ORIGINAL RESEARCH article

Front. Environ. Sci., 11 September 2026

Sec. Environmental Economics and Management

Volume 14 - 2026 | https://doi.org/10.3389/fenvs.2026.1931075

Can industrial internet drive “new quality productivity” without fueling pollution? A paradox

  • 1. Business School, East China University of Political Science and Law, Shanghai, China

  • 2. The School of Finance, Hunan University of Technology and Business, Changsha, China

Abstract

Introduction:

Whether digital technologies can deliver productivity leaps while safeguarding environmental bottom lines is a core issue shaping the path of high-quality development.

Methods:

Using the pilot demonstration projects of Industrial Internet Construction (IIC) as a quasi-natural experiment, this paper systematically investigates the dual effects of IIC on enterprises’ new quality productivity cultivation and pollution emissions via the multi-period difference-in-differences (DID) method.

Results:

The results show that the IIC significantly boosts enterprises’ green total factor productivity (GTFP), and this effect is primarily driven by improvements in green technical efficiency, whereas the contribution of green technological progress remains insignificant. Meanwhile, the policy drives a notable increase in the total volume of enterprise pollution emissions, evidencing an environmental backfire effect at the firm level, with the scale rebound effect as the core driver of rising total emissions. Heterogeneity analysis reveals that the policy effects are more pronounced in state-owned enterprises and manufacturing firms, but relatively muted in high-tech industries. Enterprises with weaker market power face stronger emission rebound pressure, and firms with weaker long-term orientation exhibit larger environmental costs. Further analysis indicates that IIC generally promotes green innovation among enterprises, yet the green innovation response is weaker among firms facing faster growth pressure.

Discussion:

This study provides micro-level empirical evidence for designing industrial policies that balance efficiency gains and emission regulation.

1 Introduction

As global climate governance advances and resource and environmental constraints tighten, China’s economy has entered a critical phase of high-quality development led by new quality productivity (Zhang et al., 2026). As the core carrier of new digital infrastructure, Industrial Internet Construction (IIC) is endowed with the dual mission of promoting industrial upgrading and green transformation at the policy level. On the one hand, data-driven production modes are widely acknowledged to optimize resource allocation and reduce energy consumption per unit of output, serving as a key pillar for fostering green growth momentum. On the other hand, the leap in production efficiency enabled by digital technology tends to incentivize enterprises to expand their output scale. The resulting scale rebound effect may offset or even outweigh the emission reduction gains per unit of output, ultimately driving up total pollution emissions (Han et al., 2023). This coexistence of efficiency improvement and total emission growth forms a potential paradox within the green effects of digital technologies.

Since 2017, the Ministry of Industry and Information Technology (MIIT) has selected and announced Industrial Internet pilot demonstration projects in multiple batches, creating an ideal quasi-natural experiment setting for identifying the actual effects of IIC (Ping et al., 2025). Existing studies mostly verify the green benefits of IIC from a single dimension, focusing on its role in boosting green production efficiency, while generally overlooking the dynamic changes in total pollution emissions. Evaluating the environmental effects of digital technologies solely through efficiency indicators can easily miss the hidden environmental costs driven by scale expansion, resulting in a biased judgment of policy performance. Meanwhile, regarding the internal driving channels of new quality productivity, most extant research remains at the level of verifying overall effects, and fails to effectively distinguish the core source of the enabling effect: whether it arises from the catch-up effect of improved management efficiency, or the breakthrough effect of advancing the technological frontier.

Building on the above background and existing research gaps, this study addresses three core research questions. First, whether IIC effectively fosters new quality productivity at the firm level, specifically, whether it significantly elevates enterprises’ green total factor productivity (GTFP). Second, alongside efficiency gains, what pattern of change is observed in the total volume of enterprise pollution emissions, and does the “green paradox” manifest at the firm level? Third, what constitutes the core driving channel for improvements in new quality productivity, and what is the internal mechanism underlying changes in total pollution emissions? Systematically answering these questions helps comprehensively clarify the dual effects of IIC, and provides micro-level empirical evidence for balancing efficiency improvement and total emission regulation in the process of digital infrastructure promotion.

Recent studies have further enriched the understanding of digital technologies’ environmental consequences. Chouksey et al. (2026) systematically review the climate and sustainability implications of artificial intelligence and articulate a sustainability paradox in which intelligent technologies deliver efficiency gains while simultaneously generating rising ecological costs through energy consumption, hardware supply chains and rebound effects. At the firm operational level, data-driven methods have also found direct applications in green manufacturing. Geda and Kwong (2018) develop a distributed lag model with Bayesian estimation to forecast used product returns for remanufacturing, demonstrating how data analytics can reduce inventory costs and support circular economy practices. Regarding industrial internet specifically, Xiao et al. (2026) provide quasi-experimental evidence that industrial internet platforms significantly enhance green total factor productivity in China’s renewable energy sector. These studies collectively confirm the dual nature of digital technologies, but most existing works examine either the efficiency dimension or the environmental cost dimension separately. Integrated firm-level evidence for targeted industrial digital policies remains scarce.

