Assessing the Impact of Industrial Robot Application on Urban Electricity Consumption in China

Abstract

The widespread application of industrial robots in China has significantly enhanced productivity, yet their impact on the energy system remains underexplored. This study empirically examines the impact of industrial robot application (IRA) on electricity intensity using panel data from 281 Chinese cities spanning 2006 to 2019 and a two-way fixed effects model. The results demonstrate that for every one-unit increase in IRA’s penetration rate, total electricity consumption and industrial electricity consumption decrease by 0.01 and 0.032 units, respectively. The effect operates through several mechanisms, including technological innovation, industrial agglomeration, and structural optimization. Despite these overall positive trends, the influence of IRA on electricity consumption exhibits notable regional heterogeneity. Furthermore, the study uncovers evidence of spatial spillover effects, indicating that the electricity-saving benefits of IRA extend beyond their immediate regions to neighboring cities. This phenomenon also contributes to narrowing the inter-city electricity gap, fostering a convergence in electricity consumption patterns among cities. These findings underscore the potential of industrial robots as a viable policy tool for advancing energy conservation and emission reduction goals.

1. Introduction

Economic development has placed a huge strain on global resources and environment. All countries are seeking to achieve a balance between economic development and environmental sustainability [1,2]. In particular, rapid population growth and urbanization have driven a significant increase in energy demand. Energy consumption and its structure are related to various pollutants and greenhouse gas emissions, so improving energy efficiency has become the key to achieving sustainable development. Electricity is an important form of energy utilization, and adequate power supply is crucial for economic and social development [3,4]. However, inefficient power use and extensive power structure can also harm environmental and resource welfare. Currently, climate warming and extreme weather have further stimulated electricity demand, which will drive overall energy consumption [5]. The International Energy Agency has stated that global electricity consumption in 2024 will be 27.8 trillion kilowatt-hours (kWh), representing a year-on-year increase of 4.3%, which is twice the average annual growth rate over the past decade. Efficient electricity utilization has become a crucial global concern.

Although electricity consumption in developed countries has been steadily decreasing as economic growth has slowed, the rapid increase in consumption in emerging economies is sufficient to reverse this trend. Developing countries are currently in an important stage of accelerated urbanization, and rising energy demand has led to a sharp increase in electricity consumption. As the world’s largest energy consumer and developing country, China is tasked with achieving multiple goals: maintaining rapid economic growth and reducing carbon emissions. However, China’s power structure, which is dominated by coal-fired power, is unlikely to change in the short term. In recent years, although the application of renewable energy has become increasingly widespread, coal and other fossil fuels still underpin regional economic growth and industrial production [6]. Meanwhile, China must enhance power utilization efficiency to meet the requirements of sustainable development. Therefore, scientifically identifying the influencing factors of electricity consumption holds practical value.

With the acceleration of technological revolution, digitization and intelligence have been integrated into industrial production. Among them, robots, as a key outcome of technological development, are closely related to digitization and intelligence. Many countries have made forward-looking arrangements for industrial robots as a strategic industry [7]. Industrial robots have been widely applied in industrial production and enterprise innovation, becoming an important tool for enhancing production efficiency [8,9]. The Chinese government has long recognized the significant role of IRA in industrial development and has introduced some policies to support the large-scale application of industrial robots. China has become the world’s largest producer and consumer of industrial robots, and it has tremendous potential in IRA. In 2024, the industrial robot output in China reached 556,400 units. China has built an advanced intelligent manufacturing system that relies on industrial robots, enhancing the competitiveness and productivity of the manufacturing industry.

To provide intelligent support for efficient electricity utilization and offer policy incentives for energy conservation and emission reduction, this research focuses on IRA. Firstly, IRA’s energy consumption has increased significantly in recent years, and the resulting electricity demand needs to be thoroughly examined. The global installation of industrial robots is expected to reach 542,000 units in 2024, with China accounting for 54% (295,000 units). As shown in Figure 1, China‘s IRA has been on an upward trend, especially since 2010. The large stock and growth of industrial robots may increase the burden on the power system. However, the role of IRA in shaping the power trajectory has not been adequately studied. Secondly, the previous literature may overlook potential endogeneity, heterogeneity, and spatial effects when assessing the influence of IRA. Our study comprehensively examines the impact of IRA on electricity consumption by integrating the instrumental variable method and spatial econometric estimation, providing more reliable empirical evidence. Thirdly, intelligence is regarded as a key factor in regional innovation [10,11,12], but the impact of digitalization and intelligence on energy utilization remains uncertain. Our research provides necessary theoretical and empirical evidence from the perspective of IRA to expand the understanding of related studies.

This paper aims to bridge these research gaps. Our research focuses on linking IRA with electricity consumption, expanding the research boundaries of IRA’s impact. We combine China’s electricity data and industrial robot data to calculate the penetration rate of IRA. Then, we examine the impact of IRA on electricity consumption from two dimensions: total electricity intensity and industrial electricity intensity. Based on this, this paper delves deeper into the influencing mechanisms and heterogeneity within it. Meanwhile, incorporating spatial effects into the research framework is highly necessary, as it helps to recognize the role of IRA in regional collaborative energy conservation.

