Can Intelligent Equipment Optimization Improve the Carbon Emissions Efficiency of the Equipment-Manufacturing Industry?

Abstract

China’s equipment-manufacturing industry accounts for a significant portion of its total carbon emissions. While intelligent equipment optimization has been found to be an effective way of reducing carbon emissions, understanding of its mechanisms remains limited. This paper takes the equipment-manufacturing industry as an example to explore the mechanisms and pathways for enhancing carbon emissions efficiency through intelligent equipment optimization. Using panel data from 243 equipment-manufacturing firms, the analysis identified a nonlinear, U-shaped relationship between intelligent equipment upgrades and carbon emissions efficiency. At the initial stage of intelligent upgrading of equipment, efficiency declines due to the high capital expenditures required for upgrading and integrating advanced systems. However, as these technologies become more integrated into production processes, carbon emissions efficiency improves significantly. This study also examines the mediating role of cost-saving effects and the moderating influence of energy intensity in this relationship. The effect of intelligent transformation on improving carbon emissions efficiency is more significant in high-energy-intensity enterprises. The findings suggest that intelligent equipment optimization not only enhances resource-utilization efficiency but also supports green and low-carbon transitions in equipment-manufacturing enterprises. These insights offer valuable guidance for policymakers and industry leaders aiming to further integrate intelligent manufacturing with carbon reduction strategies.

1. Introduction

As a pillar of the national economy, industry is not only a major area of energy consumption and carbon emissions, but also the key to realizing energy conservation and carbon reduction and promoting a transformation to a green and low-carbon economy. In the face of the severe challenges posed by global climate change, accelerating the pace of industrial energy conservation and carbon reduction is of great significance to building an ecological civilization and achieving high-quality economic development. With the development of industry, environmental problems become more prominent. Carbon emissions are one of the prominent environmental problems. China put forward China’s carbon emissions-reduction targets at the World Climate Conference. Theory and practice show that as the industrial sector tends toward technological innovation, industrial intelligence will become an important factor affecting carbon emissions; we will thus explore the impact of industrial intelligence on carbon emissions. China’s “Dual Carbon” targets—the aims of reaching peak carbon emissions by 2030 and achieving carbon neutrality by 2060—highlight the urgency of transitioning to greener practices, particularly in the equipment-manufacturing industry, which is a major contributor to energy consumption and carbon emissions [1]. For this reason, this paper takes the equipment-manufacturing industry as an example to explore the impact of intelligent technologies on the efficiency of carbon emissions.

The equipment manufacturing sector, given its significant carbon footprint, has been the focus of numerous studies aimed at improving carbon emissions efficiency. Research shows that methods for measuring carbon emissions efficiency can be categorized into single-factor and total-factor approaches. Single-factor methods calculate carbon productivity as the ratio of carbon emissions to GDP [2]. Total-factor approaches, such as stochastic frontier analysis (SFA) and data envelopment analysis (DEA), offer a more comprehensive view [3,4]. Additionally, the slack-based measure (SBM) model, which accounts for undesirable outputs, has been proposed as a more accurate method for assessing the impact of emissions on efficiency [5,6]. However, despite the advancements in methodology, there is growing recognition that achieving substantial improvements in carbon efficiency requires systemic shifts in production models, particularly through the lenses of intelligent transformation and the circular economy [7].

Intelligent transformation, driven by technologies such as the Internet of Things (IoT), Big Data, and artificial intelligence (AI), is widely regarded as essential for enhancing resource efficiency and reducing carbon emissions. These technologies enable real-time monitoring of energy consumption, predictive maintenance, and optimized resource allocation [8]. Intelligent technologies also reduce waste generation by enabling closed-loop resource management, aligning with the principles of the circular economy [9,10]. Prior research suggests that intelligent transformation not only improves resource-utilization efficiency but also allows firms to manage energy use more effectively, reducing energy intensity and improving carbon efficiency [11]. Additionally, intelligent transformation has been shown to help mitigate rising labor costs and enhance global competitiveness, contributing to reshoring efforts in medium- and high-end manufacturing sectors [12].

Despite these advancements, several research gaps remain. Few studies have explored the U-shaped relationship between intelligent equipment upgrades and carbon emissions efficiency, where initial investments may lead to temporary declines in efficiency due to high capital costs before delivering long-term improvements [13]. Additionally, the literature often overlooks mediating mechanisms, such as cost-saving effects, and moderating factors, such as energy intensity. These gaps limit our understanding of how intelligent transformation in alignment with the principles of the circular economy can enhance carbon efficiency and reduce overall environmental impact.

After a careful review of the existing literature, we found there is scant literature on the theoretical analysis and empirical testing of the effectiveness of intelligent production in reducing carbon emissions from the standpoint of intelligent equipment optimization; moreover, the current empirical research on intelligentization-enabled carbon emissions reduction is mainly focused on the industry level, and there are few published works on intelligentization-enabled low-carbon development from the enterprise level. Furthermore, previous studies mainly focus on the whole industry or the manufacturing industry, and there is a lack of research on the carbon emissions efficiency of the equipment-manufacturing industry. Moreover, the existing research that links intelligent production with sustainable manufacturing needs to be expanded.

To address the research gap, our research used panel data from the equipment-manufacturing industry from 2006 to 2019 to investigate the impact and mechanism of enterprise-level intelligent optimization of equipment on carbon emissions efficiency, and we make contributions to three areas. Firstly, the existing literature mainly studies the linear impact of intelligent technology on carbon emissions, while this article further explores the U-shaped impact that the optimization of intelligent equipment may have on the carbon emissions efficiency of the equipment-manufacturing industry. Secondly, other scholars are mainly concerned about the direct impact of intelligent technology on enterprises’ carbon emissions. This study reveals the mediating role of cost savings in increasing carbon emissions efficiency through intelligent optimization of equipment. Thirdly, the indirect effects of enterprises’ environmental impacts have not been taken into account in existing intelligent research. This study analyzes the mechanisms by which enterprises’ energy intensity regulates improvements in carbon emissions efficiency through intelligent equipment optimization in order to determine whether enterprises with different energy intensities have varying degrees of improvement in carbon efficiency after upgrading their intelligent equipment.

This research will contribute to the literature by providing new insights into how intelligent transformation, mediated by cost-saving mechanisms and moderated by energy intensity, affects carbon emissions efficiency in the equipment-manufacturing industry. By explicitly incorporating the principles of the circular economy, this study will offer a comprehensive understanding of how intelligent technologies support sustainable production and enhance resource efficiency. The findings will provide valuable guidance for policymakers and industry leaders aiming to achieve China’s “Dual Carbon” goals and drive the global shift towards low-carbon manufacturing.

2. Literature Review and Hypothesis Development

2.1. The Level of Intelligent Equipment Upgrades and the Carbon Emissions Efficiency of Equipment-Manufacturing Enterprises

Intelligent equipment upgrades and carbon emissions efficiency have become critical topics in sustainable manufacturing. Research has shown that intelligent equipment upgrades can have a significant impact on carbon emissions efficiency in manufacturing [14,15,16,17]. Scholars point out that this influence can be divided into two different effects that occur in different periods [18]. Early phases of intelligent technology adoption often come with high costs related to equipment upgrades, system integration, and employee training, which tend to lower operational efficiency temporarily. This decrease in efficiency is a common occurrence in the initial stages of industrial transformations, where large upfront investments may lead to short-term increases in energy consumption and emissions. On the other hand, the development of intelligent technologies is transforming traditional industries. The earliest industries to undergo transformation were those with high pollution and energy consumption. They urgently need to introduce intelligent technologies to improve their production capacity, which will bring about a large energy demand and an increase in carbon emissions [19]. However, as these technologies become more mature and fully integrated, improvements in efficiency typically follow, driven by optimized resource use and waste reduction, leading to enhanced carbon emissions efficiency. This progression is consistent with “circular economy principles”, which emphasize the importance of resource recovery and “closed-loop management” for reducing emissions [20].

