following double fixed effect baseline regression model:

$${\mathrm{C}\mathrm{E}\mathrm{P}}_{\mathrm{i}\mathrm{t}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_1{\mathrm{I}\mathrm{M}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\beta }\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\mu }}_{\mathrm{i}}+{\mathsf{\delta }}_{\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$$

(1)

i and t represent province i and year t, respectively. ${\mathrm{C}\mathrm{E}\mathrm{P}}_{\mathrm{i}\mathrm{t}}$ represents the CEP of the explained variable, ${\mathrm{I}\mathrm{M}}_{\mathrm{i}\mathrm{t}}$ represents the core explanatory variable IM, ${\mathsf{\alpha }}_0$ represents the constant term, ${\mathsf{\alpha }}_1$ represents the estimated parameters of IM, ${\mathrm{X}}_{\mathrm{i}\mathrm{t}}$ represents the control variables that affect the CEP, ${\mathsf{\mu }}_{\mathrm{i}}$ represents the fixed effect of the province, ${\mathsf{\delta }}_{\mathrm{t}}$ represents the fixed effect of time, and ${\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$ represents the random error term.

4.1.2. Mediation Effect Model

In analyzing the mediating effect, many scholars employ the stepwise regression method. Lin and Zhou (2021) used the three-step regression method to investigate the impact of Internet development on energy and CEP [42]. They tested the effects of the Internet on energy and CEP, the Internet on IS upgrading and technology diffusion, and the combined effects on energy and CEP. This allowed them to determine that IS upgrading and technology diffusion partially mediate the Internet’s impact on energy and CEP. For analyzing the impact path of IM on CEP, we refer to the regression method of Wen and Ye (2024) and divide the analysis into three parts to test the impact mechanism [61]. Equation (1) tests the impact of IM on CEP without considering EA; Equation (2) tests the impact of IM on EA; and Equation (3) examines the combined impact of IM and EA on CEP:

$${\mathrm{E}\mathrm{A}}_{\mathrm{i}\mathrm{t}}={\mathrm{b}}_0+{\mathrm{b}}_1{\mathrm{I}\mathrm{M}}_{\mathrm{i}\mathrm{t}}+{\mathrm{b}\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\mu }}_{\mathrm{i}}+{\mathsf{\delta }}_{\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$$

(2)

$${\mathrm{C}\mathrm{E}\mathrm{P}}_{\mathrm{i}\mathrm{t}}={\mathrm{c}}_0+{\mathrm{c}}_1{\mathrm{I}\mathrm{M}}_{\mathrm{i}\mathrm{t}}+{\mathrm{c}}_2{\mathrm{E}\mathrm{A}}_{\mathrm{i}\mathrm{t}}+{\mathrm{c}\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\mu }}_{\mathrm{i}}+{\mathsf{\delta }}_{\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$$

(3)

Among them, ${\mathrm{E}\mathrm{A}}_{\mathrm{i}\mathrm{t}}$ is the mediating variable of EA, and the a₁ coefficient in Equation (1) is the total effect of IM on CEP; the b₁ coefficient in Equation (2) is the effect of IM on EA; and the coefficient c₂ in Equation (3) is the effect of EA on CEP after controlling the impact of IM. The coefficient c₁ is the direct effect of IM on CEP after controlling the impact of EA. Firstly, it is verified whether the mediating effect exists. If the coefficient b₁ is significant in Equation (2), c₁ and c₂ are significant in Equation (3), and the absolute value of c₁ in Equation (3) is smaller than a₁ in Equation (1), it can be inferred that EA plays a partial mediating effect in the impact of IM on CEP.

4.2. Definitions of the Variables

4.2.1. Dependent Variable

With the introduction of new technologies and the establishment of the total-factor framework, the economy, energy, and environment are gradually integrated into a unified performance evaluation framework [62]. The TFCEP provides an objective and accurate assessment of whether an economic development model meets the simultaneous requirements of energy conservation, emission reduction, and economic growth [41]. However, total-factor measurement indices that solely use DEA and similar methods combined with input–output indicators typically analyze a static CEP within a cross-sectional data framework. It is necessary to introduce the Malmquist index to measure changes in CEP over time [63]. Kao and Hwang viewed the overall efficiency of the two-stage network DEA model as the product of two sub-stage efficiencies, opening the production “black box” of the traditional DEA model [36].

Based on the above analysis, this study employs the network DEA–Malmquist index model to measure the TFCEP as the dependent variable. Capital stock, labor, total energy consumption, transportation, and interest expenditure are selected as input factors, while regional GDP and the total CEs of the manufacturing industry are selected as output factors. The capital stock is estimated using the perpetual inventory method based on the 1978 base period, drawing on the research of Zhang et al. (2004) [64]. Labor is measured by the total number of employed people in urban manufacturing units; total energy consumption is calculated as the sum of various energy types consumed in the region; and transportation and interest expenses are measured by the added value of transportation, warehousing, postal services, and industrial enterprises’ interest expenses. Manufacturing CEs are the direct emissions generated by using energy sources such as natural gas and liquefied petroleum gas in the 31 manufacturing industries.