The contribution of this study is reflected in three aspects. First, by examining the relationship between IRA and electricity consumption, we expand the research domain on intelligence and green development. Unlike existing studies that mainly focus on the economic effects of IRA, our research focuses on IRA’s power costs and emphasizes its electricity-saving effect. This topic can provide important empirical evidence for a deeper understanding of IRA and its impacts. Second, this study constructs the theoretical framework for the impact of IRA on electricity consumption and provides empirical evidence based on urban panel data. Our findings not only reveal the significant role of IRA in efficient power utilization from both theoretical and empirical perspectives, but also identify its spatial spillover effects. These findings highlight the need for regional collaboration to promote intelligence and energy conservation. Third, this study can provide insights for IRA and green production in China and other developing countries. Many emerging economies are actively developing industrial robots and need to better balance energy conservation and emission reduction. China is a major producer of industrial robots and a major energy consumer worldwide, so using China as an example case can provide abundant information. Our theoretical and empirical experience can guide the formulation of intelligent and green policies, thereby reducing the resource and environmental costs associated with intelligence or digitalization.

2. Literature Review and Theoretical Analysis

2.1. Literature Review

The literature reviews closely related to this study mainly cover three aspects: (1) the determinants of electricity consumption; (2) the relationship between intelligence and energy conservation; and (3) the impact assessment of industrial robots.

Efficient utilization of electricity has always been a focus of academic circles. Numerous studies have evaluated the determinants of electricity consumption. Among them, macroeconomic growth is a key factor influencing the trajectory of electricity changes, as economic growth drives electricity consumption [13,14]. Due to differences in economic level and industrial foundation, regional disparities in electricity consumption may persist [15]. Meanwhile, electricity exhibits the typical characteristics of a public good, so the government and market play significant roles in electricity consumption. On the one hand, government actions can regulate electricity consumption through administrative means. For instance, using electricity prices to alter the enthusiasm of different entities for electricity consumption is a common practice in countries around the world [16,17,18]. There are also studies indicating that government environmental regulations may not be conducive to improving electricity utilization efficiency, as they can interfere with enterprises’ energy efficiency and energy utilization patterns [19]. On the other hand, the market mechanism can effectively allocate power resources and balance power supply and demand [20]. Especially in emerging economies, their marketization degree is low, so market-oriented reforms can often effectively improve electricity utilization efficiency [21].

In recent years, with the advancement of digital technologies, many studies have begun to focus on the compatibility of intelligent development and green low-carbon transformation. Intelligence can promptly analyze emission data during the production process, strengthen monitoring of resources and energy consumption, and improve resource utilization efficiency [22]. For instance, using intelligent technology can improve green innovation and production structure to achieve decarbonization [23]. However, the impact of intelligent development on environmental sustainability cannot be ignored [24]. As intelligent development is highly dependent on digital infrastructures such as data centers and computing power, these digital devices generate a large amount of electricity demand, resulting in increased carbon emissions [25,26]. More importantly, the significant role of artificial intelligence in addressing climate change is multifaceted. Artificial intelligence technologies can better manage and predict changes in energy and emissions during production processes, and the government’s artificial intelligence policies can also drive regional low-carbon transformation [27]. These studies emphasize the necessity of artificial intelligence in achieving carbon reduction.

It is worth noting that industrial robots are a concentrated manifestation of artificial intelligence. In practical applications, industrial robots are mainly used in enterprise production and innovation, so many studies have focused on the economic effects of IRA. Among them, an increasing number of studies indicate that IRA can replace human labor in enterprise production processes, thereby reducing employment [7]. The opposing view is that IRA increases productivity and thereby leads to an increase in employment [28,29]. These studies emphasize that IRA is an important factor influencing the employment trajectory. IRA has been proven to be a crucial engine for enhancing productivity [30,31]. Li et al. (2024) [32] found that IRA increased total factor productivity in China’s manufacturing industry by 10%. Similar conclusions have been reached in other countries, indicating that IRA has a significant positive impact on productivity [28]. Additionally, there is some attention paid to the role of IRA in environmental sustainability [33,34]. For instance, Li et al. (2022) [35] used cross-border panel data to find that IRA reduced carbon emission intensity, and this effect is more pronounced in developed countries. Although these studies acknowledge the relationship between IRA and carbon emission reduction, our research focuses on the overlooked aspect of electricity consumption.

Although the existing literature has already addressed the impact of IRA, there is still room for breakthroughs. Firstly, most studies focus on carbon emissions, neglecting the role of electricity intensity. Few studies discuss the impact of IRA on electricity intensity within a unified framework. In fact, the power industry is a major source of greenhouse gas emissions [36], and ignoring IRA’s electricity consumption may limit the understanding of IRA. Secondly, previous studies did not explore the spatial correlation between IRA and electricity consumption, nor did they consider the spatial spillover effect of IRA on electricity intensity. This research gap hinders the implementation of policies that utilize IRA to achieve regional coordinated energy conservation. Meanwhile, electricity consumption exhibits significant spatial dependence within a geographical area. Ignoring this spatial effect may lead to incorrect judgments regarding IRA’s energy consumption. Furthermore, the heterogeneity impact of IRA has received extensive attention. However, whether this effect will promote the convergence of regional power disparities remains to be investigated.

Given the existing research gaps, this study presents new theoretical and empirical findings from the following perspectives. (1) Explaining the impact of IRA on electricity intensity and constructing a unified theoretical framework, deeply analyzing the relationship between intelligence and energy consumption, and providing a theoretical reference for policymakers to achieve a win–win situation of intelligence and green development. (2) Our research combines panel econometric testing and spatial econometric estimation to conduct an empirical study on the impact of IRA on electricity intensity and its spatial spillover effects. Furthermore, it aims to reveal the underlying mechanisms to enhance the understanding of IRA’s effect on electricity consumption. (3) This paper incorporates the inter-city electricity gap into the empirical framework and examines the impact of IRA on the inter-city electricity gap, thereby addressing the issue of whether IRA can achieve regional power convergence. This work verifies the positive contribution of IRA to reducing the regional electricity gap and supplements the existing literature on intelligence and energy consumption.