Existing research mainly focuses on macro-level analysis of industries or regions, rather than investigating the specific mechanisms acting within enterprises in key industries. In addition, the current literature mainly studies the single impact of intelligent technology on carbon emissions. Therefore, based on the characteristics of the manufacturing industry, this study examines the mechanism by which the optimization of equipment intelligence affects the carbon emissions efficiency of the equipment-manufacturing industry.

Therefore, we argue that the optimization of intelligent equipment has a U-shaped impact on carbon emissions. Specifically, in the initial stages of implementing intelligent equipment upgrades, substantial upfront investments in new technologies and workforce training can result in a temporary decline in carbon emissions efficiency. This early-stage inefficiency is a well-documented phenomenon during technological transitions and is largely driven by the high costs of capital and integration. As intelligent equipment becomes more integrated into production systems, significant improvements in carbon efficiency are typically observed, largely due to more effective resource management and reductions in waste [21]. “Closed-loop resource management” strategies, which emphasize material reuse and minimizing waste, play a key role in enhancing carbon efficiency as intelligent technologies mature [22]. This development supports the argument for a “nonlinear, U-shaped relationship” between intelligent equipment upgrades and carbon emissions efficiency, where early declines give way to substantial long-term gains as the technology matures.

H1: 

There is a “nonlinear U-shaped relationship” between the level of intelligent equipment upgrades and the carbon emissions efficiency of equipment-manufacturing enterprises.

2.2. Mediating Effect: Cost-Saving Mechanism

Intelligent transformation in manufacturing, particularly in the equipment sector, requires significant investments in capital, technology, and employee training. Automation technologies, like industrial robots, enhance productivity by optimizing processes and reducing reliance on low-skilled labor [23,24,25]. However, these cost-saving benefits tend to emerge only after initial investments have been absorbed, delaying their impact on carbon emissions efficiency. Scholars have pointed out that intelligent technology also has a non-linear impact on costs [26], which manifests as a positive increase or negative decrease in enterprise costs in the early or later stages of intelligent upgrading, contributing to a non-linear “cost-saving effect” of intelligent upgrading.

Therefore, we believe that intelligent equipment optimization has a non-linear “cost-saving effect” on enterprise costs, and under this non-linear influence, intelligent technology also has a non-linear impact on carbon emissions. In the initial stage of equipment intelligence optimization, due to these high-cost investments, the operating costs of enterprises significantly increase, often with a short-term negative impact on carbon emissions efficiency. This phenomenon can be explained as a negative “cost-saving effect”, where the high costs brought about by intelligent equipment optimization may initially suppress improvements of carbon emissions efficiency. As technology integration progresses, the cost-saving effect shifts from negative to positive. Early capital expenditures, training, and technology integration increase operational costs, but as systems mature, automation optimizes resource use, reduces waste, and improves efficiency [27]. Intelligent systems help firms transition to closed-loop management, improving both economic and environmental outcomes [28]. This progression highlights the mediating role of cost-saving mechanisms in enhancing carbon efficiency over time.

H2: 

The cost-saving effect mediates the relationship between intelligent transformation and carbon emissions efficiency. Early investments lead to an initial decline, but as the transformation progresses, the cost-saving effect enhances carbon emissions efficiency through resource optimization and reuse.

2.3. Moderating Effect: Energy Intensity as a Mechanism

Energy intensity, defined as energy consumption per unit of output, is a critical measure of an enterprise’s energy efficiency [29]. In the equipment-manufacturing industry, energy intensity not only affects carbon emissions but also moderates the relationship between intelligent transformation and carbon emissions efficiency [30]. Firms with lower energy intensity typically exhibit high baseline energy efficiency, meaning the improvements from intelligent transformation may be less significant [31,32]. On the other hand, enterprises with higher energy intensity have more room for improvement. By optimizing processes and upgrading equipment, these firms can significantly reduce energy consumption and emissions [11,33,34].

Therefore, we believe that the level of energy intensity may affect the energy-saving and emission-reduction effects brought about by intelligent optimization and further affect the improvement of carbon emissions efficiency in equipment-manufacturing enterprises. Specifically, for low-energy-intensity firms, intelligent transformation optimizes production efficiency but offers limited improvement in carbon emissions efficiency due to their already-efficient energy use. This reflects a key principle of the circular economy, where technological advancements have diminishing returns as resource use approaches optimal levels [35]. Conversely, high-energy-intensity firms see greater benefits from intelligent transformation [36]. The introduction of intelligent control systems and optimized production processes not only cuts energy consumption and emissions but also supports the goals of the circular economy by reducing waste and improving resource efficiency [37]. As a result, intelligent transformation is more effective in high-energy-intensity firms.

H3: 

Energy intensity moderates the relationship between intelligent transformation and carbon emissions efficiency. The positive impact of intelligent transformation is stronger in high-energy-intensity firms, while in low-energy-intensity firms, the effect is less pronounced.

3. Research Design

3.1. Model Construction

To verify the mechanisms by which the level of intelligent transformation, cost-saving effect, and energy intensity influence the carbon emissions efficiency of equipment-manufacturing enterprises, several regression models were established in this study. They are given below.

$${{\displaystyle CEEE}}_{i,t}={{\displaystyle \beta }}_0+{{\displaystyle \beta }}_1{{\displaystyle LIR}}_{i,t}+{{\displaystyle \beta }}_2{{\displaystyle LIR}}_{i,t}^2+{\displaystyle \sum {{\displaystyle \beta }}_3}{{\displaystyle control}}_{i,t}+{{\displaystyle \mu }}_t{{\displaystyle +\delta }}_t+{{\displaystyle \epsilon }}_{i,t}$$

(1)

In Model (1), ${{\displaystyle CEEE}}_{i,t}$ represents the carbon emissions efficiency of equipment-manufacturing enterprise t in period I; ${{\displaystyle LIR}}_{i,t}$ represents the level of intelligent transformation of enterprise t in period I; and ${{\displaystyle LIR}}_{i,t}^2$ is the squared term of the intelligent transformation level for enterprise t in period i. ${\displaystyle \sum {{\displaystyle \beta }}_3}{{\displaystyle control}}_{i,t}$ denotes a series of control variables, ${{\displaystyle \mu }}_i$ represents the time-invariant firm fixed effects for enterprise t, and ${{\displaystyle \delta }}_t$ controls for time fixed effects. Lastly, ${{\displaystyle \epsilon }}_{i,t}$ represents the random error term. This model was used to test the nonlinear, U-shaped relationship between intelligent transformation and carbon emissions efficiency.