The marginal contributions of this study are mainly reflected in three aspects. First, few studies have placed new quality productivity cultivation and total emission cost assessment under the same micro analytical framework. Existing research on the digital rebound effect has widely discussed the environmental trade-offs of technological progress (Zhu and Lan, 2023; Liu et al., 2025), but most works focus on carbon intensity or aggregate regional evidence. Scant firm-level causal evidence is available for targeted industrial digital policies such as IIC. This paper simultaneously investigates the dual effects of IIC on GTFP and total pollution emissions, and systematically verifies the double-edged sword effect of digital infrastructure construction. This framework compensates for the one-sidedness of existing studies that evaluate digital environmental effects only from the efficiency dimension. Second, based on the index decomposition results of GTFP, this study accurately identifies the core enabling channel of new quality productivity improvement, and interprets the internal mechanism underlying changes in total pollution emissions from the perspective of the scale rebound effect. This analytical approach deepens the academic understanding of how digital technologies exert their environmental effects. Third, leveraging multi-batch IIC pilot programs to construct a quasi-natural experiment setting, this study achieves more rigorous causal identification between policy shocks and firm performance. The findings provide micro-level empirical evidence at the firm level for formulating digital infrastructure promotion policies that balance efficiency improvement and total emission regulation.

The remainder of this paper is organized as follows. Section 2 provides a literature review that systematically sorts out the relevant research strands on the green empowerment and environmental rebound effects of digital technologies. Section 3 presents the theoretical analysis and research hypotheses. Section 4 describes the research design, elaborating on model specification, variable selection and data sources. Section 5 analyzes the empirical results. Section 6 carries out extended discussions. Section 7 concludes the study and puts forward policy recommendations.

2 Literature review

2.1 Economic effects of IIC

Existing studies have confirmed the positive effects of IIC on corporate operating performance from multiple dimensions. In terms of efficiency, IIC significantly improves enterprises’ total factor productivity and resource allocation efficiency through equipment interconnection, data-driven decision-making and production process reengineering (Cao et al., 2025; Wang and Wu, 2025). In terms of innovation, relying on knowledge reuse mechanisms and collaborative R&D networks, IIC platforms effectively reduce innovation costs, thereby accelerating the process of enterprise product iteration and process innovation (Yi et al., 2024). In terms of the industrial chain, IIC breaks down information barriers between enterprises and promotes supply chain collaboration and the modernization of the industrial chain (Cheng, 2023). However, most of these studies are based on theoretical deduction or industry-level data, and causal identification evidence at the micro-firm level remains relatively scarce.

2.2 Research on the green empowerment effect of digital technologies

Existing studies generally agree that the deep integration of new-generation information technology and industrial systems is a key driver of enterprises’ green development. From the perspective of resource allocation, digital technologies can break down data barriers across all production links, reduce ineffective losses in the production process, and provide stable support for GTFP (Liu and Li, 2025; Zhang and Mao, 2026). From the perspective of technological innovation, digital platforms reduce the R&D and diffusion costs of green technologies, promote the penetration of cleaner production technologies along the industrial chain, and provide technical support for long-term green growth (Bibi et al., 2026; Xu et al., 2026).

Targeted studies on IIC also support the above positive conclusions. Equipment networking and digitalized production processes can significantly improve enterprises’ energy use efficiency and reduce energy consumption per unit of output (Chao et al., 2024). When pollutant emissions are incorporated into the efficiency accounting framework, the boosting effect of IIC on GTFP remains valid, with resource allocation optimization and technological innovation as the core functional paths (Jian et al., 2023; Zhang et al., 2024). Overall, existing studies widely acknowledge the green efficiency gains of IIC, but insufficiently decompose the internal composition of efficiency growth. The relative contributions of the two channels, namely, technical efficiency catch-up and technological frontier breakthrough, have not yet been clearly verified by micro-level evidence.

2.3 Research on the environmental rebound effect of digital technologies

Efficiency improvement does not necessarily lead to a decline in total emissions, and the scale rebound effect is a critical factor that cannot be ignored when evaluating the environmental effects of technologies. The theoretical logic of the rebound effect has been fully validated in academia: energy efficiency gains from technological progress reduce the marginal cost of production, incentivize firms to expand their output scale, and ultimately offset part or even all of the emission reduction outcomes (Zhu and Lan, 2023; Liu et al., 2025). In extreme cases, a “backfire effect” may occur, in which total emissions rise rather than fall (Du et al., 2025).