2.2. Theoretical Analysis and Research Hypotheses

The research framework of this study is shown in Figure 2.

2.2.1. Direct Effect Analysis

Industrial robots are hailed as “the most precious gem on the crown of the manufacturing industry”. China is a major manufacturing country with strong demand for IRA. Especially with the rise in labor costs and technological advances, it is an inevitable trend to use industrial robots to replace heavy and repetitive tasks. From practical experience, industrial robots have been widely applied in China’s manufacturing sectors, such as home appliances and chemicals, becoming a driving force for optimizing the industrial structure.

From the direct impact of IRA on electricity consumption, IRA may reduce electricity consumption by adopting clean production methods and green technologies. Compared to human labor, robots can precisely control each stage of production process, minimizing the corresponding electricity consumption and avoiding unnecessary losses caused by manual operations. With advances in technology, the green and low-carbon attributes of IRA have become increasingly prominent [37]. The capacity expansion brought about by IRA may lead to greater electricity demand, but the corresponding increase in productivity may significantly enhance the efficiency of electricity use. For instance, the factory of Xiaomi Automobile has over 700 industrial robots that use photovoltaic power to enhance the efficiency of power supply and utilization. At the same time, IRA led to the establishment of numerous dark-light factories, significantly reducing the electricity demand. Additionally, the collaborative operation of IRA has optimized the production process and generated significant economic benefits. In fact, although industrial robots are not energy-efficient devices, the resulting increase in productivity and economic growth can reduce the electricity intensity. The Chinese government has introduced a series of standards and regulations to promote new-type industrialization, emphasizing green, low-carbon production in manufacturing [38], especially by promoting the coordinated development of digitalization and greening. Figure 3 shows the initial fit of the correlation between IRA and electricity intensity, indicating a negative correlation between IRA and electricity intensity. Therefore, this paper presents Hypothesis 1.

H1: 

IRA significantly reduces electricity consumption.

2.2.2. Mechanism Analysis

In terms of the influencing mechanisms, IRA may reduce electricity intensity through technological innovation, industrial agglomeration, and structural optimization.

Firstly, IRA can accelerate technological innovation and reduce electricity consumption. This can be analyzed from two aspects: innovation motivation and innovation capability. Intelligent production requires advanced technological capabilities as a guarantee, thereby creating a strong demand for innovation [39]. Industrial robots can accelerate the development of big data and AI and promote the construction of digital infrastructure. Meanwhile, reductions in production costs and improvements in economic benefits from industrial robots may encourage enterprises to expand their production scale. During this process, the technological innovation demands of enterprises will accelerate technological progress. Furthermore, applying industrial robots to the production process can improve resource allocation in innovation processes and enhance innovation efficiency [40]. Industrial robots can address the shortcomings of traditional labor in innovation activities, such as long working hours, high productivity, and the need for adaptability to harsh environments. More importantly, industrial robots are more likely to meet government-set environmental standards and avoid penalties for noncompliance [41]. Therefore, enterprises can utilize industrial robots to unlock their innovative potential and achieve efficient electricity utilization.

Secondly, IRA may promote industrial agglomeration to achieve energy conservation. Industrial agglomeration is often an important means of achieving green development. Many studies have confirmed the positive environmental externalities arising from industrial agglomeration [42,43]. The sharing of infrastructure, such as transportation, energy, and electricity, within a spatial scope helps to significantly enhance the efficiency of resource utilization. The robots themselves, as typical representatives of the intelligent industry and digital technology, can attract big data, artificial intelligence and manufacturing enterprises to a region. For instance, the “Robot Valley” in Shenzhen has produced a large number of intelligent robots, and has also attracted surrounding industries such as artificial intelligence, automobile manufacturing, and universities, eventually forming a highly competitive high-tech industry. These industries not only have high industrial added value and innovation capabilities, but also exhibit low energy consumption. It can be seen that IRA can lead to significant economies of scale, enhancing economic efficiency while improving electricity utilization efficiency.

Thirdly, IRA may achieve electricity conservation through industrial structure optimization. From an industrial perspective, industrial structure optimization eliminates outdated industries that consume substantial electricity [44]. IRA not only enhances the productivity and competitiveness of various industries but also drives the industrial structure to a higher level. Specifically, industrial intelligent production can optimize outdated production capacity and promote the transformation of traditional industries, for instance, by using industrial robots to manage energy consumption and monitor emissions in coal and power enterprises, enabling green and efficient production. Meanwhile, the various products enabled by intelligent manufacturing have further stimulated the development of emerging industries and modern service sectors, contributing to industrial structure optimization.

In sum, this study presents Hypothesis 2.

H2: 

IRA can achieve electricity conservation through technological innovation, industrial agglomeration, and structural optimization.

2.2.3. Spatial Effect Analysis

Some studies have found that the development of digitalization and intelligence is often not constrained by spatial distance [45], suggesting that spatial effects may influence the integration of digital technologies and industrial intelligence. IRA may have a demonstration effect, encouraging enterprises in nearby areas to introduce industrial robots to achieve intelligent production [46]. In the context of local economic competition, local governments are keen to introduce robots and related industries to boost the economy. This economic competition model may rapidly alter regional energy consumption and intensify the burden on the power system. In fact, industrial robots leverage various carriers, such as the technology market, population mobility, and logistics, to enable cross-regional movement, thereby promoting and penetrating green technologies. At the same time, the spatial clustering of the robot industry may also lead to significant technological spillovers, thereby optimizing the power utilization patterns of surrounding cities. Overall, the impact of IRA on electricity intensity of surrounding cities depends on the extent of the influence of these two effects. Thus, we propose the following hypothesis.

H3: 

IRA may have significant spatial effects, not only affecting local electricity consumption but also altering electricity consumption in surrounding cities.