In addition to the direct effect presented in Equation (1), this paper adopted the methods of Mei et al. [38]. To further explore how intelligent equipment upgrades influence carbon emissions efficiency through the cost-saving effect, a mediation-effect model was constructed.

$${{\displaystyle CER}}_{i,t}={{\displaystyle \alpha }}_0+{{\displaystyle \alpha }}_1{{\displaystyle LIR}}_{i,t}+{{\displaystyle \alpha }}_2{{\displaystyle LIR}}_{i,t}^2+{\displaystyle \sum {{\displaystyle \alpha }}_3}{{\displaystyle control}}_{i,t}+{{\displaystyle \mu }}_t{{\displaystyle +\delta }}_t+{{\displaystyle \epsilon }}_{i,t}$$

(2)

$${{\displaystyle CEEE}}_{i,t}={{\displaystyle \beta }}_0+{{\displaystyle {{\displaystyle \beta }}_1{{\displaystyle CER}}_{i,t}+\beta }}_2{{\displaystyle LIR}}_{i,t}+{{\displaystyle {{\displaystyle \beta }}_{3}LIR}}_{i,t}^2+{\displaystyle \sum {{\displaystyle \beta }}_4}{{\displaystyle control}}_{i,t}+{{\displaystyle \mu }}_t{{\displaystyle +\delta }}_t+{{\displaystyle \epsilon }}_{i,t}$$

(3)

In models (2) and (3), ${{\displaystyle CER}}_{i,t}$ represents the cost-saving effect of equipment-manufacturing enterprises at time t in period i. Model (2) was used to examine the impact of the intelligent equipment upgrade level, represented by LIR and LIR², on the cost-saving effect CER, while model (3) tests the effects of LIR and LIR² on the carbon emissions efficiency (CEEE) of equipment-manufacturing enterprises by controlling for the cost-saving effect CER. The existence of a mediation effect can be determined by the significance of regression coefficients α_₁, α_₂, β_₁, β_₂, and β_₃.

Finally, to further explore the moderating effect of energy intensity on the relationship between intelligent equipment upgrades and carbon emissions efficiency in equipment-manufacturing enterprises, a term representing the interaction between energy intensity and the level of intelligent upgrades was introduced into Equation (1), extending it into the following model:

$${{\displaystyle CEEE}}_{i,t}={{\displaystyle \beta }}_0+{{\displaystyle {{\displaystyle \beta }}_1{{\displaystyle EI}}_{i,t}+\beta }}_2{{\displaystyle EI\times LIR}}_{i,t}+{{\displaystyle {{\displaystyle \beta }}_{3}EI\times LIR}}_{i,t}^2{{\displaystyle +LIR}}_{i,t}+{{\displaystyle {{\displaystyle \beta }}_{3}LIR}}_{i,t}^2+{\displaystyle \sum {{\displaystyle \beta }}_6}{{\displaystyle control}}_{i,t}+{{\displaystyle \mu }}_t{{\displaystyle +\delta }}_t+{{\displaystyle \epsilon }}_{i,t}$$

(4)

In model (4), the interaction terms ${{\displaystyle EI\times LIR}}_{i,t}$ and ${{\displaystyle EI\times LIR}}_{i,t}^2$ represent the energy intensity and the level of intelligent equipment upgrades for equipment-manufacturing enterprises at time t in period i.

3.2. Measurement and Description of Variables

3.2.1. Measurement of Carbon Emissions Efficiency in Equipment-Manufacturing Enterprises

Based on the industry classification from the “National Economic Industry Classification” (GB/T 4754—2017), this study selected enterprises in the equipment-manufacturing industry as the research subjects. The equipment-manufacturing industry includes eight sectors: metal products; general equipment; specialized equipment; automobiles; railway, shipbuilding, aerospace, and other transportation equipment; electrical machinery and apparatus; and instruments and meters.

Total-factor carbon emissions efficiency is more comprehensive than single-factor carbon emissions efficiency. It includes various factors such as economic growth, resource-based and non-resource-based inputs, and energy carbon emissions. As a result, it can more accurately assess the overall efficiency of economic activities. The data envelopment analysis (DEA) method constructs the production frontier of decision-making units (DMUs) using linear programming techniques. It measures the technical efficiency of each unit based on its distance from the frontier and has been widely applied in measuring total-factor carbon emissions efficiency. However, traditional DEA models are mostly based on radial and angular characteristics. They do not fully consider slack in inputs and outputs, which leads to deviations between evaluation results and actual economic conditions, thus affecting the accuracy of carbon emissions-efficiency assessments. To address this issue, the SBM model [39], which handles the slack of undesirable outputs, provides a more accurate reflection of carbon emissions efficiency in real economic environments. On this basis, the super-efficiency SBM model further enhances the ranking ability of decision-making-unit efficiency values, significantly improving the accuracy of the evaluation. This paper used the super-efficiency SBM model based on undesirable outputs to measure industrial carbon emissions efficiency, as it provides a more precise reflection of efficiency values when undesirable outputs are present. The specific model is as follows:

$$\begin{array}{l}\min \rho ={\displaystyle \frac{1+\frac{1}{m}{\displaystyle \sum _{i=1}^m\frac{s_i^{-}}{x_{i0}}}}{1-\frac{1}{q_1+q_2}\left[{\displaystyle \sum _{r=1}^{q_1}\frac{s_r^{+}}{y_{r0}}}+{\displaystyle \sum _{i=1}^{q_2}\frac{s_i^{-}}{b_{r0}}}\right]}}\\ \mathsf{s}\mathbf{.}\mathsf{t}\mathbf{.}\left\{\begin{array}{ll}{\displaystyle \sum _{j=1}^n{x}_j}{\lambda }_j-s^{-}\le x_{i0},\hfill & i=1,\dots ,m\hfill \\ {\displaystyle \sum _{j=1}^n{y}_j}{\lambda }_j+s^{+}=y_{r0},\hfill & r=1,\dots ,q_1\hfill \\ {\displaystyle \sum _{j=1}^n{b}_j}{\lambda }_j-s^{-}\le b_{r0},\hfill & r=1,\dots ,q_1\hfill \\ {\lambda }_j,s^{-},s^{+},s_r^{+},s_b^{-}\ge 0,\hfill & j=1,\dots ,n,j\ne j_0\hfill \end{array}\right.\end{array}$$

(5)

This model assumes there are m decision-making units (DMUs). $\rho$ represents the carbon emissions-efficiency value of equipment-manufacturing enterprises. $\rho$, $q_1$ and $q_2$ represent the numbers of inputs, desirable outputs, and undesirable outputs, respectively. $x_{i0}$ is the input matrix. $y_{r0}$ is the desirable output matrix. $b_{r0}$ is the undesirable output matrix. $s^{-}$, $s^{+}$ and $s_b^{-}$ are the slack variables for desirable and undesirable outputs. ${\lambda }_j$ represents the weight assigned to different decision-making units.

In calculating the carbon emissions efficiency of equipment-manufacturing enterprises, this study uses the number of employees, net fixed assets, and total energy consumption as input indicators. It uses operating income as the desirable output indicator and total carbon emissions as the undesirable output indicator.

The calculation of carbon dioxide emissions at the level of the equipment-manufacturing industry was based on work in a study by Shao et al. [40]. It included data on the consumption of seven types of fossil energy: coal, coke, fuel oil, gasoline, kerosene, diesel, and natural gas. The carbon emissions from energy consumption were calculated based on the carbon emission factors provided by IPCC (2006).