The rebound effect in digital contexts has also been supported by abundant empirical evidence. Digital transformation generally reduces the carbon emission intensity and pollution emission intensity of enterprises, but the growth of total emissions driven by output expansion often weakens the actual effect of intensity-based emission reduction (Lu and Wang, 2024). The magnitude of rebound varies across different types of digital technologies: intelligent manufacturing technologies targeting the production side usually have a more prominent rebound effect, while management-oriented digital systems exert a weaker stimulus on output scale, with a relatively smaller rebound magnitude (Chen and Li, 2024). Cross-industry comparisons in existing studies also show that the manufacturing sector has strong production scalability, and efficiency improvements are more readily translated into capacity growth, resulting in a generally higher rebound effect than the service sector (Font Vivanco et al., 2022). Existing research has confirmed the prevalence of the rebound effect in the digital field, but micro-level evidence for the IIC remains relatively scarce. In particular, there is a lack of systematic analysis that simultaneously examines efficiency improvement and total emission changes within a unified analytical framework.

3 Theoretical analysis and research hypotheses

3.1 Enabling effect of IIC on new quality productivity

New quality productivity takes scientific and technological innovation as its core driving force. It pursues higher quality of economic output alongside lower environmental costs. At the micro-firm level, production efficiency improvement represents the most direct manifestation of new quality productivity. Existing studies widely use green total factor productivity as the core proxy for firm-level new quality productivity. This indicator incorporates pollutant emissions as undesirable outputs into efficiency accounting. It captures both economic efficiency gains and environmental performance improvements. Its connotation fits well with the inherent requirements of high-quality development driven by technological progress (Gao and Li, 2025). This paper focuses on the production efficiency dimension of new quality productivity and uses GTFP as the core measurement indicator. The driving effect of IIC on enterprises’ green efficiency is mainly realized through step-by-step optimization on the production side. With full network connectivity of production equipment, enterprises can obtain real-time operation parameters, energy consumption data and material consumption progress of each process, thereby reducing ineffective losses in production. This refined management path has been validated by multiple studies on industrial digitalization (Lin and Teng, 2023; Guo and Yang, 2025).

At the energy management level, IIC can align production plans with energy supply and dynamically allocate resources including water, electricity and gas according to actual production demand. Such dynamic scheduling not only cuts idle energy consumption from equipment no-load operation, but also smooths energy load during peak production periods, lifting overall energy use efficiency. At the industrial chain collaboration level, IIC opens up information channels between upstream and downstream enterprises, reducing inventory backlog and turnover losses caused by supply-demand mismatch. Optimized resource allocation across the entire chain further amplifies green efficiency gains at the firm level (Azumah et al., 2026). Accordingly, this paper proposes Hypothesis 1:

Hypothesis 1The IIC significantly improves enterprises’ GTFP and exerts a positive driving effect on the cultivation of new quality productivity.

3.2 Scale rebound effect of IIC on pollution emissions

Improvement in green production efficiency is not directly equivalent to a decline in total pollution emissions. The rebound effect theory in energy economics has long established that efficiency gains from technological progress reduce the marginal cost of production, which in turn incentivizes firms to expand their production scale and ultimately partially or fully offsets the resource savings brought by efficiency improvement (Greening et al., 2000).

IIC exerts a remarkable enhancement effect on production efficiency. With efficiency gains, enterprises’ unit production cost decreases and capacity release accelerates significantly. Against the backdrop of unsaturated market demand, firms generally have incentives to expand production scale and capture larger market share. For manufacturing enterprises in particular, their production systems feature strong scalability, enabling efficiency dividends to be rapidly translated into actual output growth (Wen et al., 2022). At this stage, although the pollution emission intensity per unit of output declines, the rapid expansion of output scale drives a synchronous rise in total emissions. Once the emission increment from scale expansion exceeds the emission reduction from lower intensity, total enterprise pollution emissions will register a net increase. In energy economics, this scenario where efficiency gains lead to higher total emissions is known as the backfire effect. It represents the extreme form of the rebound effect. For consistency with existing digital environmental research, we describe this coexistence of green efficiency improvement and rising total emissions as a firm-level green paradox phenomenon. The scale rebound effect serves as its core mechanism. This pattern has been preliminarily verified in relevant studies (Wang, 2026). Accordingly, this paper proposes Hypothesis 2:

Hypothesis 2The IIC will trigger a significant scale rebound effect, which in turn drives up total enterprise pollution emissions. That is, the “green paradox” objectively exists at the firm level.

3.3 Mechanism analysis

3.3.1 Driving channels of new quality productivity: efficiency catch-up and technological breakthrough

According to the measurement logic of the SBM-ML index, GTFP growth can be decomposed into two core components, and this decomposition framework has been widely applied in the field of environmental efficiency assessment (Wang et al., 2026; Wu and Zhao, 2026). The first is green technical efficiency improvement, which captures the process whereby enterprises improve the utilization of existing technologies through management optimization and resource allocation and gradually converge to the industry’s technological frontier; this process essentially constitutes a catch-up effect. The second is green technological progress, which represents the outward expansion of the production technology frontier driven by firms’ R&D innovation, and forms a substantive technological breakthrough.

The driving effects of IIC on these two components are notably asymmetric. For green technical efficiency improvement, IIC exerts a direct and fast-acting impact. Enterprises can quickly unlock efficiency potential simply by connecting equipment to the network, deploying production management and control systems, and implementing process optimization and refined management, without incurring additional large-scale R&D investment. Existing studies on the green effects of digital technologies also find that the short-term benefits of digital transformation mostly stem from efficiency gains brought by management optimization (Shen and Zhang, 2024).