3. Research Design and Data

3.1. Model

To examine the impact of IRA on electricity consumption, we estimate a panel fixed effects model following Zhang and Ye (2026) [47], as shown in Equation (1).

$$E{C}_{it}={\alpha }_0+{\alpha }_{1}IR{A}_{it}+\beta Contro{l}_{it}+{\omega }_i+{\rho }_t+{\epsilon }_{it}$$

(1)

where EC represents electricity consumption, IRA is the industrial robot, and Control includes a series of control variables. ${\omega }_i$ and ${\rho }_t$ represent the urban and year fixed effects, and ${\epsilon }_{it}$ denotes the random error term. i and t stand for cities and years.

3.2. Variables

Independent variable: Industrial robot application (IRA). Referring to previous studies [48,49], the industrial robots’ penetration rate is constructed using industrial robot data from the International Federation of Robotics (IFR). We calculate IRA using the Bartik instrumental variable method. The specific formula is as follows.

$$IR{A}_{it}={\displaystyle \sum _{h=1}^m{l}_{jh{t}_0}\times \frac{robo{t}_{ht}}{L_{h{t}_0}}}$$

(2)

j and h respectively represent cities and industries, t represents the year, and m represents the number of industries. $robo{t}_{ht}$ is the robot stock of industry h in year t. $l_{jh{t}_0}$ represents the proportion of employment number in industry h of city j in the base period, and $L_{h{t}_0}$ is the total employment number in industry h. We set the base year as 2006 because robot data became available from that year. Meanwhile, during the calculation process, the robot data from IFR is matched with industry employment data from China’s Urban Statistical Yearbook to obtain robot data for different industries.

Dependent variable: Electricity consumption (EC). According to Perillo et al. (2022) [50], it is measured as electricity consumption per unit of GDP, namely, electricity intensity (EI). Meanwhile, since IRA is closely related to industrial development, we also use industrial electricity intensity (IEI) as the dependent variable.

Control variables. Some control variables are included in the econometric model to enhance the estimation accuracy. The population factor and economic factor are respectively measured by the logarithm of population density (PD) and per capita GDP (GDP). Government intervention (GOV) is measured by the proportion of fiscal expenditure in GDP. Technological investment (TI) is measured as proportion of fiscal spending on technology relative to total fiscal expenditure. Trade openness (TRADE) is measured by the proportion of import and export trade to GDP. Financial development (FD) is characterized by the proportion of loan balance of financial institutions to GDP.

3.3. Data

The research sample consists of panel data from 281 prefecture-level cities and above (2006–2019). The data is sourced from the following channels: IFR, China Statistical Yearbook, and China Urban Statistical Yearbook. The descriptive statistics are presented in

Table 1. Descriptive statistics.

Variables	Definition	Obs	Mean	Std. Dev.	Min	Max
EI	Electricity intensity	3934	0.1376	0.1187	0.0136	2.0963
IEI	Industrial electricity intensity	3934	0.2083	0.2440	0.0017	4.9883
IRA	Industrial robot application	3934	0.1914	0.4988	0.0001	9.4832
PD	Population density	3934	470.7977	565.4310	5.0672	8564.8470
GDP	Per capita GDP	3934	42,014.27	30,305.76	2767	203,489
GOV	Government intervention	3934	0.1807	0.1006	0.0427	1.4852
TI	Technological investment	3934	0.0145	0.0150	0.0003	0.2068
TRADE	Trade openness	3934	0.1930	0.3450	0.00001	3.4989
FD	Financial development	3934	0.8852	0.5613	0.1122	9.6221

.

4. Results and Discussion

4.1. Baseline Results

Table 2. Benchmark estimate.

	EI (1)	IEI (2)	EI (3)	IEI (4)
IRA	−0.025 *** (−9.32)	−0.034 *** (−7.72)	−0.010 *** (−3.29)	−0.032 *** (−4.54)
Control	×	×	√	√
Year FE	×	×	√	√
City FE	×	×	√	√
Obs	3934	3934	3934	3934
R2	0.0109	0.0045	0.6501	0.5961

Note: This table presents the results of the panel fixed effects model. The dependent variables are electricity intensity (EI) and industrial electricity intensity (IEI), while the core independent variable is industrial robot application (IRA). Control encompasses a set of control variables, as shown in Section 3.2. The specifications in column (1) and column (2) do not include control variables and fixed effects, and the specifications in column (3) and column (4) include control variables and fixed effects. t-value in parentheses. *** p < 0.01.

provides the baseline results. Columns (1) and (2) do not include fixed effects and control variables, and columns (3) and (4) incorporate fixed effects and control variables. The coefficients for IRA are consistently significantly negative at the 1% level, indicating that IRA significantly reduces total and industrial electricity intensity. Specifically, IRA led to decreases of 0.01 and 0.032 units in EI and IEI, respectively, suggesting the potential of IRA to promote electricity conservation. This finding supports Qi et al. (2024) [51]. The benchmark results indicate that IRA can significantly improve the electricity utilization efficiency. Although IRA requires a large amount of electricity and energy support, it can enhance electricity utilization efficiency through big data, artificial intelligence, and production process optimization. Meanwhile, IRA increases productivity and output value, and reduces electricity intensity. Our findings indicate that IRA can alleviate the burden on the power system and enhance environmental sustainability.

4.2. Robustness Test and Endogeneity Discussion

First, replace the dependent variable. We use per capita electricity consumption and industrial electricity consumption as the dependent variables for the regression. Columns (1) and (2) of

Table 3. Robustness test.