4.2.2. Independent Variable

The core explanatory variable is IM. The intelligence of the manufacturing industry encompasses not only the basic investment in intelligent equipment such as AI but also the social and economic benefits generated by the production, application, and intelligence of intelligent products [33]. Firstly, the integration of intelligent technology and manufacturing requires the input of fundamental elements such as intelligent equipment and human resources, as well as support from infrastructure like the Internet, communication, and transportation. Secondly, the promotion and application of intelligent products and services, software, and information technology are crucial standards for measuring the integration effect of intelligent technology with the manufacturing industry. Finally, the economic, environmental, innovation, and other social benefits brought by intelligence are key drivers for the manufacturing industry to engage in the new industrial revolution. Based on this, this paper draws on the research of Liu et al. (2021) to construct a comprehensive index evaluation system for IM from three aspects, intelligent foundation, intelligent application, and intelligent benefit, ensuring the evaluation system is scientific, systematic, comprehensive, and representative [33]. The specific indicators and calculation methods are detailed in

Table 2. Intelligent comprehensive index system of manufacturing industry.

Global Indicator	First Grade Indexes	Second Grade Indexes	Third Grade Indexes	Method of Calculation
IM	Intelligent foundation	Intelligent equipment investment	Robot installation density	Number of regional industrial robot installations * (number of regional industrial employees/total number of national employees)
		Intelligent manpower input	Human resources input level	Number of research and development (R&D) personnel in high-tech industries/manufacturing employees
		Communication capability	Coverage rate of long-distance optical cable	Long-distance optical cable line length/provincial area
		Product circulation ability	Grade highway coverage rate	Grade highway mileage/provincial area
		Information resource collection ability	Internet penetration rate	Number of Internet users/provincial population
	Intelligent application	Intelligent products	Profitability of high-tech products	Embedded system income/main business income of enterprises above designated size
		Intelligent service	The importance of service industry	Information technology service income/main business income of enterprises above designated size
		Intelligent management	Popularization and application of software	Software product income/main business income of enterprises above designated size
		Software and information technology	Popularity of information technology	Information transmission, software, and information technology services urban unit employees/manufacturing employees
	Intelligent benefits	Economic benefits	Production and operation of new products	New product sales revenue of high-tech industry/main business income of above-scale industry
		Environmental benefits	Energy intensity	Energy consumption/GDP
		Innovation benefits	The degree of intelligent innovation	AI patents/total patents

Note: * in the table indicates the multiplication of two variables; / indicates that two variables divide.

. Using the established evaluation index system, the entropy weight method calculates the comprehensive development level of IM.

4.2.3. Control Variables

In addition to intelligent technology, many factors will affect the CEP of the manufacturing industry. To reduce the potential estimation error and ensure the comprehensiveness of the analysis, we add control variables ${\mathrm{X}}_{\mathrm{i}\mathrm{t}}$ to the fixed effect model. Based on previous studies, we choose six variables that may have an impact on CEP, including IS, education level, technological innovation ability, environmental regulation, employment density, and urbanization level.

(1) Industrial Structure (IS): Measured by the proportion of industrial added value in GDP. The greater the proportion of industry in the IS, the lower the level of CEP [48]. Therefore, the proportion of industrial added value is used to represent the impact of IS on CEP, and the expected coefficient is negative.

(2) Educational Level (EL): Measured by the number of years of education per capita [59]. The higher the level of education, the higher the people’s low-carbon awareness, which is beneficial to improving CEP; however, a high education level also represents a high population density and a high economic development level, which will consume more energy and reduce CEP. The expected coefficient symbol is uncertain.

(3) Technological Innovation Capability (TIC): Measured by the proportion of new product development funds of industrial enterprises in a regional GDP. The higher the technological innovation ability, the stronger the ability of enterprises to use energy-saving technologies through independent innovation and promote the improvement of energy efficiency, which is conducive to reducing pollution emissions in the production process and improving CEP. The expected coefficient symbol is positive.

(4) Environmental Regulation (ER): Measured by the proportion of investment in industrial pollution control to industrial added value. Environmental regulation is considered to be an effective way to curb CEs [43]. In this paper, the proportion of industrial pollution control is used to measure the “forced effect” of environmental regulation. The greater the value, the more serious the pollution of waste gas and wastewater, the weaker the intensity of environmental regulation, the lower the CEP. The expected coefficient symbol is negative.

(5) Employment Density (ED): Measured by the ratio of manufacturing urban employment population to urban construction land area. The higher the employment density, the higher the energy consumption, traffic congestion, etc., and the lower the CEP level. The expected coefficient symbol is negative.

(6) Urbanization Level (UL): Measured by population urbanization rate. According to the existing research, the relationship between urbanization level and CEP has not yet reached a consensus, such as positive, negative, insignificant, inverted U-shaped, nonlinear, etc. [43]. The expected coefficient symbol is uncertain.

4.2.4. Mediating Variable

The mediating variable of this study is EA. Since there is no unified definition of EA measurement indicators, scholars mostly use economic density [50], employment density [51], and population density [52] to measure it. Based on understanding its definition and drawing on the existing literature, this study establishes a comprehensive index system to measure the level of EA. The specific indicators are shown in

Table 3. Comprehensive index system of EA.