Since the government does not mandate companies, including listed companies, to disclose their carbon emissions and energy consumption data, it is currently difficult to directly obtain relevant data from equipment-manufacturing enterprises. In the process of evaluating the carbon emissions and energy consumption of enterprises in this paper, it was found that few companies voluntarily disclose such data in their annual reports, limiting the availability of firm-level microdata. To address this issue, this paper adopted the method used in Wang’s [41] study, estimating the carbon emissions and energy consumption of enterprises based on the proportion of operating costs and thus compensating for the data shortage. The calculation formula is as follows:

$$C=\left({\displaystyle \frac{Operating costs}{Cost of main business}}\times Carbon emissions\right)/Operating income$$

(6)

$$E=\left({\displaystyle \frac{Operating costs}{\mathrm{Cos}t of main business}}\times Energy consumption\right)/Operating income$$

(7)

3.2.2. Level of Intelligent Equipment Upgrades

Referring to Jia et al. [42], the level of intelligent equipment upgrades was measured by robot density, which is defined as the number of robots installed per 10,000 employees. Based on the industrial robot statistics provided by IFR, this paper further estimated the robot density at the firm level. Specifically, the calculation was performed as follows:

$$R{D}_t={\displaystyle \frac{IF{R}_t}{AC{U}_t}}$$

(8)

Here, $R{D}_t$ represents the robot density of the equipment-manufacturing industry in year t, $IF{R}_t$ is the number of robots installed in the equipment-manufacturing industry in year t, and $AC{U}_t$ is the average number of employees in the equipment-manufacturing industry in year t.

Based on the calculated industrial robot density of the equipment-manufacturing industry, this paper calculated the firm-level industrial robot penetration rate as follows:

$$LI{R}_{i,t}={\displaystyle \frac{S{N}_{i,t}}{ManuS{N}_t}}\times R{D}_{i,t}$$

(9)

$S{N}_{i,t}$ represents the number of employees in company j of industry i, and $ManuS{N}_t$ represents the median number of employees of all companies in year t. The ratio between the two can measure the position and significance of a listed company within the industry.

3.2.3. Control Variables

Referring to the existing literature, we used several control variables that may potentially have influenced our analysis results: company size (Size), leverage ratio (Lev), revenue growth rate (Growth), company age (Age), return on equity (ROE), ownership concentration (Top1), and board size (Board) [43,44]. The definitions and calculation methods of the main variables are detailed in

Table 1. Variable definitions.

	Variable Name	Variable Symbol	Variable Description
Dependent Variable	Carbon Emissions Efficiency of Equipment-Manufacturing Enterprises	CEEE	Carbon emissions efficiency, calculated using the super-efficiency SBM model.
Independent Variable	Level of Intelligent Equipment Upgrades	LIR	Robot density at the firm level, calculated by formulas (8) and (9).
Mediating Variable	Cost-Saving Effect	CER	Ln (total operating costs).
Moderating Variable	Energy Intensity	EI	Energy consumption of the enterprise/operating income.
Control Variables	Company Age	Age	Observation year − year of establishment + 1.
	Company Size	Size	Ln (total assets).
	Return on Equity	Roe	Net profit/shareholders’ equity balance.
	Leverage Ratio	Lev	Total liabilities/total assets.
	Revenue Growth Rate	Growth	(Operating income for this quarter–operating income for the previous quarter)/(Operating income for the previous quarter).
	Ownership Concentration	Top1	Shareholding ratio of the largest shareholder.
	Board Size	Board	Ln (number of board members).
	Firm Dummy Variable	Industry	243 firms involved, generating 243 dummy variables.
	Year Dummy Variable	Year	14 years involved, generating 14 dummy variables.

.

This study empirically examined the impact of intelligent equipment upgrades on the carbon emissions of equipment-manufacturing enterprises and the mechanisms driving these impacts. The study is based on data from A-share listed equipment-manufacturing companies on the Shanghai and Shenzhen stock exchanges from 2006 to 2019. The data in this paper come from the following sources: (1) industrial robot data were sourced from the International Federation of Robotics (IFR). This organization surveys global robot manufacturers and provides the most authoritative statistics on robot applications to date [45,46,47]. (2) Corporate operating data were obtained from the CSMAR database. The database covers China’s capital market, macro-economy, industry, company, and other dimensions, including the financial data and governance data of listed companies required by this paper. (3) Industry-level data for the equipment manufacturing sector came from the “China Statistical Yearbook” and the “China Energy Statistical Yearbook”.

In order to reduce statistical errors, improve the effectiveness of statistical results, and eliminate interference from abnormal values, the sample companies were selected based on the following criteria, excluding: (1) listed companies marked as ST or *ST during the observation period; (2) samples with missing key variables. After filtering and processing, a total of 3402 research samples were obtained.

4. Results

4.1. Descriptive Statistical Results

Table 2. Descriptive statistics.

	Variable	Observations	Mean	Standard Deviation	Min	Max
Dependent Variable	CEEE	3402	0.125	0.197	−0.065	1.880
Independent Variable	LIR	3402	0.0113	0.0357	1.22 × 10−6	0.951
Control Variables	Age	3402	22.190	1.316	18.157	27.468
	Size	3402	2.782	0.389	1.099	4.143
	Roe	3402	−0.017	1.854	−76.764	0.938
	Lev	3402	0.501	0.179	0.044	0.995
	Growth	3402	0.262	1.891	−0.955	58.487
	Top1	3402	34.181	14.195	5.278	83.827
	Board	3402	8.916	1.698	2.000	19.000

provides a summary of the descriptive statistics, showing variability across key variables, including carbon emissions efficiency (CEEE), firm age, size, leverage, and growth.

4.2. Baseline Regression Results

Table 3. Baseline regression results for intelligent equipment upgrades and carbon emissions efficiency in equipment-manufacturing enterprises.

Variable	CEEE
	(1)	(2)	(3)	(4)
LIR	−0.232 *** (−3.29)	−0.525 *** (−4.03)	−0.35 *** (−4.99)	−0.870 *** (−6.54)
LIR2		0.507 ** (2.67)		0.873 *** (4.58)
Age			0.033 *** (7.70)	0.036 *** (8.41)
Size			0.021 (0.80)	0.014 *** (0.53)
Roe			0.001 (0.85)	0.001 (0.90)
Lev			0.067 *** (3.39)	0.070 *** (3.55)
Growth			0.0001 (0.24)	0.0001 (0.12)
Top1			0.0006 ** (2.29)	0.0006 ** (2.29)
Board			−0.002 (−1.52)	−0.003 (−1.56)
cons	0.422 *** (14.79)	0.455 *** (15.05)	−0.422 *** (−3.64)	−0.468 *** (−4.04)
Year	YES	YES	YES	YES
Industry	YES	YES	YES	YES
R2	0.726	0.727	0.735	0.7368
F	32.61	32.57	33.10	33.26
Observations	3402	3402	3402	3402

Note: (1) The values below the coefficients are t-values. (2) *** and ** indicate that the variable coefficients have passed significance tests at the 1% and 5% levels, respectively.

reports the regression results for the impact of intelligent equipment upgrades on the carbon emissions efficiency of equipment-manufacturing enterprises. As shown in columns (1) and (2), after controlling for firm fixed effects and year fixed effects, the coefficient of LIR was −0.870 and significantly negative, while the coefficient of LIR² was 0.873 and significantly positive. This indicates a U-shaped relationship between intelligent equipment upgrades and carbon emissions efficiency, suggesting that upgrades initially have a negative impact on carbon emissions efficiency but that as the upgrades become more significant, the impact gradually turns positive. As shown in columns (3) to (4), after control variables such as firm size and company age were introduced, the U-shaped effect remained significant, confirming hypothesis H₁.