	Replace the Dependent Variable		Adjust the Sample		Change the Sample Period
	EI (1)	IEI (2)	EI (3)	IEI (4)	EI (5)	IEI (6)
IRA	−0.040 * (−1.90)	−0.057 *** (−2.85)	−0.018 *** (−3.47)	−0.058 *** (−4.97)	−0.008 *** (−2.69)	−0.026 *** (−3.92)
Control	√	√	√	√	√	√
Year FE	√	√	√	√	√	√
City FE	√	√	√	√	√	√
Obs	3934	3934	3850	3850	3091	3091
R2	0.7007	0.7045	0.6382	0.5957	0.6651	0.6255

Note: This table presents the robustness check results of the panel fixed effects model. The dependent variables are electricity intensity (EI) and industrial electricity intensity (IEI), while the core independent variable is industrial robot application (IRA). Control encompasses a set of control variables, as shown in Section 3.2. The specifications include control variables and fixed effects. t-value in parentheses. *** p < 0.01, * p < 0.1.

show that the coefficients of IRA are significantly negative, indicating that IRA significantly reduces electricity consumption. This finding is consistent with the previous results.

Second, eliminate special cities. The sample cities include municipalities and ordinary cities. There are significant differences in administrative status and economic conditions between them. Therefore, we exclude four municipalities (Beijing, Tianjin, Shanghai, and Chongqing) and two megacities (Guangzhou and Shenzhen). The regression results are shown in columns (3) and (4) of Table 3. The findings indicate that IRA’s coefficients are significantly negative, suggesting that the estimated results did not change significantly after adjusting the sample.

Third, change the sample period. The 2008 financial crisis, as an exogenous event, might interfere with the empirical results. Therefore, we exclude the samples from 2008 to 2010, which are the most severely affected by the crisis. After re-estimating in columns (5) and (6) of Table 3, it is found that the coefficients of IRA remain consistent with the baseline regression.

Fourth, data truncation processing. To further eliminate the influence of outliers, we apply 1% and 99% truncation to all the variables (columns (1) and (2) of

Table 4. Other robustness tests and endogeneity discussion.

	Data Truncation Processing		Lag Effect Regression		Instrumental Variable Method
	EI (1)	EI (2)	IEI (3)	EI (4)	IEI (5)	IEI (6)
IRA	−0.012 *** (−3.73)	−0.047 *** (−6.40)	−0.012 *** (−3.20)	−0.040 *** (−4.45)	−0.013 *** (−3.56)	−0.036 *** (−4.43)
Control	√	√	√	√	√	√
Year FE	√	√	√	√	√	√
City FE	√	√	√	√	√	√
Kleibergen–Paap rk Wald F statistic					448.185	448.185
Kleibergen–Paap rk LM statistic					55.786 ***	55.786 ***
Obs	3934	3934	3653	3934	3934	3934
R2	0.7142	0.6793	0.6557	0.6017	0.6767	0.6269

Note: This table presents the results of other robustness tests and endogeneity discussion. Columns (1) and (2) perform data truncation processing, columns (3) and (4) conduct regression using the one-period-lagged IRA (L1.IRA), and columns (5) and (6) employ instrumental variable estimation. The dependent variables are electricity intensity (EI) and industrial electricity intensity (IEI), while the core independent variable is industrial robot application (IRA). Control encompasses a set of control variables, as shown in Section 3.2. The specifications include control variables and fixed effects. t-value in parentheses. *** p < 0.01.

). The result supports the robustness of the main conclusion.

Fifth, consider the lag effect. Due to the endogeneity caused by reverse causality, the estimated results may be biased. To alleviate the potential endogeneity, we use the lagged value of IRA for estimation, as shown in columns (3) and (4) of Table 4. The coefficients for one-period-lagged IRA are significantly negative, meaning that the electricity-saving effect of IRA is persistent, thereby verifying the robustness of the results.

Sixth, the instrumental variable method. The endogeneity caused by omitted variables and measurement errors in data needs to be effectively addressed in empirical tests. This study uses the instrumental variable method to control for potential endogeneity as much as possible. Following He et al. (2024) [52], robot data from the United States in the same period is used to construct the robot penetration rate as an instrumental variable, as shown in columns (5)–(6). The reason is that the United States’ IRA holds a leading position in the global robot production network; it has a demonstration effect on China’s industrial robot development, and its changing trend is similar to that of China’s IRA. Thus, the two countries have a close relationship in adopting industrial robots. However, the robots in the United States generally only affect the economic development of their own country and do not significantly affect China’s electricity consumption.

$$usaROBO{T}_{it}={\displaystyle \sum _{h=1}^m{l}_{jh{t}_0}\times \frac{usarobo{t}_{ht}}{usa{L}_{h{t}_0}}}$$

(3)

where $usarobo{t}_{ht}$ represents the robot stock of industry h in year t in the United States, and $usa{L}_{h{t}_0}$ represents the employment number of industry h in the base period in the United States.

The results reported by the instrumental variable method are shown in columns (5) and (6) of Table 4. The coefficients for IRA remain significantly negative, indicating that the estimation result is robust after addressing the endogeneity issue.

4.3. Heterogeneity Analysis

Urban heterogeneity may affect the motivation and ability to adopt industrial robots. We further consider the heterogeneity of regression results. By constructing dummy variables, we examine the heterogeneous impact of IRA on electricity intensity.

First, geographical location heterogeneity. Define EAST = 1 if the city belongs to the eastern provinces, and 0 otherwise. Then, construct the interaction term between IRA and EAST.

Table 5. Heterogeneity analysis: Geographical location and economic level.