For green technological progress, by contrast, the role of IIC is relatively indirect. Breakthroughs in core green technologies require long-term R&D input and technological accumulation, and IIC can only provide data support and R&D assistance, rather than directly enabling leapfrog technological advancement. Driven by short-term business objectives, firms tend to prioritize immediate gains from capacity release and adopt a prudent attitude toward green R&D investment with a long payback period. Under this constraint, the driving effect of technological frontier breakthroughs on GTFP improvement may present a time-lag effect. Accordingly, this paper proposes Hypothesis 3:

Hypothesis 3The enhancement effect of IIC on GTFP is mainly realized through improving green technical efficiency, while the driving effect of green technological progress is not yet significant.

3.3.2 Formation logic of environmental costs: the offset between scale expansion and intensity decline

Changes in total enterprise pollution emissions are essentially the result of the combined effects of scale effect and intensity effect (Zhang et al., 2025). Specifically, the scale effect is determined by changes in total output, while the intensity effect is determined by the level of pollution emissions per unit of output. The relative strength of the two directly determines the final direction of change in total emissions. IIC affects both effects simultaneously, and the final direction of total emission change depends on the relative magnitude of the two effects.

In terms of the intensity effect, IIC can effectively reduce the level of pollution emissions per unit of output and achieve relative emission reduction through production process optimization and refined energy management and control. In terms of the scale effect, the capacity dividend released by IIC will drive enterprises to expand their output scale, exerting upward pressure on total emissions. Considering the development stage of China’s manufacturing industry, most enterprises are still in the scale expansion phase, and there remains sufficient room for market growth (Deng et al., 2024). The capacity released by efficiency improvement is mostly converted into actual output rather than idle capacity. Under this circumstance, the emission increment brought by scale expansion usually exceeds the emission reduction brought by intensity decline, ultimately pushing up total emissions. This is precisely the core logic behind the formation of the “green paradox”. Based on the above analysis, this paper sorts out the transmission path through which IIC simultaneously affects new quality productivity and environmental costs. The specific mechanism framework is shown in Figure 1.

FIGURE 1

4 Research design

4.1 Model specification

Given the staggered timing of sample enterprises’ inclusion in the IIC pilot program, this paper adopts the multi-period DID method to identify the causal effects of IIC on enterprises’ new quality productivity cultivation and pollution emissions. The specific model specifications are presented as follows:where subscript denotes the firm and subscript denotes the year. and are the explained variables, corresponding to the GTFP and pollutant emissions of firm in year , respectively. is the core explanatory variable, a dummy variable indicating whether firm is selected into the IIC pilot program in year . is a set of firm-level control variables. represents firm fixed effects and represents year fixed effects, which control for macroeconomic shocks and policy environment changes common to all sample firms. is the random disturbance term. In estimating Equations 1, 2, we cluster standard errors at the firm level to ensure valid inference in the presence of within-firm error correlations.

4.2 Variable definitions

4.2.1 Explained variables

This paper selects two types of explained variables, corresponding to the two research dimensions of new quality productivity improvement and environmental cost change.

The first explained variable is GTFP (), which serves as the core indicator for measuring enterprises’ new quality productivity. Different from traditional total factor productivity that only focuses on economic output, this indicator incorporates pollutant emissions as undesirable outputs into the efficiency accounting framework, capturing enterprises’ high-quality development performance more comprehensively. Following the measurement approach of Chen (2024) and Yin et al. (2025), we adopt the non-radial SBM-ML index for estimation. On the input side, the number of employees, net value of fixed assets, and converted enterprise energy consumption are selected to represent labor, capital and energy inputs, respectively. Operating revenue is taken as desirable output, and industrial SO2, wastewater and smoke-dust are included as undesirable outputs. Given limited firm-level emission data, we allocate city-level industrial energy and pollution data to firms by employment share, following standard practice in the literature. The allocation formula is shown in Equation 3:where is firm-level energy/pollution, is city-level industrial energy/pollution, is firm employees, and is city secondary industry employment. The base year value is set to 1, and annual GTFP is obtained by accumulating the ML index.

The second explained variable is total pollution emissions (). Notably, its raw data are directly sourced from firm-level environmental disclosures, independent from the city-allocated data used in GTFP calculation. Following Mao et al. (2022), we select COD, ammonia nitrogen, SO2 and NOx, and convert them into pollution equivalents per the official standard. The calculation formula is shown in Equation 4:

This yields the final firm-level total pollution emission indicator. With independent data sources and different pollutant sets, and avoid mechanical correlation and provide separate evidence. In terms of economic connotation, GTFP belongs to the efficiency dimension. It measures comprehensive productivity after accounting for environmental costs, and essentially reflects emission intensity per unit of output. The total pollution emission indicator belongs to the scale dimension. It reflects the absolute discharge level generated by overall enterprise production activities. This distinction between intensity efficiency and total emission scale forms the logical basis for the scale rebound effect analysis in this paper.