The regression results show that the coefficient of LIR was −0.870 (p < 0.01), while the coefficient of LIR² was 0.873 (p < 0.01),which indicates that the negative impact of intelligent upgrades on carbon emissions efficiency is significant in the initial phase. Within the framework of the circular economy, the initial negative impact of intelligent upgrades reflects the temporary inefficiency experienced by equipment-manufacturing enterprises when implementing intelligent technologies due to early capital investments and the complexity of technology integration; this leads to a decrease in carbon emissions efficiency. This negative effect is linked to the low resource-utilization efficiency at the early stages, which indicates that closed-loop resource management has not yet been fully achieved, prevents the cost-saving effect from materializing, and results in higher-than-expected resource consumption.

The advancement of intelligent upgrades is reflected in the regression, with results showing that the coefficient of LIR² was 0.873 (p < 0.01). As shown in Figure 1, When the LIR value reaches approximately 0.498, carbon emissions efficiency starts to turn around and shows significant improvement, indicating that after the lowest point of the curve, the maturity of the technology and improvements in operational efficiency begin to generate positive effects in the middle and later stages of the upgrades, significantly enhancing carbon emissions efficiency. This aligns with the resource-optimization concept of the circular economy: when intelligent technologies enable real-time monitoring and feedback on resources, enterprises gradually adopt a closed-loop resource-management model, reducing energy waste and resource consumption and thereby improving carbon emissions efficiency.

As shown in

Table 4. U-test results.

	Lower Bound	Upper Bound
Range of Values	1.22 × 10−6	0.9514
Slope	−0.87	0.79
t-value	−4.845	1.990
Significance	0.000	0.000

Note: U-shaped test result: t = 1.99, p > t = 0.02.

, to test the U-shaped relationship between LIR and CEEE, this paper used the U-test command and verified three necessary conditions: first, the estimated inflection point falls within the range of values for the independent variable; second, the slopes on both sides of the curve have opposite signs; third, the slopes are overall significant. The test results in column (4) indicate a significant U-shaped relationship between LIR and CEEE. First, the extreme point of 0.498 (as shown in Figure 1) lies within the range of values for the independent variable [1.22 × 10⁻⁶, 0.9514] and the slopes have opposite signs within the interval (the left slope is −0.87, and the right slope is 0.79). Both slopes are significant at the 1% statistical level (t-values of −4.845 and 1.990, with p-values of 0.000. This result confirms a valid U-shaped relationship between LIR and carbon emissions efficiency.

As shown from the results of the U-shaped test on LIR and its squared term, hypothesis H₁ was further validated, which suggests that within the framework of the circular economy, intelligent upgrades initially bring negative effects, but as the upgrades progress, carbon emissions efficiency significantly improves. This outcome aligns with the core concept of the circular economy, which emphasizes improving resource-utilization efficiency through technological advancement.

As shown in Figure 1, there is a clear U-shaped relationship between LIR and CEEE, which further supports this study’s hypothesis H₁. In the early stages of intelligent upgrades, the high costs and technological adaptation challenges may lead to a decline in carbon emissions efficiency for equipment-manufacturing enterprises. However, as the upgrades progress, companies gradually overcome these initial challenges and carbon emissions efficiency begins to recover and significantly improve. Notably, when the LIR value reaches approximately 0.498, carbon emissions efficiency starts to turn around and shows significant improvement. This visualization underscores the necessity of continuously advancing intelligent upgrades to ensure the long-term optimization of carbon emissions efficiency for enterprises.

This U-shaped relationship aligns closely with the core concept of the circular economy. The circular economy emphasizes optimizing resource efficiency through closed-loop resource management, waste minimization, and resource reuse. In the early stages of intelligent upgrades, due to high costs and the complexity of technology integration, companies may experience a decline in resource-utilization efficiency, leading to a temporary decrease in carbon emissions efficiency. This reflects the high initial investment and low returns characteristic of the early implementation phase of the circular economy.

However, as intelligent upgrades progress, companies introduce smart control systems and optimize production processes to reduce energy consumption and waste emissions, thereby improving resource-utilization efficiency. This transition reflects the principles of efficient resource use and energy minimization in the circular economy. Intelligent upgrades not only enhance carbon emissions efficiency but also promote resource reuse and waste management, fostering the comprehensive implementation of the circular economy.

Therefore, the research results show that intelligent upgrades provide an effective pathway to achieving the goals of the circular economy by improving resource-utilization efficiency and reducing waste. This verifies the positive impact of intelligent upgrades on carbon emissions efficiency and supports the development framework of the circular economy.

4.3. Test for Mediation by the Cost-Saving Effect

To verify the mediating role of the cost-saving effect (CER) between the level of intelligent equipment upgrades (LIR) and the carbon emissions efficiency (CEEE) of equipment-manufacturing enterprises, this paper adopted a stepwise regression method. Based on Model (1), Models (2) and (3) were added to form the mediation-effect model shown below. This model tests the mediating transmission role of the cost-saving effect between the level of intelligent equipment upgrades and carbon emissions efficiency in equipment-manufacturing enterprises.

The detailed results of the mediation-effect regression are shown in

Table 5. Results of test for mediation by the cost-saving effect.

Variable	CER	CEEE
	(1)	(2)
CER		0.817 *** (19.777)
LIR	−0.090 *** (−7.572)	−0.136 *** (−6.106)
LIR2	0.004 *** (3.849)	0.078 *** (4.163)
Size	0.330 *** (27.169)	−0.444 *** (−10.068)
Age	0.012 (1.717)	0.019 (0.385)
Roe	0.007 *** (3.500)	0.003 (0.286)
Lev	0.045 (7.500)	0.025 (1.483)
Growth	−0.004 ** (−2.000)	−0.011 (−1.190)
Top1	0.025 ** (2.500)	0.027 (1.339)
Board	−0.009 (−1.800)	−0.026 (−1.680)
cons	−0.372 (−7.019)	0.053 (−7.019)
Year	YES	YES
Industry	YES	YES
R2	0.845	0.846
F	125.6194	125.6194
Observations	3402	3402

Note: (1) The values below the coefficients are t-values. (2) *** and ** indicate that the variable coefficients have passed significance tests at the 1% and 5% levels, respectively.

.

Column (1) shows the impact of the level of intelligent equipment upgrades (LIR) on the cost-saving effect (CER). The results show that LIR has a significant negative impact on CER (α₁ = −0.090, p < 0.01), while the impact of LIR² is significantly positive (α₂ = 0.004, p < 0.05), indicating a U-shaped relationship between LIR and CER. This means that in the early stages of intelligent upgrades, due to the need for substantial capital investment and complex technology integration, operating costs increase, resulting in a negative effect. However, as LIR reaches a certain level, with technology gradually maturing and operational efficiency improving, the cost-saving effect turns positive, reflecting significant improvements in cost control.

Column (2) shows the relationship between LIR and CEEE after controlling for the cost-saving effect. The results indicate that CER has a significantly positive impact on CEEE (β₁ = 0.817, p < 0.01). At the same time, the direct negative impact of LIR on CEEE remains significant (β₂ = −0.136, p < 0.01) and the positive impact of LIR² also continues (β₃ = 0.078, p < 0.01). These results suggest that the cost-saving effect partially mediates the relationship between intelligent equipment upgrades and carbon emissions efficiency, but the direct impact of intelligent upgrades on carbon emissions efficiency remains significant.