	EI (1)	IEI (2)	EI (3)	IEI (4)
IRA	−0.032 *** (−5.48)	−0.078 *** (−5.84)	−0.010 *** (−6.23)	−0.202 *** (−5.86)
IRA × EAST	0.025 *** (4.80)	0.052 *** (4.37)
IRA × HG			0.088 *** (6.07)	0.166 *** (5.31)
Control	√	√	√	√
Year FE	√	√	√	√
City FE	√	√	√	√
Obs	3934	3934	3934	3934
R2	0.6510	0.5970	0.6524	0.5980

Note: This table presents the heterogeneity results of the panel fixed effects model. The dependent variables are electricity intensity (EI) and industrial electricity intensity (IEI), while the core independent variable is industrial robot application (IRA). EAST represents the dummy variable for eastern cities, and HG represents the dummy variable for cities with high economic development levels. Control encompasses a series of control variables, and the set of control variables is shown in Section 3.2. The specifications include control variables and fixed effects. t-value in parentheses. *** p < 0.01.

shows that IRA is significantly negative, but IRA × EAST is significantly positive, indicating that the inhibitory effect of IRA on electricity intensity is more pronounced in central and western cities. One possible explanation is that economic development in the central and western regions drives higher energy and electricity consumption, so IRA significantly enhances the electricity utilization efficiency. In recent years, these regions have accelerated the development of digitalization and intelligence, actively introduced industrial robots and related industries, and promoted the coordinated development of urban digitalization and greening.

Second, economic heterogeneity. Calculate the mean of per capita GDP (MEAN_GDP) for all cities within the sample period. Next, define HG = 1 if the per capita GDP of a city is higher than MEAN_GDP. Finally, we construct IRA × HG and incorporate it into the econometric model in

Table 6. Heterogeneity analysis: Resource endowment and administrative hierarchy.

	EI (1)	IEI (2)	EI (3)	IEI (4)
IRA	−0.011 *** (−3.75)	−0.035 *** (−4.94)	−0.021 *** (−2.94)	−0.060 *** (−3.71)
IRA × RC	−0.113 *** (−6.54)	−0.233 *** (−6.42)
IRA × AC			0.013 ** (2.22)	0.031 ** (2.38)
Control	√	√	√	√
Year FE	√	√	√	√
City FE	√	√	√	√
Obs	3934	3934	3934	3934
R2	0.6557	0.6018	0.6503	0.5964

Note: This table presents the heterogeneity results of the panel fixed effects model. The dependent variables are electricity intensity (EI) and industrial electricity intensity (IEI), while the core independent variable is industrial robot application (IRA). RC represents the dummy variable for resource-based cities, and AC represents the dummy variable for higher-administrative cities. Control encompasses a series of control variables, and the set of control variables is shown in Section 3.2. The specifications include control variables and fixed effects. t-value in parentheses. *** p < 0.01, ** p < 0.05.

. The results show that IRA and IRA × HG are significantly negative and positive, respectively, indicating that the electricity-saving effect of IRA is more pronounced in economically underdeveloped cities. The electricity intensity in economically underdeveloped cities is significantly higher. Adopting industrial robots for intelligent production can rapidly shift traditional electricity utilization patterns and help promote the green transformation of underdeveloped cities.

Third, resource endowment heterogeneity. Resource-based cities often face challenges in balancing economic growth and green transformation [9]. According to the “National Sustainable Development Plan for Resource-based Cities (2013—2020)”, we divide the entire sample into resource-based cities and non-resource-based cities. Define RC = 1 if the city is a resource-based city. We find that IRA × RC is significantly negative in Table 6, indicating that IRA can achieve electricity conservation for resource-based cities. Generally speaking, resource-based cities rely on resource extraction as their growth driver and are highly dependent on resource-based industries. These cities can optimize energy utilization by adopting industrial robots in the process of resource extraction and industrial development, thereby unlocking greater potential for electricity conservation.

Fourth, administrative hierarchy heterogeneity. The provincial capital cities, municipalities, and sub-provincial cities are classified as high-administrative cities, while other cities are classified as general-administrative cities. In addition, the dummy variable AC is constructed. If a city belongs to the higher-administrative cities, then AC = 1. The results show that IRA is significantly negative while IRA × AC is significantly positive. This indicates that the electricity-saving effect of IRA is weakened in high-administrative cities. This finding implies that IRA helps narrow the electricity consumption gap between cities of different administrative levels.

4.4. Mechanism Test

(1) Technological innovation mechanism. Most of the literature uses patents to measure regional technological innovation [53,54]. To comprehensively assess the scale and quality of technological innovation, we employ per capita patent applications (PCP) and per capita invention patent applications (PCIP) as indicators. Additionally, we further use per capita green patent applications (PCGP) and per capita green invention patent applications (PCGIP) as dependent variables to examine the influence of IRA on green technological innovation. The results are shown in

Table 7. Technological innovation channels.

	PCP (1)	PCIP (2)	PCGP (3)	PCGIP (4)
IRA	9.369 *** (11.45)	4.137 *** (13.45)	0.980 *** (14.34)	0.531 *** (14.27)
Control	√	√	√	√
Year FE	√	√	√	√
City FE	√	√	√	√
Obs	3934	3934	3934	3934
R2	0.8511	0.7622	0.8177	0.7871

Note: This table presents the results of the panel fixed effects model (testing the technological innovation channels). The dependent variables are per capita patent applications (PCP), per capita invention patent applications (PCIP), per capita green patents (PCGP), and per capita green invention patent applications (PCGIP), while the core independent variable is industrial robot application (IRA). Control encompasses a series of control variables, and the set of control variables is shown in Section 3.2. The specifications include control variables and fixed effects. t-value in parentheses. *** p < 0.01.