Global Indicator	First-Grade Indexes	Second-Grade Indexes
EA	Employment density	Labor concentration
	Economic density	Geographical concentration
		Railway business index
		Grade highway coverage rate
		Concentration of freight transport
	Population density	Urban population concentration
	Consumption density	Consumer demand index
		Foreign trade concentration

. To comprehensively investigate the relationship between regional economic output and geographical area, and draw lessons from the calculation method of Zhu et al. (2023) using geographical concentration as the proxy variable of EA [65], the calculation of secondary indicators in this study follows the following formula:

$${\mathrm{E}\mathrm{A}}_{\mathrm{i}\mathrm{t}}={\displaystyle \frac{\raisebox{1ex}{${\mathrm{A}}_{\mathrm{i}\mathrm{t}}$}\!\left/ \!\raisebox{-1ex}{${\sum }_{\mathrm{i}=1}^{\mathrm{N}}{\mathrm{A}}_{\mathrm{i}\mathrm{t}}$}\right.}{\raisebox{1ex}{${\mathrm{S}}_{\mathrm{i}\mathrm{t}}$}\!\left/ \!\raisebox{-1ex}{${\sum }_{\mathrm{i}=1}^{\mathrm{N}}{\mathrm{S}}_{\mathrm{i}\mathrm{t}}$}\right.}}$$

(4)

Among them, ${\mathrm{A}}_{\mathrm{i}\mathrm{t}}$ represents the second-level indicator proxy variable of the region i in t years, which represents the number for manufacturing labor employment, regional GDP, railway business mileage, grade highway mileage, freight volume, urban population, consumption level, and the total import and export of foreign-invested enterprises. ${\sum }_{\mathrm{i}=1}^{\mathrm{N}}{\mathrm{A}}_{\mathrm{i}\mathrm{t}}$ represents the sum of the proxy variables of the secondary indicators in all regions, and ${\mathrm{S}}_{\mathrm{i}\mathrm{t}}$ and ${\sum }_{\mathrm{i}=1}^{\mathrm{N}}{\mathrm{S}}_{\mathrm{i}\mathrm{t}}$ represent the sum of the urban construction land area of region i in year t and the urban construction land area of all regions in China.

4.3. Data Source

This study selects panel data from 30 provinces (autonomous regions and municipalities directly under the central government) in mainland China from 2010 to 2022. Given the availability of data, the samples do not include Taiwan, Hong Kong, Macao, and Tibet. All data are from the China Carbon Accounting Database, China Labor Statistics Yearbook, China Population and Employment Statistics Yearbook, Ministry of Industry and Information Technology of the People’s Republic of China, State Intellectual Property Office, National Bureau of Statistics, China Provincial Statistical Yearbook, and The International Federation of Robotics. The descriptive statistical results of all variables are shown in

Table 4. Definition and summary of key variables.

Variable	Obs	Mean	Std. Dev.	Min	Max
CEP	360	1.627	0.496	0.719	3.449
IM	360	0.100	0.076	0.019	0.623
IS	360	0.334	0.080	0.100	0.574
EL	360	9.899	0.738	6.361	12.586
TIC	360	0.012	0.007	0.002	0.040
ER	360	0.003	0.003	0.000	0.031
ED	360	0.071	0.044	0.014	0.250
UL	360	0.589	0.125	0.338	0.896

.

5. Empirical Results and Analysis

5.1. Estimation Results

According to the Hausman test results, the fixed effect model was selected. To control for the effect of IM on CEP, the benchmark regression results are presented in

Table 5. Estimation results of regression results.

Variables	(1)	(2)
	CEP	CEP
IM	1.164 ***	0.610 *
	(0.349)	(0.350)
IS		1.026 *
		(0.472)
EL		0.003
		(0.019)
TIC		13.246 ***
		(4.376)
ER		−4.195
		(4.628)
ED		−1.278 *
		(0.673)
UL		3.962 ***
		(0.658)
Constant	0.943 ***	−0.457 ***
	(0.042)	(0.457)
Year FE	YES	YES
Province FE	YES	YES
Observations	360	360
R-squared	0.704	0.396

Note: *, *** represent the significant level of 10%, and 1% respectively; the value in brackets below the coefficient is the robust standard error t value.

. Column (1) shows the regression results for IM, which only includes the core explanatory variable. The regression coefficient of the IM industry is 1.164, and it is significantly positive at the 1% statistical level, indicating that a 1% increase in the level of IM leads to a 1.16% increase in CEP. Column (2) displays the regression results after adding control variables. The regression coefficient of MI is 0.610, which is significantly positive at the 10% statistical level, indicating that a 1% increase in the level of IM leads to a 0.61% increase in CEP. Although the estimated coefficient and significance level decrease after including control variables, they remain statistically significant. This indicates that, in both economic and statistical terms, IM has a significant positive effect on CEP, confirming H1.

From the regression results of the control variables, the coefficient for IS is positive, contrary to the expected sign, suggesting that the proportion of industrial added value positively affects CEP. This result may be due to significant measures taken by the Chinese government to address climate change, such as adjusting the IS, improving energy efficiency, and establishing a carbon emission trading mechanism. These efforts have led to substantial reductions in industrial CEs and improvements in CEP [66]. The regression coefficient for education level is positive but not significant, indicating that while the education level promotes CEP improvement to some extent, this effect is not pronounced. The regression coefficient for technological innovation capability is positive and significant at the 1% statistical level, suggesting that enhancing technological innovation capability significantly boosts CEP. The regression coefficient for environmental regulation is negative, consistent with the expected sign, but not significant, indicating that higher investment in industrial pollution control corresponds to weaker environmental regulation intensity and a non-significant reduction in CEP, drawing on the research of Zhao et al. (2023) [43]. The regression coefficient for employment density is negative and significant at the 10% statistical level, indicating that increased employment density somewhat reduces CEP. The regression coefficient for urbanization level is positive and significant at the 1% statistical level, indicating that urbanization level significantly promotes CEP, consistent with Tian and Wu’s (2023) research [67].