In conclusion, the results of the mediation-effect test in this study support hypothesis H₂. That is, the level of intelligent equipment upgrades indirectly improves carbon emissions efficiency by enhancing the cost-saving effect of enterprises. This finding reveals that intelligent upgrades not only improve resource-utilization efficiency by directly influencing a company’s operational model but also amplify their positive effects within the circular economy framework by optimizing cost structures. Policymakers, when promoting intelligent upgrades in enterprises, should fully consider the mediating role of the cost-saving effect to maximize both carbon emissions efficiency and resource-utilization efficiency, both of which align closely with the sustainability goals of the circular economy.

4.4. Test for a Moderating Effect of Energy Intensity

Hypothesis H₃ proposes that energy intensity may play a significant moderating role in this relationship, one that specifically manifests as a varying impact of intelligent equipment upgrades on carbon emissions efficiency depending on changes in energy intensity. To verify the moderating effect of energy intensity (EI) on the relationship between the level of intelligent equipment upgrades (LIR) and the carbon emissions efficiency (CEEE) of equipment-manufacturing enterprises, this paper conducted an empirical analysis by introducing the interaction terms LIR × EI and LIR² × EI.

First, Model (1) in

Table 6. Results of test for a moderating effect of energy intensity.

Variable	CEEE
	(1)	(2)	(3)
LIR	−0.870 *** (−6.54)	−0.152 *** (−6.42)	−2.436 *** (−5.74)
LIR2	0.873 *** (4.58)	0.006 *** (4.49)	1.160 *** (3.25)
EI		−0.010 (−0.74)	−3.526 *** (−5.90)
LIR×EI			−9.462 *** (−5.23)
LIR2×EI			4.937 *** (3.23)
Size	0.036 *** (8.41)	0.240 *** (7.87)	0.206 *** (6.72)
Age	0.014 *** (0.53)	0.031 (0.59)	0.054 (1.03)
Roe	0.001 (0.90)	0.009 (0.90)	0.007 (0.77)
Lev	0.070 *** (3.55)	0.063 *** (3.55)	0.064 *** (3.62)
Growth	0.0001 (0.12)	0.001 (0.09)	0.002 (0.21)
Top1	0.0006 ** (2.29)	0.047 ** (2.26)	0.052 ** (2.51)
Board	−0.003 (−1.56)	−0.026 (−1.57)	−0.022 (−1.32)
cons	−0.468 *** (−4.04)	1.461 *** (8.29)	0.722 *** (3.33)
Year	YES	YES	YES
Industry	YES	YES	YES
R2	0.73	0.736	0.7387
F	33.26	32.9991	33.1889
Observations	3402	3402	3402

Note: (1) The values below the coefficients are t-values. (2) *** and ** indicate that the variable coefficients have passed significance tests at the 1% and 5% levels, respectively.

shows the direct impact of LIR and LIR² on CEEE. The results indicate that the coefficient of LIR is −0.870 (p < 0.01) and the coefficient of LIR² is 0.873 (p < 0.01), suggesting a significant U-shaped relationship between LIR and CEEE. This means that in the early stages of intelligent upgrades, carbon emissions efficiency decreases as the level of intelligent upgrades increases. However, once LIR exceeds a certain threshold, carbon emissions efficiency begins to improve. This reflects the fact that, after the initial phase of resource consumption and inefficiency, intelligent upgrades gradually enter the closed-loop resource-management mode promoted by the circular economy, ultimately leading to a significant improvement in carbon emissions efficiency.

In Model (2), energy intensity was introduced as a control variable to further examine its moderating effect on the relationship between LIR and CEEE. The results show that after controlling for energy intensity, the negative effect of LIR (−0.152) and the positive effect of LIR² (0.006) remain significant, but their coefficients decrease slightly. This indicates that energy intensity has a relatively small direct impact on carbon emissions efficiency (p > 0.1) but that it does have a moderating effect on the relationship between intelligent upgrades and carbon emissions efficiency. The change in energy intensity causes a shift in the U-shaped curve’s inflection point, suggesting that under high-energy-intensity conditions, the negative impact of intelligent upgrades on carbon emissions efficiency is alleviated earlier and the positive effect appears sooner. This aligns with the core principles of the circular economy, which emphasize optimizing energy use, reducing waste, and accelerating the process of resource reuse.

In Model (3), through the introduction of the interaction terms LIR × EI and LIR² × EI, the moderating effect of energy intensity is further explored. The empirical results show that the coefficient of LIR × EI is −9.462 (p < 0.01) and the coefficient of LIR² × EI is 4.937 (p < 0.01), both of which are significant. This indicates that at the lower stages of intelligent upgrades, higher energy intensity exacerbates initial resource consumption and carbon emissions but also accelerates the improvement in carbon emissions efficiency. As intelligent upgrades progress, the increase in energy intensity further strengthens its positive impact on carbon emissions efficiency. This suggests that through the combination of optimized energy usage and intelligent technologies, enterprises can more quickly enter the closed-loop resource-management mode, maximizing resource-utilization efficiency, reducing waste, and achieving the goals of the circular economy. Therefore, hypothesis H₃ is confirmed.

Figure 2 visually presents the U-shaped relationship between LIR and CEEE and further reveals how this relationship changes under different levels of energy intensity.

The study finds that a significant U-shaped curve exists between LIR and CEEE, and this relationship exhibits different characteristics under varying levels of energy intensity. The results indicate that under low-energy-intensity conditions, the inflection point of the U-shaped curve appears later, suggesting that in low-energy-intensity firms, the positive effects of intelligent upgrades require a higher LIR level to become evident. As energy intensity increases, the inflection point shifts to the left and the curve becomes steeper, indicating that in high-energy-intensity firms, intelligent upgrades can improve carbon emissions efficiency earlier and more significantly. This analysis compares the differences in the U-shaped relationship between LIR and CEEE at different levels of energy intensity (EI). Specifically, the relationship is as follows:

Under low-energy-intensity conditions, the inflection point of the U-shaped curve appears relatively late and the curve is relatively flat. This indicates that in firms with lower energy intensity, the impact of intelligent upgrades on carbon emissions efficiency is milder and LIR needs to reach a higher level for the positive effects to manifest. Specifically, the inflection point occurs at LIR = 1.682, suggesting that in low-energy-intensity firms, intelligent upgrades must reach a higher level for carbon emissions efficiency to shift from negative to positive.

For moderate-energy-intensity conditions (dashed line), as energy intensity increases, the inflection point of the U-shaped curve shifts significantly to the left and the curve becomes steeper. This means that under moderate energy intensity conditions, firms can reduce the negative impact on carbon emissions efficiency at lower LIR levels and achieve efficiency improvements earlier. The inflection point shifts to LIR = 1.218, indicating that with moderate energy intensity, firms can experience a positive effect at lower LIR levels.

For high-energy-intensity conditions (dotted line), the inflection point of the U-shaped curve shifts further to the left compared to medium energy intensity and the curve becomes even steeper. This indicates that in firms with higher energy intensity, the positive effects of intelligent upgrades on carbon emissions efficiency not only appear earlier but also result in a greater efficiency improvement as LIR increases. Specifically, the inflection point shifts further to LIR = 1.065, suggesting that in high-energy-intensity firms, even at lower LIR levels, intelligent upgrades can significantly improve carbon emissions efficiency.

4.5. Robustness Test

To verify the robustness of the empirical results, this study employed multiple methods for robustness checks, including narrowing the sample scope, replacing the independent variable, and using the instrumental variable method. The results are shown in

Table 7. Results of robustness test.