. Columns (1) and (2) indicate that IRA is significantly positive, suggesting that IRA can significantly enhance the quantity and quality of technological innovation. Meanwhile, the coefficients in columns (3) and (4) show that IRA is also significantly positive, indicating that IRA promotes green technological innovation. This result indicates that IRA has the potential to drive technological innovation and green innovation. The development of industrial intelligence is conducive to the emergence of more advanced and green technologies [55], promoting clean production and efficient utilization of resources, and providing the impetus for urban energy conservation.

(2) Agglomeration effect mechanism. Theoretical analysis indicates that industrial agglomeration may improve electricity utilization efficiency through positive environmental externalities. We use the domestic gross domestic product per unit area to represent industrial agglomeration [56]. Meanwhile, we also use the output density of different industries as the dependent variable to test the impact of IRA on industrial agglomeration across industries, as shown in

Table 8. Agglomeration effect channels.

	IAGG (1)	PIAGG (2)	SIAGG (3)	TIAGG (4)
IRA	0.295 *** (7.13)	−0.0004 ** (−2.03)	0.072 *** (4.80)	0.224 *** (7.95)
Control	√	√	√	√
Year FE	√	√	√	√
City FE	√	√	√	√
Obs	3934	3934	3934	3934
R2	0.9024	0.8702	0.9183	0.8845

Note: This table presents the results of the panel fixed effects model (testing the agglomeration effect channels). The dependent variables are industrial agglomeration (IAGG), primary industrial agglomeration (PIAGG), secondary industrial agglomeration (SIAGG), and tertiary industrial agglomeration effect (TIAGG), while the core independent variable is industrial robot application (IRA). Control encompasses a series of control variables, and the set of control variables is shown in Section 3.2. The specifications include control variables and fixed effects. t-value in parentheses. *** p < 0.01, ** p < 0.05.

. Column (1) shows that the coefficient for IRA is significantly positive, indicating that IRA is conducive to promoting industrial agglomeration. Furthermore, the coefficient of IRA on the primary industrial agglomeration (PIAGG) is significantly negative, whereas the coefficients for secondary industrial agglomeration (SIAGG) and tertiary industrial agglomeration (TIAGG) are significantly positive. This indicates that industrial robots can drive the agglomeration of manufacturing and service industries, which is similar to the findings of Lin and Xu (2024) [57]. IRA improves enterprise productivity and production factor allocation, thereby motivating the manufacturing and service industries to cluster spatially. This result indicates that IRA can facilitate the transformation of agriculture into manufacturing and services, providing the impetus for optimizing the economic structure.

(3) Structural optimization mechanism. Theoretical analysis suggests that the structural optimization effect may become an important influencing mechanism. In this section, we conduct tests on the industrial structure. Firstly, industrial structure optimization is represented by industrial structure rationalization (ISR) and industrial structure advancement (ISA). Among them, ISR is measured by the reciprocal of the Theil index [58]. ISA is represented by the product of the three industries’ labor productivity and the proportion of their added value [59], and labor productivity is averaged to eliminate the dimension.

Table 9. Structural optimization channels.

	ISR (1)	ISA (2)
IRA	−0.093 (−0.14)	0.097 *** (4.46)
Control	√	√
Year FE	√	√
City FE	√	√
Obs	3934	3934
R2	0.6131	0.7117

Note: This table presents the results of the panel fixed effects model (testing the structural optimization channels). The dependent variables are industrial structure rationalization (ISR) and industrial structure advancement (ISA), while the core independent variable is industrial robot application (IRA). Control encompasses a series of control variables, and the set of control variables is shown in Section 3.2. The specifications include control variables and fixed effects. t-value in parentheses. *** p < 0.01.

indicates the coefficient for IRA on ISR is not significant, but the coefficient for IRA on ISA is significantly positive. This indicates that IRA significantly improves industrial structure advancement but fails to promote industrial structure rationalization. One possible explanation is that IRA drives the development of high-end manufacturing and modern service industries, enabling the industrial structure to move towards a higher level. However, China’s industrial structure is still at the mid-to-low end of the industrial chain. The positive impact of IRA on many less developed and low-end industries is relatively weak.

4.5. Spatial Effect Test

Given IRA’s potential for significant spatial effects, we conduct a test using a spatial econometric model. Following Zhou and Lin (2025) [60], a Spatial Durbin Model is constructed, as shown in Equation (4).

$$E{C}_{it}=\alpha +\rho W_{ij}\cdot E{C}_{it}+{\beta }_{1}IF{A}_{it}+{\beta }_2{W}_{ij}\cdot IF{A}_{it}+\eta Contro{l}_{it}+\eta W_{ij}\cdot Contro{l}_{it}+{\omega }_i+{\rho }_t+{\epsilon }_{it}$$

(4)

where W represents the spatial weight, which is measured using a geographical distance matrix. Specifically, it is represented by the reciprocal of the square of the geographical distance calculated based on the city’s latitude and longitude [61]. Other variables are consistent with Equation (1). The results are presented in

Table 10. Spatial effect test.

	EI (1)	IEI (2)
IRA	−0.007 * (−1.91)	−0.023 *** (−3.01)
Control	√	√
W × IRA	−0.025 *** (−2.59)	−0.070 *** (−3.33)
W × Control	√	√
Direct effect	−0.007 * (−1.85)	−0.023 *** (−2.88)
Indirect effect	−0.026 *** (−2.75)	−0.070 *** (−3.47)
Total effect	−0.033 *** (−3.51)	−0.093 *** (−4.72)
Year FE	√	√
City FE	√	√
Obs	3934	3934
R2	0.0244	0.0139
Log-likelihood	5040.042	1930.990

Note: This table presents the results of the spatial econometric model and maximum likelihood estimation. W represents the geographical distance weight. The dependent variables are electricity intensity (EI) and industrial electricity intensity (IEI), while the core independent variable is industrial robot application (IRA). Control encompasses a series of control variables, and the set of control variables is shown in Section 3.2. The specifications include control variables and fixed effects. t-value in parentheses. *** p < 0.01, * p < 0.1.