5.2. Robustness Test

To substantiate the reliability of our empirical findings, we have conducted a series of robustness checks. Drawing on established research practices, we have employed a variety of techniques, including the substitution of explanatory variables, the exclusion of outlier observations, and the utilization of instrumental variables. The process of addressing extreme values involved both the removal of outlier provinces and cities, as well as shrinking tail processing. The outcomes of these robustness tests demonstrate that the regression coefficient associated with IM remains statistically significant. This consistency affirms the robustness of our empirical findings regarding the impact of IM on CEP, thereby validating the key assumptions underpinning our research. The specific robustness test content is as follows.

Firstly, we replaced the explained variable. Referring to the research of Wang et al., we used CEI instead of CEP as the explanatory variable. CEI is the ratio of a province’s total CE to its gross regional product. A lower CEI indicates a higher energy efficiency and better CEP [41]. As shown in

Table 6. Robustness regression results.

Variables	(1)	(2)	(3)	(4)
	CEI	CEP	CEP	CEP
IM	−11.260 ***	2.390 **	2.203 ***	4.107 ***
	(2.760)	(0.996)	(0.488)	(0.543)
IS	17.544 ***	−1.410 ***	−2.087 ***	−1.491 ***
	(2.215)	(0.292)	(0.368)	(0.253)
EL	0.221	−0.151 ***	−0.026	−0.039
	(0.164)	(0.037)	(0.037)	(0.030)
TIC	−110.060 ***	18.139 ***	28.479 ***	18.636 ***
	(22.040)	(5.414)	(3.279)	(3.692)
ER	178.553 ***	−19.444 ***	−17.404 ***	−13.445 ***
	(60.313)	(4.396)	(4.546)	(3.657)
ED	−26.177 ***	−2.165 ***	−2.103 ***	−2.555 ***
	(3.189)	(0.596)	(0.527)	(0.475)
UL	6.026 ***	−1.490 ***	−1.426 ***	−1.500 ***
	(1.286)	(0.279)	(0.260)	(0.175)
Constant	−2.774	4.205 ***	3.056 ***	2.996 ***
	(1.825)	(0.367)	(0.378)	(0.303)
Observations	360	300	360	360
R-squared	0.505	0.595	0.563	0.618

Note: **, *** represent the significant level of 5%, and 1% respectively; the value in brackets below the coefficient is the robust standard error t value.

, Column (1) presents the regression results for CEI as the explained variable. The regression coefficient of the IM industry is −11.260, and it is significant at the statistical level of 1%, which indicates that the improvement of the intelligent comprehensive development level of the manufacturing industry has a significant inhibitory effect on carbon emission intensity. According to the above analysis, as the level of IM increases, the carbon emission intensity decreases, thereby improving the level of CEP; that is, IM promotes the level of CEP, which is consistent with the previous research hypothesis, indicating that the results are robust.

Secondly, we addressed extreme values using two methods. The first method eliminated specific provinces and cities. Given Guangdong Province’s high robot installation density and the complex administrative structures of Beijing, Tianjin, Shanghai, and Chongqing—regions with high economic development, green technology concentrations, and intelligence levels [28]—we excluded these five regions to ensure the empirical results’ general applicability. The results, shown in Column (2) of Table 6, indicate that IM still significantly enhances CEP. The regression coefficient of IM is 2.390, which is significant at the 5% statistical level, indicating that after excluding individual provinces and cities, IM still significantly promotes CEP, consistent with the expected conclusion and proving robustness. The second method involved trimming extreme values. To avoid regression result deviations caused by outliers, we trimmed all variables at the 1% and 5% levels [68] and conducted an ordinary least squares regression on the processed data. The results are shown in Columns (3) and (4) of Table 6. The regression coefficients of IM are 2.203 and 4.107, respectively, and both are significant at the 1% statistical level, indicating that after the 1% and 5% tail reduction, IM still significantly promotes CEP, consistent with the baseline regression results, proving robustness.

Finally, we addressed endogeneity. Given the many factors influencing CEP, this study only includes certain control variables, making it challenging to avoid estimation bias from omitted variables. To improve CEP, enterprises may engage in low-carbon technology R&D and optimize production processes to improve their intelligence level, potentially leading to a reverse causality [40]. First and foremost, as international cooperation intensifies, China’s manufacturing industry is increasingly integrated with the global landscape of intelligent development. Consequently, the sophistication of China’s IM is inherently linked to the AI advancements in other nations. However, the scale of robot installations in other countries is unlikely to exert a substantial influence on China’s CEP. Furthermore, there exists a correlation between the intelligence levels of the manufacturing industry in the lag period and the current period. Nevertheless, this historical correlation does not appear to have a direct bearing on the CEP in the present period. Based on the aforementioned analysis, to address estimation bias from omitted variables and reverse causality, we draw on the research of Chen et al. (2022) [40] and Ye and Zhang (2024) [58], introducing the average robot installation density (ID) in the United States, Germany, Japan, and South Korea, and the lagged core explanatory variable (L.IM) as instrumental variables. These instrumental variables passed over-identification and weak instrument tests, meeting correlation and homogeneity requirements. The regression results, shown in

Table 7. Instrumental variable regression results.