Variable	Narrowing the Sample Scope	Replacing the Independent Variable	Instrumental Variable Method
	(1)	(2)	(3)
LIR	−0.227 *** (−7.908)	−0.584 *** (−9.004)	−0.231 *** (−9.066)
LIR2	0.153 *** (5.268)	0.313 *** (6.339)	0.131 *** (6.339)
Size	0.227 *** (7.331)	0.041 *** (9.298)	0.272 *** (9.298)
Age	−0.028 (−0.437)	0.007 (0.281)	0.015 (0.281)
Roe	0.015 (1.132)	0.001 (0.911)	0.009 (0.911)
Lev	0.078 *** (4.231)	0.074 *** (3.790)	0.067 *** (3.790)
Growth	0.008 (0.852)	0.001 (0.006)	0.001 (0.006)
Top1	0.040 (1.834)	0.001 ** (2.301)	0.048 ** (2.301)
Board	−0.020 (−1.129)	−0.003 (−1.586)	−0.026 (−1.586)
cons	−0.011 *** (−1.328)	−0.515 *** (−4.462)	−0.520 *** (−4.500)
Year	YES	YES	YES
Industry	YES	YES	YES
R2	0.774	0.740	0.740
F	34.636	33.866	33.866
Observations	1816	2284	2284

Note: (1) The values below the coefficients are t-values. (2) *** and ** indicate that the variable coefficients have passed significance tests at the 1% and 5% levels, respectively.

.

4.5.1. Narrowing the Sample Scope

As shown in column (1), we narrowed the sample scope by excluding abnormal or potentially biased years (such as 2006 and 2009) and re-ran the regression analysis. The results show that the coefficient of the level of intelligent equipment upgrades (LIR) is −0.227, which remains significantly negative (p < 0.01), and the coefficient of LIR² is 0.153, which is significantly positive (p < 0.01). These findings confirm that the original model’s conclusion of a U-shaped relationship between LIR and the carbon emissions efficiency (CEEE) of equipment-manufacturing enterprises is robust.

4.5.2. Replacing the Independent Variable

As shown in column (2), we used the density of robot installations as a substitute variable for the regression analysis. The results show that the coefficient of LIR is −0.584 (p < 0.01) and that the coefficient of LIR² is 0.313 (p < 0.01), both of which are significant. This indicates that regardless of the independent variable used, the impact of intelligent equipment upgrades on carbon emissions efficiency remains robust, further confirming the existence of the U-shaped relationship.

4.5.3. Instrumental Variable Method

As shown in column (3), to address potential endogeneity issues, the instrumental variable method was employed. Specifically, the density of industrial robots in the U.S. equipment-manufacturing industry was used as an instrumental variable. The regression results show that the coefficient of LIR is −0.231 (p < 0.01) and the coefficient of LIR² is 0.131 (p < 0.01), consistent with the original model. Furthermore, the F-statistic of the instrumental variable is significantly higher than the commonly used threshold of 10, indicating that the selected instrumental variable is valid.

In conclusion, the results from various robustness tests suggest that, whether they are tested by narrowing the sample scope, replacing the independent variable, or using the instrumental variable method, the findings on the impact of intelligent equipment upgrades on the carbon emissions efficiency of equipment-manufacturing enterprises remain robust. This further enhances the reliability and generalizability of the conclusions drawn in this study.

5. Conclusions and Implications

5.1. Conclusions

Through empirical analysis, this study explored the mechanism by which the level of intelligent equipment upgrades affects the carbon emissions efficiency (CEEE) of equipment-manufacturing enterprises. The results confirm the U-shaped impact of intelligent equipment upgrades on carbon emissions efficiency, with the cost-saving effect playing a partial mediating role in this process. Furthermore, the study reveals a significant moderating effect of energy intensity on the relationship between intelligent upgrades and carbon emissions efficiency. It should be noted that this study used data from A-share equipment-manufacturing listed companies from 2006 to 2019. At present, China’s manufacturing industry is still far from highly intelligent, and there are still a large number of manufacturing enterprises in the process of transformation. The 14th Five Year Plan proposes the goal of “basically achieving new industrialization by 2035”, includes plans to preliminarily complete the informatization and digitization of industry by 2035, and describes the expectation of long-term development of intelligent manufacturing in the future. Therefore, the existing research model can still be applied for the foreseeable future, for the next decade and beyond.

5.1.1. The U-Shaped Impact of Intelligent Upgrades on Carbon Emissions Efficiency

The empirical analysis results indicate there is a significant, U-shaped relationship between the level of intelligent equipment upgrades (LIR) and carbon emissions efficiency. According to the regression analysis data, in the early stages of intelligent upgrades, the coefficient of LIR is −0.870, which is significant at the 1% level. This suggests that in the initial phase, intelligent upgrades have a negative impact on carbon emissions efficiency, primarily due to the substantial capital required for purchasing equipment, integrating technology, and employee training, leading to a short-term increase in operational costs and a decline in carbon emissions efficiency.

However, as intelligent upgrades progress, the coefficient of the squared term of LIR is 0.873, which is also significant at the 1% level, further confirming the existence of the U-shaped relationship. The results show that once intelligent upgrades reach a certain threshold, carbon emissions efficiency begins to recover. This inflection point occurs when the LIR value reaches approximately 0.498, at which point the carbon emissions efficiency improves significantly, indicating that the long-term positive effects of intelligent upgrades on carbon emissions efficiency gradually emerge. The initial negative effects of intelligent upgrades are effectively reversed once firms overcome the initial high costs and challenges of technology integration, resulting in a notable improvement in carbon emissions efficiency.

5.1.2. The Mediating Effect of Cost-Saving

The study also reveals that the cost-saving effect plays a significant mediating role between intelligent equipment upgrades and carbon emissions efficiency. The empirical results show that the negative impact of intelligent equipment upgrades (LIR) on the cost-saving effect (CER) is significant (coefficient = −0.090), indicating that in the early stages, due to high expenditures on equipment purchases and technology integration, operational costs increase, suppressing carbon emissions efficiency in the short term. However, as the level of intelligent upgrades increases, the positive impact of the squared term of LIR on the cost-saving effect becomes significant (coefficient = 0.004). This indicates that in the mid-to-late stages, as equipment utilization improves and technology matures, intelligent upgrades gradually reduce operational costs, leading to a positive effect on carbon emissions efficiency.

Even after controlling for the cost-saving effect, the negative impact of LIR on carbon emissions efficiency remains significant (coefficient = −0.136) and the positive effect of the squared term of LIR persists (coefficient = 0.078). This further confirms the critical role of the cost-saving effect as a mediating mechanism between intelligent equipment upgrades and carbon emissions efficiency: in the early stages, the cost-saving effect negatively affects carbon emissions efficiency, but as intelligent upgrades progress, the cost-saving effect turns positive, driving improvements in carbon emissions efficiency.

5.1.3. The Moderating Role of Energy Intensity

Energy intensity (EI), as a moderating variable, significantly influences the relationship between intelligent equipment upgrades (LIR) and carbon emissions efficiency (CEEE). The empirical data show that the interaction term between LIR and energy intensity has a significant negative moderating effect on carbon emissions efficiency (coefficient = −9.462, p < 0.01), while the interaction term between LIR² and energy intensity exhibits a positive moderating effect (coefficient = 4.937, p < 0.01). This indicates that in high-energy-intensity firms, intelligent upgrades can improve carbon emissions efficiency earlier and more significantly. In particular, for companies with higher energy intensity, the marginal benefits of intelligent upgrades are more pronounced, as they optimize energy use, reduce energy waste, and significantly enhance carbon emissions efficiency.