4.2.2 Explanatory variable

The core explanatory variable is the IIC (). Selected and announced in annual batches by the Ministry of Industry and Information Technology (MIIT) since 2017, the IIC pilot demonstration projects cover multiple development directions including network, platform, security and application, providing a quasi-natural experiment setting for identifying the effectiveness of enterprises’ IIC construction. Following the identification strategy of Chao et al. (2024), we determine the sample of pilot-participating enterprises by manually matching the applicant entities of pilot projects with the list of A-share listed companies. This variable takes a value of 1 in the year the enterprise is selected into the pilot and all subsequent years, and 0 otherwise. Pilot projects are selected by the MIIT through a formal review process. The core evaluation criteria cover technical innovation level, practical application effect, industrial replication value and network security capability. Selected firms usually have larger operation scale and better digital foundation. To mitigate the potential self-selection bias brought by non-random entry, we adopt the propensity score matching strategy in the robustness check.

4.2.3 Control variables

To mitigate endogeneity bias caused by omitted variables, we select the following control variables based on existing literature (Chen et al., 2025; Xu and Ren, 2025; Cobbinah et al., 2026). Firm size () is measured by the natural logarithm of total assets. Leverage ratio () is calculated as the ratio of total liabilities to total assets. Return on assets () is measured by the ratio of net profit to total assets. Cash flow ratio () is calculated as net cash flow from operating activities divided by total assets. Fixed asset ratio () is measured by the ratio of net fixed assets to total assets. Tobin’s Q () is calculated as the ratio of enterprise market value to asset replacement cost. Listing age () is measured by taking the natural logarithm of (current year minus listing year plus 1). Ownership concentration () is measured by the shareholding ratio of the largest shareholder in total share capital. Total asset turnover () is calculated as the ratio of operating revenue to average total assets.

4.3 Data sources and processing

This paper selects A-share listed companies on the Shanghai and Shenzhen Stock Exchanges from 2012 to 2024 as the initial research sample. Firm-level financial data are from Wind and CSMAR. Firm-level pollutant data for are from the CSMAR Environmental Research Database. City-level energy and “three wastes” data for GTFP are from the China City Statistical Yearbook and China Environmental Statistical Yearbook. The IIC pilot list is manually collected from MIIT announcements (2017–2022).

We apply the following screening procedures to the initial sample: ① Exclude listed firms in the financial and insurance sectors; ② Exclude firms under special treatment (ST and *ST) during the sample period; ③ Exclude observations with severe missing values for key variables; ④ Winsorize all continuous variables at the 1st and 99th percentiles to mitigate the interference of extreme values on estimation results. After the above processing, we finally obtain a panel dataset containing 17251 firm-year observations. To provide a transparent picture of the policy implementation timeline, Supplementary Table A1 reports the distribution of the 37 unique treated firms across the six pilot batches from 2017 to 2022. As shown, the treated group is not dominated by any single cohort, which further justifies our adoption of a staggered DID framework. In all regression specifications, standard errors are clustered at the firm level to account for potential serial correlation and heteroskedasticity within firms over time. The descriptive statistics of the main variables are reported in Table 1.

TABLE 1

VariablesNMeanSdMinMax
172511.0290.0930.8531.255
1725115.4790.22914.84516.000
172510.0130.1120.0001.000
1725122.7511.32919.56326.452
172510.4410.1980.0350.935
172510.0390.056−0.3750.255
172510.0530.064−0.1990.266
172510.2220.1650.0020.725
172511.8681.2010.78916.647
172512.6070.5850.6933.466
172510.3430.1500.0740.758
172510.6430.4350.0542.891

Descriptive statistics.

5 Empirical results

5.1 Baseline regression

Table 2 reports the baseline regression results of the IIC on enterprises’ GTFP and pollutant emissions. Specifically, columns (1) and (3) only control for firm fixed effects and year fixed effects, while columns (2) and (4) further include all firm-level control variables based on the above specification. According to the estimation results in column (2), the regression coefficient of the core explanatory variable is 0.028 and statistically significant at the 1% level, indicating that the IIC increases the GTFP of treatment group firms by approximately 2.8 percentage points on average compared with the control group. This finding demonstrates that IIC effectively improves enterprises’ green production efficiency through paths such as data-driven process optimization, equipment interconnection and network collaboration, which is consistent with the expectation of Hypothesis 1. Existing studies have provided micro-level empirical evidence for the positive role of digital technologies in empowering enterprises’ green development from two complementary dimensions: green investment efficiency and energy efficiency (Liao et al., 2024; Jing et al., 2025).