Variables	(1)	(2)	(3)
	IM	CEP	CEP
ID	0.012 ***
	(0.121)
L.IM			3.120 ***
			(0.568)
IM		3.855 *
		(1.970)
IS	−0.394 ***	−1.419	−0.478
	(0.083)	(0.976)	(0.512)
EL	−0.004	−0.025	0.012
	(0.013)	(0.047)	(0.025)
TIC	2.882 ***	21.917 **	15.175 ***
	(0.590)	(9.984)	(5.452)
ER	−5.050 ***	−8.567	−17.376 ***
	(1.113)	(10.341)	(5.253)
ED	−0.197 **	−1.865 ***	−6.570 ***
	(0.092)	(0.586)	(0.787)
UL	0.251 ***	−1.814 ***	−2.699 ***
	(0.054)	(0.547)	(0.548)
Constant	0.104 ***	2.916 ***	3.318 ***
	(0.121)	(0.447)	(0.290)
Observations	360	360	330
R-squared	0.521	0.521	0.731

Note: *, **, *** represent the significant level of 10%, 5%, and 1% respectively; the value in brackets below the coefficient is the robust standard error t value.

, indicate that the regression coefficient for IM remains significantly positive, proving robustness.

5.3. Heterogeneity Analysis

Considering the differences in economic development level and resource endowment, there may be great differences in the level of IM and CEP in different regions. This study analyzes the heterogeneity from three dimensions—geographical location, EA degree, and low-carbon pilot provinces and cities—to more accurately understand the relationship between IM and CEP.

First of all, according to the regional classification standard formulated by the National Bureau of Statistics of China, the provinces are divided into eastern, central, and western regions. The regression results are shown in

Table 8. Regional heterogeneity regression results.

Variables	(1)	(2)	(3)
	CEP	CEP	CEP
IM	1.545 ***	5.804 ***	2.752 **
	(0.516)	(1.423)	(1.095)
IS	−3.634 ***	−1.948 ***	−1.640 ***
	(0.570)	(0.381)	(0.562)
EL	0.046	0.193 **	−0.127 **
	(0.037)	(0.078)	(0.058)
TIC	42.895 ***	7.932	41.798 ***
	(5.942)	(5.747)	(13.998)
ER	−43.010 **	−10.888	−16.637 ***
	(17.793)	(11.520)	(5.856)
ED	0.361	−4.784 ***	−6.739 ***
	(0.941)	(0.884)	(1.218)
UL	−1.010 **	−1.363 ***	−1.349 ***
	(0.393)	(0.447)	(0.421)
Constant	2.163 ***	0.838	4.016 ***
	(0.500)	(0.887)	(0.526)
Observations	132	96	132
R-squared	0.568	0.786	0.699

Note: **, *** represent the significant level of 5%, and 1% respectively; the value in brackets below the coefficient is the robust standard error t value.

, and there are indeed regional differences in the impact of IM on CEP. Columns (1), (2), and (3) represent the regression results of the eastern, central, and western regions respectively. The IM industry can significantly promote the improvement of CEP in the central and eastern regions, and the promotion effect in the western region is relatively weak. It may be because the eastern region, as the most developed region, has a higher level of intelligence, but the high concentration of intelligent industries may lead to increased energy consumption and increased pollutant emissions, offsetting the promotion effect of some intelligence on CEP. The rise of the central region is strong, with both a high intelligence development level and an emphasis on environmental protection, which has the best effect on promoting CEP. The level of economic development in the western region is relatively lagging; the integration of intelligent technology and traditional manufacturing industry has not yet reached a high level, energy efficiency and low-carbon technology levels are relatively low, and the improvement effect of IM on CEP is relatively poor.

Secondly, whether it is a low-carbon pilot province or city is used as a regional classification standard [58]. The regression results are shown in Columns (1) and (2) of

Table 9. Other heterogeneity test regression results.

Variables	(1)	(2)	(3)	(4)
	CEP	CEP	CEP	CEP
IM	2.237 ***	2.882 ***	0.155	2.852 ***
	(0.584)	(0.559)	(0.785)	(0.349)
IS	−3.058 ***	−1.976 ***	−7.831 ***	−0.731 **
	(0.542)	(0.402)	(0.823)	(0.284)
EL	−0.023	−0.125 ***	0.098 *	−0.049 **
	(0.038)	(0.041)	(0.054)	(0.024)
TIC	48.252 ***	14.131 ***	30.842 ***	18.559 ***
	(7.578)	(4.214)	(5.607)	(4.186)
ER	−32.489	−24.728 ***	−63.316 ***	−16.067 ***
	(21.190)	(5.386)	(21.895)	(5.053)
ED	−0.759	−2.849 ***	−1.419	−5.608 ***
	(1.229)	(0.607)	(1.006)	(0.809)
UL	−2.408 ***	−0.357	−0.183	−1.829 ***
	(0.403)	(0.301)	(0.397)	(0.200)
Constant	3.592 ***	3.604 ***	3.476 ***	3.271 ***
	(0.463)	(0.386)	(0.600)	(0.247)
Observations	120	240	96	264
R-squared	0.570	0.602	0.756	0.647

Note: *, **, *** represent the significant level of 10%, 5%, and 1% respectively; the value in brackets below the coefficient is the robust standard error t value.