.

Regardless of whether the dependent variable is EI or IEI, the coefficients for IRA and the direct effect are significantly negative, indicating that IRA significantly reduces the electricity intensity in this region. Meanwhile, the coefficients for W × IRA and indirect effect are significantly negative, indicating that IRA significantly reduces the surrounding cities’ electricity intensity. Additionally, the coefficients for total effect are significantly negative. This is consistent with previous research [46]. It can be seen that IRA has a positive demonstration effect on surrounding cities. Luo et al. (2026) [62] indicated that geographical proximity was conducive to the spatial expansion of IRA and actively contributed to carbon emission reduction in surrounding areas. On one hand, the advancement of industrial intelligence often leads to typical spatial clustering, such as artificial intelligence demonstration zones and robot industrial parks. This enables the electricity-saving effect of IRA to spread spatially. On the other hand, the development of regional industrial robots prompts surrounding cities to imitate and learn from it, thereby forming a positive interaction on a larger scale.

4.6. Impact of IRA on Electricity Gap

Electricity consumption accurately reflects economic development status. It is worth noting whether there is a digital divide or digital dividend in the impact of IRA on electricity consumption. According to Luo et al. (2025) [63], the variation coefficient of electricity intensity is used to construct the inter-city electricity consumption gap (EG) and industrial electricity consumption gap (IEG). The results are shown in

Table 11. Impact of IRA on the electricity gap.

	EG (1)	IEG (2)	EG (3)	IEG (4)
IRA	−0.058 *** (−6.34)	−0.097 *** (−7.10)	−0.060 *** (−5.88)	−0.110 *** (−7.26)
Control	×	×	√	√
Year FE	√	√	√	√
City FE	√	√	√	√
Obs	3934	3934	3934	3934
R2	0.0291	0.0608	0.0771	0.1251

Note: This table presents the results of the panel fixed effects model. The dependent variables are inter-city electricity consumption gap (EG) and industrial electricity consumption gap (IEG), while the core independent variable is industrial robot application (IRA). Control encompasses a series of control variables, and the set of control variables is shown in Section 3.2. The specifications in column (1) and column (2) do not include control variables, and the specifications in column (3) and column (4) include control variables. t-value in parentheses. *** p < 0.01.

.

Regardless of whether control variables and fixed effects are included, the coefficients for IRA are significantly negative, indicating that IRA significantly reduces the inter-city electricity consumption gap. Therefore, the impact of IRA has a clear digital dividend. Previous studies have confirmed the positive role of digital economy in narrowing the regional electricity gap [56]. Currently, industrial intelligence development in regions with high energy consumption can achieve energy-saving effects. By introducing advanced robots and production equipment, it can improve the enterprise productivity and the electricity utilization efficiency. This result implies that IRA enhances resource allocation among cities and improves factor flow and power convergence.

5. Conclusions and Policy Implications

The nexus between industrial intelligence and energy consumption is a critical consideration, bearing significant implications for the resource and environmental costs of technological advancement. This study utilizes urban panel data from 2006 to 2019 to empirically examine the impact of IRA on urban and industrial electricity intensity.

Our analysis concludes that IRA significantly reduces electricity intensity, a finding that remains robust across various model specifications. The electricity-saving effects are particularly pronounced in midwestern, economically underdeveloped, resource-based, and non-administrative-center cities. The mechanisms mediating this relationship include technological innovation, industrial agglomeration, and structural optimization. Crucially, the energy-saving benefits of IRA exhibit significant spatial spillover effects, reducing electricity intensity not only within the host city but also in neighboring regions. Furthermore, IRA contributes to the convergence of electricity consumption among cities, playing a demonstrable role in narrowing the inter-city electricity gap.

These findings offer some valuable insights. First, policymakers should champion the wider adoption of industrial robots to synergize intelligent development with green economic goals. As the deployment of industrial robots requires substantial capital investment, governments should leverage fiscal and financial instruments to provide necessary support. Concurrently, fostering deep integration between industry, academia, and research is essential to accelerate technological R&D and assist enterprises in phasing out obsolete production capacity. Second, targeted policies should address regional disparities. In less developed regions, promoting IRA offers a dual opportunity to narrow the economic gap with developed areas while advancing sustainable development. These regions should capitalize on the ongoing digital transformation to integrate robotics into green innovation and industrial upgrading. However, the potential for increased electricity demand from large-scale IRA deployment must be managed proactively. The focus should therefore be on the green low-carbon development of robotics itself. Third, it is vital to leverage the positive spillover effects of industrial intelligence. Local governments should consider establishing regional robotics industrial parks to cultivate spatial agglomeration and maximize the positive externalities of IRA. Furthermore, neighboring cities should enhance cooperation in IRA, big data, and artificial intelligence to facilitate the cross-regional application of new technologies for energy conservation.

This article also has some limitations. Firstly, our research is limited to Chinese cities. Future studies can be expanded to other countries and regions worldwide, providing intelligent support for power utilization from a broader perspective. In particular, it is possible to explore the differences in IRA’s impact between developed and developing countries. Secondly, the research subjects of this paper focus on industrial robots, but robots in other industries can also be further considered, for instance, agricultural robots and logistics robots. It is hoped that more data will become available to conduct empirical analysis of IRA and electricity usage. Thirdly, this study analyzes the impact of IRA on electricity consumption through literature and case studies. Further research can attempt to construct a mathematical model to simulate the effect of IRA.