TABLE 2

Variables(1) (2) (3) (4)
0.080***0.028***0.198***0.072***
(0.007)(0.004)(0.017)(0.010)
​0.028***​0.067***
​(0.001)​(0.002)
​−0.071***​−0.188***
​(0.004)​(0.010)
​−0.059***​−0.163***
​(0.008)​(0.023)
​0.008​−0.012
​(0.006)​(0.017)
​−0.053***​−0.097***
​(0.005)​(0.013)
​−0.003***​−0.005***
​(0.000)​(0.001)
​0.162***​0.391***
​(0.001)​(0.004)
​0.030***​0.063***
​(0.006)​(0.016)
​−0.005***​−0.015***
​(0.002)​(0.005)
1.050***0.00615.632***13.104***
(0.001)(0.018)(0.004)(0.051)
YESYESYESYES
YESYESYESYES
17251172511725117251
0.49310.84530.45270.7921

Baseline regression results.

***, **, * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Standard errors in parentheses.

Column (4) reports the estimation results with total enterprise pollution emissions as the explained variable. The regression coefficient of the core explanatory variable is 0.072 and significantly positive at the 1% statistical level, indicating that the IIC increases the pollution emissions of treatment group firms by approximately 7.2% on average relative to the control group. This finding verifies that while IIC drives the improvement of new quality productivity, it is indeed accompanied by rising environmental costs, and the “green paradox” objectively exists at the firm level. The core explanatory logic is as follows: the leap in production efficiency brought by IIC stimulates the strong expansion of enterprises’ output scale, and the resulting scale rebound effect outweighs the pollution reduction per unit of output in magnitude, ultimately leading to a net increase in total pollution emissions. In a study on the energy rebound effect in Beijing, Kong et al. (2023) found that energy efficiency improvements from technological progress were not fully translated into energy savings, and the overall economy exhibited partial rebound or even a “backfire effect” in most years—that is, total energy consumption rose rather than fell. This conclusion also confirms that scale expansion induced by efficiency improvement may significantly weaken or even reverse the expected environmental benefits.

5.2 Robustness checks

5.2.1 Parallel trend test and sensitivity analysis

Causal inference using the multi-period DID method requires satisfying the parallel trend assumption. Subfigures (a) and (b) in Figure 2 present the parallel trend test results for the two explained variables, respectively. For GTFP, the estimated coefficients for all pre-policy periods are statistically insignificant, indicating no systematic trend difference between the treatment and control groups before the pilot policy was launched, and the parallel trend assumption holds. From the year of policy implementation onward, the estimated coefficients turn from negative to positive and rise continuously, suggesting that the improvement effect of the IIC pilot on GTFP is gradually released over time with an obvious dynamic cumulative characteristic. For pollution emissions, the coefficients for all pre-policy periods are also insignificant, and the emission trends of the two groups remain parallel ex ante. After policy implementation, the estimated coefficients quickly turn positive and remain significant, demonstrating that the scale rebound effect of pollution emissions emerges rapidly after the policy takes effect. Overall, both explained variables satisfy the parallel trend assumption, supporting the identification premise of the baseline regression.

FIGURE 2

The ex-ante trend test of the conventional event study method is limited by insufficient statistical power. Even if the test fails to reject the null hypothesis, it cannot fully eliminate the interference of potential trend deviations on estimation results (Roth et al., 2023). To further reinforce the reliability of our findings, we use the sensitivity analysis framework developed by Rambachan and Roth (2023) as a supplementary test. In Figure 3, subfigures (a) and (b) present the sensitivity analysis results for GTFP under the relative deviation bound and smoothness bound specifications, respectively, while subfigures (c) and (d) report the corresponding results for pollution emissions. The results confirm that the parallel trend assumption in this study is robust: potential deviations from the parallel trend are insufficient to overturn the core baseline findings, thus ruling out the possibility that ex ante trend differences account for the baseline results.

FIGURE 3

5.2.2 Placebo test

To eliminate the impact of unobservable omitted variables and random disturbances on the baseline regression results, we conduct a placebo test by randomly assigning fictitious treatment groups (Chetty et al., 2009). Following the standard design in existing literature (Dreber et al., 2024), we construct fictitious policy variables by randomly assigning treatment status to sample firms, and then repeat the full baseline estimation procedure. We iterate the above random sampling and regression process 500 times to obtain 500 estimated coefficients of the fictitious policy variables, and plot the kernel density distribution of the coefficients along with the corresponding p-value distribution, as presented in Figures 4A,B. The distribution patterns show that the estimated coefficients of the fictitious policy for both GTFP and pollution emissions are concentrated around zero, whereas the true estimated coefficients from the baseline regression fall in the right tail of the distribution. This confirms that the two core findings of this study are not driven by random factors, and the baseline regression results are robust.

FIGURE 4

5.2.3 Endogeneity treatment

First, we adopt the PSM-DID method to alleviate self-selection bias arising from non-random pilot selection. Taking all control variables from the baseline regression as matching covariates, we use the nearest-neighbor matching method to match each treatment group firm with control group firms with similar characteristics. The balance test results are shown in Figures 5A,B. After matching, the standardized bias of each covariate narrows substantially compared with the pre-matching level, and all values remain below the 5% critical value, indicating a favorable matching effect. Columns (1) and (2) in Table 3 report the re-estimation results based on the matched sample: the estimated coefficients of the IIC pilot on both GTFP and pollution emissions are positive and remain statistically significant at the 1% level, which is highly consistent with the baseline regression results.