. Column (1) indicates the regression results of the six provinces and four cities covered by the first two batches of low-carbon pilot provinces (autonomous regions and municipalities directly under the central government), and Column (2) indicates the regression results of the areas not covered by the first two batches of low-carbon pilots. Whether it is a low-carbon pilot area or not, the regression coefficient of IM is significantly positive at the statistical level of 1%, and the promotion effect of IM on CEP in non-low-carbon pilot areas is greater. It may be because non-low-carbon pilot areas have not well combined the work of adjusting the IS, optimizing the energy structure, and increasing carbon sinks, while the development of IM can not only save energy and increase efficiency, it also improves the potential of non-pilot areas to improve CEP.

Finally, taking the degree of EA as the division standard, the regions above and below the national average agglomeration level are divided into high agglomeration level and low agglomeration level regions, respectively. The regression results are shown in Columns (3) and (4) of Table 9. Column (3) indicates the regression results of regions with high EA levels. The regression coefficient of IM is positive, but not significant. It can be seen that when the agglomeration degree is high; IM can promote the improvement of CEP, but the effect is not obvious. Column (4) shows the regression results of a low agglomeration degree. The regression coefficient of IM is significantly positive at the statistical level of 1%, indicating that there are positive externalities such as factor optimization allocation and knowledge technology spillover before the EA is low or excessive agglomeration is achieved. IM can significantly promote the improvement of CEP.

5.4. Mechanism Analysis

The empirical test results of this study show that IM can significantly promote the improvement of CEP in the economic and statistical sense. After a series of robustness tests and the use of instrumental variables to solve the endogenous problems, the results are still significant. However, the path of IM affecting CEP needs to be further explored. Based on a theoretical analysis, IM may affect CEP through the path of EA. Therefore, we incorporate EA as an intermediary variable into the research framework for regression testing. As shown in

Table 10. Intermediate effect regression results.

Variables	(1)	(2)	(3)
	CEP	EA	CEP
IM	2.136 ***	0.121 **	1.666 ***
	(0.414)	(0.053)	(0.362)
EA			3.898 ***
			(0.374)
IS	−1.278 ***	−0.115 *	−0.830 **
	(0.479)	(0.062)	(0.417)
EL	0.031	0.005	0.011
	(0.025)	(0.003)	(0.022)
TIC	19.086 ***	1.779 **	12.151 **
	(5.411)	(0.697)	(4.734)
ER	−15.653 ***	−0.702	−12.918 ***
	(5.343)	(0.688)	(4.637)
ED	−5.606 ***	−0.018	−5.535 ***
	(0.731)	(0.094)	(0.634)
UL	−3.196 ***	0.116 *	−3.648 ***
	(0.519)	(0.067)	(0.452)
Constant	4.013 ***	0.073	3.729 ***
	(0.461)	(0.059)	(0.400)
Observations	360	360	360
R−squared	0.792	0.905	0.844

Note: *, **, *** represent the significant level of 10%, 5%, and 1% respectively; the value in brackets below the coefficient is the robust standard error t value.

, Columns (1), (2), and (3) correspond to econometric models (1), (2), and (3) respectively, reflecting the test results of the intermediary effect of EA. In Column (2), the regression coefficient of IM is significantly positive at the 5% statistical level, indicating that with the improvement in IM, the degree of EA can be significantly improved. In Column (3), the regression coefficient of EA is significantly positive at the 1% statistical level, indicating that EA can significantly promote the improvement of CEP.

6. Discussions and Suggestions

6.1. Discussions

As the core of the new generation of the industrial revolution, IM facilitates low-carbon emission reduction by reducing energy consumption, enhancing production efficiency, and promoting the low-carbon transformation of the industrial chain. Utilizing provincial panel data from China, this study empirically demonstrates that IM significantly enhances CEP, aligning with the findings of many scholars. Chen and Jin (2023) explored the relationship between manufacturing enterprise intelligence and CE from a micro perspective, concluding that intelligent technology positively impacts carbon emission reduction [37]. Similarly, Chen et al. (2022) found that AI at the urban level significantly inhibits CEI [40], while Lv et al. (2022) demonstrated that industrial robots can substantially decrease CE in the industrial sector [20]. However, Liang et al. (2022) argue that the overall efficiency of China’s AI-driven low-carbon manufacturing performance is low, indicating considerable room for improvement [36]. Additionally, Zhang et al. (2022) highlight that industrial intelligence may introduce new sources of pollution [69].