Conversely, in firms with lower energy intensity, the positive effects of intelligent upgrades on carbon emissions efficiency are slower, with the inflection point appearing later. This suggests that firms with high energy intensity have more potential for improvement in carbon emissions efficiency, while the impact of intelligent upgrades in low-energy-intensity firms is relatively smaller.

5.2. Recommendations

5.2.1. Increase Financial Support and Incentives for Intelligent Upgrades

Establish Special Funds: The government can create dedicated funds to support equipment-manufacturing enterprises in purchasing smart technologies and equipment related to the circular economy. These funds can be provided through low-interest loans, direct subsidies, or tax breaks, reducing the financial burden on firms during the initial stages of intelligent upgrades, particularly in equipment procurement and technology integration. By encouraging increased investment in intelligent upgrades, especially in waste recycling, resource regeneration, and energy-saving areas, the overall transformation and upgrading of the equipment-manufacturing industry can be effectively promoted, supporting the development of the circular economy.

Tax Incentives: To further encourage the integration of circular economy principles in intelligent upgrades, the government can offer tax incentives for relevant R&D expenditures, equipment investments, and technology imports. For instance, the costs of acquiring energy-saving equipment, resource-recycling technologies, and related smart equipment could be tax-deductible or benefit from accelerated depreciation policies. Additionally, firms that improve resource efficiency, reduce carbon emissions, and achieve green production through intelligent upgrades could be further incentivized through reductions in environmental taxes or emission taxes, motivating more enterprises to embrace intelligent upgrades and the circular economy.

Public Technology Service Platform: The government can establish a public technology service platform that consolidates technology, training, consulting services, and equipment-sharing resources related to intelligent upgrades and the circular economy. This platform can provide equal access to technical resources for small and medium-sized enterprises, helping them master core technologies in efficient resource utilization, waste management, and resource regeneration. By facilitating industry-wide technical exchanges and training, firms can simultaneously improve the efficiency of their intelligent upgrades and promote sustainable resource recycling, reducing the environmental impact of production.

5.2.2. Optimize Policies for Intelligent Upgrades and Cost Management

Guide Enterprises in Formulating Strategies for Intelligent Upgrades: The government should guide enterprises in developing scientific roadmaps for intelligent upgrades, ensuring that these upgrades align with circular economy principles to achieve efficient resource utilization and minimal waste. Firms must not only focus on the initial high-cost investment but also implement effective cost-control measures to sustain resource efficiency and intelligent upgrades. For instance, the government can encourage enterprises to use investment-benefit-evaluation mechanisms to assess the cost–benefit ratios of resource reuse and energy-saving equipment, ensuring that firms balance intelligent upgrades with the development of the circular economy at every stage.

Promote Best Practices: The government can promote successful case studies of the integration of intelligent upgrades with the circular economy within the industry, showcasing how firms have optimized resource utilization and reduced waste emissions during the upgrade process, thereby improving cost management and production efficiency. Through widespread publicity and knowledge sharing, the government can encourage more enterprises to adopt these successful practices, accelerating the intelligent transformation of the entire industry and advancing the circular economy.

Enhance Internal Training for Enterprises: Intelligent upgrades not only involve technological and equipment updates but also require employees to have relevant operational skills and knowledge of the circular economy. Therefore, the government can provide special training funds to help enterprises enhance their employees’ capabilities in resource management, operation of intelligent equipment, and circular economy-related processes. This will assist enterprises in reducing resource waste, improving production efficiency, and advancing the deep integration of intelligent upgrades with the circular economy, ultimately maximizing production cost savings.

5.2.3. Strengthen Energy-Intensity Management and Promote Green Manufacturing

Implement Energy Quota Systems and Incentive Policies: For high-energy-consuming enterprises, the government should implement energy quota systems, setting annual energy-consumption limits and encouraging firms to reduce energy usage through technological upgrades and intelligent management. Additionally, the government can create special awards and subsidies to incentivize enterprises that achieve significant results in energy conservation, emission reduction, and resource recycling. This combination of rewards and penalties can increase firms’ awareness of energy conservation and help them transition to green manufacturing while gradually integrating into the circular economy model.

Promote the Application of Green Energy: The government should actively promote the widespread adoption of clean and renewable energy, offering financial subsidies or preferential loans to support enterprises in installing solar panels, wind power, and other equipment, reducing their reliance on traditional fossil fuels. In the process, the government can also encourage enterprises to implement intelligent energy-management systems, which enable real-time monitoring and optimization of energy usage, improving energy efficiency and reducing waste and thus supporting intelligent upgrades under the circular economy.

Promote Energy Optimization Technologies: Intelligent energy-management technologies, which utilize big data and artificial intelligence to achieve real-time monitoring and optimization of energy consumption, should be encouraged by the government. Enterprises can use these technologies to optimize energy efficiency and minimize energy waste in production. Additionally, the government can offer research and development subsidies to firms developing intelligent energy-management technologies, encouraging them to create customized solutions for various production environments, further achieving efficient energy utilization and recycling.

5.2.4. Establish Long-Term Mechanisms to Ensure Policy Continuity and Effectiveness

Link Intelligent Upgrades with Green Development: The government should combine intelligent upgrades with goals related to the circular economy and green development, creating a medium-to-long-term policy framework that ensures enterprises consider low-carbon economics and sustainable resource use while undergoing intelligent upgrades. By establishing long-term incentive mechanisms, the government can guide firms to integrate intelligent upgrades with resource reuse and carbon emissions reduction, ensuring that production efficiency improvements are accompanied by resource recycling and low-carbon development.

Enhance Cross-Departmental Coordination: The development of intelligent upgrades and the circular economy requires coordinated efforts across multiple government departments, particularly in the areas of industry, energy, and environmental protection. The government can establish a cross-departmental coordination committee to ensure consistency in resource utilization, intelligent management, and carbon emission goals, and to unify related policies. Departments should communicate regularly to avoid policy conflicts and ensure that intelligent upgrades positively impact efficiency in carbon emissions and the development of a circular economy.

Regularly Evaluate and Adjust Policies: The government should establish a regular policy-evaluation mechanism to assess the effectiveness of policies related to intelligent upgrades and the circular economy, addressing new issues and challenges that arise during their integration. By gathering industry feedback and conducting data analysis, the government can dynamically adjust policies to ensure they are more precise, allowing various industries and firms to effectively integrate intelligent upgrades with the circular economy based on their specific circumstances, maximizing policy effectiveness.

5.3. Limitations

Although this study provides valuable insights into the relationship between intelligent upgrades in equipment manufacturing and carbon emissions efficiency, it has several limitations. First, due to constraints in data availability on corporate carbon emissions and energy consumption, some estimations were necessary, potentially affecting the accuracy of the findings. Additionally, the study primarily examines short- and medium-term impacts without fully exploring long-term effects, particularly those arising from ongoing technological advancements and evolving market conditions. External factors such as government policies, fluctuations in market demand, and international economic influences were also not fully considered, despite their potential to significantly affect the outcomes of intelligent upgrades.

Future research should address these limitations by improving data-collection methods and using more precise analytical tools to enhance the reliability of conclusions. It would be valuable to investigate the long-term effects of intelligent upgrades, especially how they impact carbon emissions efficiency over time. Further, incorporating external factors such as policy changes, market dynamics, and the global economic environment could provide a more comprehensive understanding of the mechanisms driving intelligent upgrades. Exploring emerging intelligent technologies aligned with sustainability goals could also offer new directions for improving carbon efficiency and supporting the green transition of enterprises, providing actionable insights for policymakers and industry leaders.