FIGURE 5

TABLE 3

VariablesPSM-DIDControl variable lag by one period
​(1) (2) (3) (4)
0.028***0.071***0.025***0.065***
(0.004)(0.010)(0.004)(0.010)
−0.01313.025***0.092***13.260***
(0.019)(0.055)(0.018)(0.052)
YESYESYESYES
YESYESYESYES
YESYESYESYES
N16422164221592415924
R20.84330.79040.83030.7729

Endogeneity tests.

***, **, * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Standard errors in parentheses.

Second, to address endogeneity caused by contemporaneous reverse causality, we re-estimate the model with all firm-level control variables lagged by one period. The lagged specification eliminates the reverse impact of contemporaneous explained variables on firms’ operating characteristics, and mitigates endogeneity interference without excessive sample loss. The regression results are reported in columns (3) and (4) of Table 3. The magnitude and significance level of the core explanatory variable’s coefficient remain consistent with the baseline regression, further verifying the robustness of the core findings.

5.2.4 Excluding the influence of concurrent policies

During the sample period, a number of city-level development policies have been rolled out in China. Three of them—the Broadband China demonstration cities, national smart city pilots, and national innovative city pilots—partially overlap with IIC construction in terms of implementation timing and geographic scope, and all may exert impacts on enterprises’ production efficiency and environmental performance. To eliminate the interference of these concurrent policies on baseline estimation results, we construct dummy variables for each of the three policies and include them in the baseline regression model one by one for re-estimation. The regression results are presented in Table 4. After controlling for the three policy shocks separately, the core conclusions of this study remain robust, verifying that the driving effect of the IIC on new quality productivity and its accompanying environmental costs are not driven by other concurrent city-level policies.

TABLE 4

VariablesBroadband ChinaSmart cityInnovative city
(1) (2) (3) (4) (5) (6)
0.028***0.071***0.028***0.072***0.029***0.073***
(0.004)(0.010)(0.004)(0.010)(0.004)(0.010)
0.022***0.057***​​​​
(0.001)(0.004)​​​​
​​0.012***0.044***​​
​​(0.002)(0.007)​​
​​​​0.014***0.039***
​​​​(0.002)(0.006)
YESYESYESYESYESYES
YESYESYESYESYESYES
YESYESYESYESYESYES
172511725117251172511725117251
0.84820.79550.84550.79260.84570.7926

Excluding the impact of other policies.

***, **, * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Standard errors in parentheses. refers to the “Broadband China” pilot policy. refers to the Smart City pilot policy. refers to the Innovative City pilot policy.

5.2.5 Additional robustness checks

First, we replace the measurement of explained variables. For GTFP, we substitute it with the traditional total factor productivity indicator excluding undesirable outputs, to eliminate the interference of undesirable output accounting rules on efficiency estimation. For pollution emissions, the baseline regression calculates pollution equivalents based on four pollutants (COD, ammonia nitrogen, sulfur dioxide, nitrogen oxides); here we further include total nitrogen, total phosphorus and soot, and recalculate the total pollution equivalent of enterprises under a broader pollutant scope before taking the natural logarithm. The regression results after measurement replacement are shown in columns (1) and (2) of Table 5. The core coefficients of both explained variables remain statistically significant, indicating that the baseline conclusions do not rely on specific variable measurement methods or pollutant selection criteria.

TABLE 5

VariablesReplace the explained variableHigh-dimensional fixed effectsIncrease control variablesCSDID
(1) (2) (3) (4) (5) (6) (7) (8)
0.042*0.001***0.028***0.070***0.028***0.071***0.058*0.098**
(0.022)(0.000)(0.004)(0.010)(0.004)(0.010)(0.032)(0.039)
YESYESYESYESYESYESYESYES
YESYESYESYESYESYESYESYES
YESYESYESYESYESYESYESYES
NONOYESYESNONONONO
172511725117251172511725117251365365
0.82730.84560.84650.79330.84550.7923//

Other robustness tests results.

***, **, * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Standard errors in parentheses. Columns (7)–(8) report aggregated overall ATT, from the Callaway and Sant’Anna (2021) staggered-DID, estimator, using not-yet-treated firms as the comparison group. Wild-bootstrap standard errors (500 replications) are clustered at the firm level. The observation count for CSDID, columns reflects the number of group-time ATT estimates used in aggregation. Cohort-specific group-time ATT, results are shown in Supplementary Table A3.

Second, we control for high-dimensional fixed effects. Building on the baseline model specification, we further incorporate industry fixed effects to absorb the potential impacts of time-invariant industry-level factors, such as technical attributes and environmental regulation intensity. The regression results are presented in columns (3) and (4) of Table 5. The magnitude and significance level of the core explanatory variable’s coefficient remain largely unchanged, and the conclusion stays robust after controlling for industry-level fixed effects.