EA exerts dual effects of energy saving and emission reduction [48]. On the one hand, it reduces production costs and enhances energy efficiency through factor sharing; on the other hand, it improves factor utilization and decreases marginal CEs via economies of scale, knowledge spillovers, and other mechanisms. This study reveals that IM influences CEP by fostering EA. Meng and Zhao (2023) corroborate that industrial intelligence indirectly boosts carbon productivity mainly through diversified agglomeration [47], consistent with our findings. However, when some scholars study EA as an intermediary variable, the conclusion is not that IM promotes EA and EA promotes CEP. For example, in their analysis of the influence of IM on innovation performance, with EA serving as a mediating factor, Tang et al. (2022) posit that IM exerts an “inverted U” effect on diversified agglomeration [57]. This suggests that IM stimulates diversified agglomeration prior to reaching a specific threshold. However, once a certain level of intelligentization is achieved, the industry’s advancement in intelligence begins to restrict further diversified agglomeration. Similarly, when exploring the dynamics between EA and carbon emission intensity, Shao et al. (2019) have demonstrated an “inverted N” curve relationship. This relationship is characterized by two distinct inflection points in the level of agglomeration. Below the first inflection point, EA tends to suppress carbon emission intensity. Between the first and second inflection points, there is a notable increase in the promoting effect on carbon emission intensity. Beyond the second inflection point, the trend reverses, and EA once again exerts a dampening effect on carbon emission intensity [48]. The divergent conclusions in the above research regarding the relationship between intelligence, EA, and CE may stem from variations in sample composition, research methodologies, and index selection. Furthermore, most existing studies focus on the relationship between AI, carbon emission reduction, and industrial agglomeration, yet AI and other technologies merely enhance the production efficiency of various aspects of IM [30]. Additionally, there are conceptual and measurement differences between EA and industrial agglomeration. Therefore, examining the relationship between the entire manufacturing industry and CEP through the lens of IM, as well as exploring the influence path of EA, holds significant theoretical value.

Based on the findings of this paper, as the intelligent development level of China’s manufacturing industry advances, economic activities tend to concentrate within specific geographic regions, consequently enhancing the efficiency of CE. However, the extent to which IM improves CEP varies across countries or regions with distinct economic, regulatory, technological, and policy environments. In the United States, economic planning is notably decentralized among states. California, for instance, has been proactive in enacting policies that leverage IM to foster renewable energy and energy efficiency, effectively utilizing intelligent and digital technologies to optimize CEP. In contrast, other states, due to policy disparities or variations in IS, may not achieve the same level of carbon emission reductions as California within the framework of IM [70]. India’s industrial development and environmental policies differ significantly from those of China, and the investment in IM is often constrained by infrastructural limitations. The ambitious goal of transitioning from coal-fired to renewable energy power generation has not yet yielded the anticipated improvements in CEP, highlighting the complex interplay between policy objectives and practical outcomes [71].

However, the journey towards enhancing CEP through IM is not without its challenges. These include the difficulty of the widespread adoption of intelligent technologies, the suboptimal intelligence levels of energy management systems, and the complexities involved in carbon emission accounting for IM enterprises. Drawing from the theoretical conclusions of this paper, to address these issues, it is imperative to bolster the integration of intelligent technology with the traditional manufacturing industry. This should be done with a focus on intelligent production, aiming to establish a tailored carbon emission mechanism for each manufacturing enterprise. For instance, the Donghua Technology Company Limited (Donghua Technology Co., Ltd.) encountered significant challenges in CEs during its transformation and development. Inspur Intelligent Production addressed these by integrating AI, digital twin technologies, and intellectual property models to create an intelligent carbon asset management platform for Donghua Technology Co., Ltd. By implementing a “dual carbon and dual control” three-step method, they achieved a precise measurement and analysis of CEs, setting a benchmark for green manufacturing in the building materials industry and significantly reducing CEs. In this intelligent transformation, Donghua Technology Co., Ltd. exemplified the immense potential of IM in advancing CEP improvements. The company was recognized as one of the top ten exemplary cases of equipment renewal and technological transformation at the provincial level in China [72]. These examples of integrating intelligent technology with traditional manufacturing to enhance CEP underscore that examining the relationship between IM and CEP not only enriches the theoretical framework of intelligence but also amplifies the impact of green carbon emission reduction efforts. This has profound practical significance for China and other developing countries around the world in their pursuit of high-quality economic development and the sustainable, healthy progression of environmental development.

6.2. Suggestions

In order to promote the further coupling between the theoretical research conclusions and actual carbon emission reduction needs, we propose the following suggestions from two perspectives. From the perspective of different subjects, first of all, the government should comprehensively evaluate their investment cost and carbon emission reduction benefits, and give priority to investing in intelligent technologies and industries with high cost-effectiveness. It should formulate clear preferential policies and regulations, guide the flow of social funds to IM enterprises that have a significant effect on carbon emission reduction, and support enterprises to develop innovative intelligent technologies through subsidies such as tax relief; strengthen international cooperation, docking international standards, and the introduction of foreign advanced intelligent technology and management experience. Secondly, industry associations should build an intelligent, green, and low-carbon information-sharing platform, publish the latest technology trends, policy information and enterprise cases, and promote the common progress of the industry. Finally, enterprises should implement green supply chain management, establish a transparent carbon emission reporting mechanism, and reduce energy consumption through intelligent equipment such as industrial robots. At the same time, manufacturing enterprises should strengthen cooperation with intelligent production technology companies, integrate intelligent technology and traditional manufacturing at the lowest cost, and create a unique intelligent carbon asset management platform.

From the perspective of heterogeneity and, first of all, regional heterogeneity, different regions in China are different in terms of economic development level, technological infrastructure, and human resources, so that the improvement effect of IM on CEP is different. Therefore, from the realistic perspective of improving CEP by intelligence, the eastern region of China should encourage the transformation of high-energy-consuming industries to low-energy-consuming and high-value-added industries, pay attention to the application of green technologies such as clean energy while upgrading their intelligence, and promote the optimization of the industrial chain. The central region should coordinate economic development and environmental protection, strengthen the cultivation of intelligent talents, further promote the application of intelligent technology in the manufacturing process, and achieve the efficient use of energy. The western region should strengthen the construction of information infrastructure and transportation logistics facilities, strengthen cooperation with the eastern and central regions, introduce intelligent technology and management experience, and develop characteristic industries such as an ecological manufacturing industry according to its own resource endowment and environmental characteristics. Secondly, the degree of agglomeration is heterogeneous. Different regions in China are different in resource endowment, policy orientation, human capital, and other aspects, so that the level of EA is different. Therefore, from the perspective of EA as an intermediary variable for IM to improve CEP, high-EA areas should optimize resource allocation through the combination of policy guidance and market mechanisms, support the development of a circular economy and green manufacturing, and avoid the waste of resources. Low-EA areas should improve their logistics operation system, reduce logistics costs, and promote the agglomeration of talents, capital, technology, and other factors to the manufacturing industry through policy guidance; at the same time, each region should improve the regional coordination mechanism to promote resource sharing and complementary advantages among different regions.

7. Conclusions

7.1. Main Conclusions

This study explores the impact of IM on CEP from a theoretical perspective. Utilizing panel data from 30 provinces (including autonomous regions and municipalities directly under the central government) in mainland China from 2010 to 2022, a comprehensive index system was constructed to measure the levels of IM and EA. The network DEA–Malmquist index model was employed to assess the TFCEP. The direct and indirect effects of IM on CEP were empirically tested using fixed effect and mediating effect models. The conclusions are as follows. (1) The intellectualization of the manufacturing industry significantly promotes the improvement of CEP. After a series of robustness tests, including variable replacement, the exclusion of extreme values, and the use of instrumental variables, this conclusion remains valid, thereby verifying H1. (2) From the perspective of impact mechanisms, IM indirectly enhances CEP by fostering EA, with EA playing a partial mediating role, thereby verifying H2. (3) The impact of IM on CEP exhibits heterogeneity based on regional geographical location, EA, and whether the region is a low-carbon pilot province. In the central and eastern regions, regions with a low EA, and non-low-carbon pilot regions, the promotion effect of intelligence on CEP is more pronounced.

7.2. Limitations and Future Study

As China is the largest manufacturing country and the largest emitter of greenhouse gases, studying the impact of IM on CEP is not only crucial for its economic development and environmental improvement but also of significant importance for the global achievement of dual carbon goals, the implementation of the Paris Agreement, and the promotion of sustainable development. The limitations of this paper are as follows. Firstly, in terms of research indicators, while we have integrated the existing research and introduced innovations, the index system retains a degree of subjectivity and requires further refinement. For instance, when considering the intelligent index system for the manufacturing industry developed by Liu et al. (2021) [33], we encountered a shortfall in our data collection, specifically the absence of information on high-tech manufacturing practitioners in the fundamental input section. Utilizing a substitute proxy variable for the index could potentially lead to discrepancies in the measured level of intelligence. To minimize the impact of such potential deviations, we conducted a comparative analysis of this index with those used in other studies. Ultimately, we opted to employ Yang’s (2024) metric for the intelligent human resource input level as a representative index for this paper [31]. This decision was made in an effort to mitigate the subjectivity associated with the proxy variables used for the alternative index. Secondly, in terms of sample data, this paper utilizes panel data from China’s provincial manufacturing industry for empirical analysis. On the one hand, due to constraints in data availability, the scope of this study is confined to China’s provincial manufacturing sector, precluding an in-depth examination of the micro-level impacts of intellectualization on CEP in prefecture-level cities and enterprises. The potential for data discrepancies and gaps in certain aspects may lead to variations in the research findings. For instance, micro-level studies might reveal a nonlinear relationship between EA and CEP, rather than a straightforward promotional effect. To address this, we have initiated the collection of micro-level data from other industries and are committed to achieving breakthroughs in this domain in the future. On the other hand, the current study does not encompass research on manufacturing industries in other countries, resulting in sample data that lack the breadth and depth necessary for a comprehensive analysis. Thirdly, in terms of the impact path, we have focused our analysis on the relationship between EA, IM, and CEP, without extending our scope to encompass a range of other potential influencing factors. These include micro-level factors such as the aggregation of skilled labor, the establishment of information infrastructure, and the effects of green technology innovation, as well as macro-level considerations like policy interventions, socio-economic conditions, and political contexts. Additionally, our analysis does not fully account for the multi-level dynamics of green infrastructure development, technological innovation, technological diffusion, and human capital development, which are critical for a comprehensive understanding of the issue. To augment the current research and address these gaps, we are actively compiling data that include mediating variables such as socio-economic factors, green infrastructure, and human capital development. Our aim is to integrate these variables into a more extensive analysis in future studies, thereby providing a more thorough and multidimensional examination of the factors that influence CEP in the context of IM and EA.

In future research, we will continue to enrich and refine the measurement indicators of each variable based on the existing studies and expand the breadth and depth of our research. We will investigate the carbon emission reduction effect of China’s manufacturing industry at the micro level, enhance the analysis of impact pathways, and extend our focus globally to analyze the impact of IM on energy conservation and emission reduction in various countries. This will contribute richer theoretical insights into achieving global net zero emissions and mitigating the global warming crisis.