Third, we supplement control variables. To alleviate estimation bias caused by omitted firm governance variables, we add two extra variables to the baseline control set: the proportion of independent directors and capital intensity. The proportion of independent directors reflects corporate governance quality, which may affect enterprises’ digital transformation decisions and environmental management performance; capital intensity is directly associated with enterprises’ production modes and energy consumption intensity. Columns (5) and (6) of Table 5 report the estimation results with the supplementary variables. The coefficients of are 0.028 and 0.071 respectively, both significant at the 1% level, showing no significant difference from the baseline regression results.

A further concern is that the staggered rollout of the IIC pilot across cities constitutes a typical multi-period policy shock. The conventional two-way fixed effects DID estimator pools all cohorts into a single average treatment effect and implicitly assumes that the policy effect is homogeneous across adoption times. When treatment effects vary over time, this estimator can produce systematically biased coefficients because early treated firms may serve as controls for later treated firms in subsequent periods. To address this issue, we re-estimate the policy effects using the CSDID estimator proposed by Callaway and Sant'Anna (2021). This method groups firms according to their first treatment year, estimates cohort-specific average treatment effects for each period, and aggregates them using only not-yet-treated or never-treated firms as the comparison group, thereby avoiding the negative weighting problem. The results are reported in columns (7) and (8) of Table 5. The estimated ATT is 0.058 for green total factor productivity and 0.098 for pollution emissions, both of which are consistent with the baseline results in sign and statistical significance, indicating that the main conclusions are robust to treatment effect heterogeneity across adoption cohorts. Inference is based on wild-bootstrap standard errors clustered at the firm level with 500 replications.

5.3 Mechanism tests

5.3.1 Internal driver of new quality productivity: efficiency catch-up or technological breakthrough?

Following the measurement and decomposition logic of the SBM-Malmquist-Luenberger index, GTFP growth can be decomposed into two components: green technical efficiency improvement and green technological progress. Green technical efficiency reflects enterprises’ utilization of existing technologies, embodying the catch-up effect brought by management optimization and resource allocation improvement. Green technological progress represents the outward expansion of the production technology frontier, derived from substantive breakthroughs in green technology innovation. We take the decomposed green technical efficiency change index and green technological progress change index as explained variables respectively and estimate them with the baseline model. The results are presented in columns (1) and (2) of Table 6.

TABLE 6

VariablesGTFP decompositionPollution channels
​(1) Efficiency(2) Technical change(3) Revenue(4) Pollution intensity
0.030***−0.0020.040**−0.037*
(0.004)(0.002)(0.018)(0.020)
YESYESYESYES
YESYESYESYES
YESYESYESYES
17251172511725117251
0.84470.00070.89600.8272

Mechanism test results.

***, **, * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Standard errors in parentheses.

The estimation results in column (1) show that the regression coefficient of the IIC on green technical efficiency change is 0.030, significantly positive at the 1% statistical level. In column (2), the coefficient of the policy on green technological progress is negative and statistically insignificant. This indicates that the enhancement effect of IIC on enterprises’ GTFP is mainly achieved through the improvement of green technical efficiency, while the driving effect of green technological progress has not yet become apparent.

5.3.2 Causes of environmental costs: scale rebound and effect decomposition

The rise in total pollutant emissions may stem from two distinct mechanisms. First, improved production efficiency stimulates enterprises to expand production scale, and the resulting additional emissions offset or even exceed the emission reduction achieved per unit of output. Second, the pollution emission intensity per unit of output itself fails to decline or even rises. Clarifying the relative role of the two mechanisms is central to understanding the causes of the “green paradox”. This paper conducts tests with two explained variables: the natural logarithm of enterprise operating revenue as a proxy for output scale, and pollution emission intensity calculated as total pollutant emissions divided by operating revenue.

The estimation results in column (3) of Table 6 show that the regression coefficient of the core explanatory variable on enterprise operating revenue is 0.040 and significantly positive at the 5% statistical level, indicating that the IIC has indeed induced a significant expansion of enterprises’ output scale, verifying the precondition for the scale rebound effect. Column (4) reports the regression results of pollution emission intensity. The coefficient of is significantly negative, suggesting that IIC construction has actually reduced the level of pollution emissions per unit of output, and the environmental cost corresponding to each unit of enterprise output has decreased. This finding aligns well with the classic rebound effect theory in energy economics (Berkhout et al., 2000).

We further quantify the rebound magnitude following the classical energy rebound effect framework. Because pollution intensity is constructed as total emissions divided by revenue, the scale and intensity components are inherently linked through an accounting relationship. Our causal estimates are obtained from separate DID regressions under the policy shock. If output scale remained unchanged, the estimated 3.7 percent reduction in pollution intensity would imply an equivalent decline in total emissions. In practice, however, IIC-induced output expansion leads to an actual 7.2 percent increase in total emissions. The implied rebound magnitude is approximately 295 percent ((3.7% + 7.2%)/3.7%), far exceeding the 100 percent backfire threshold. We note that this estimate combines coefficients from separate regressions and should be interpreted as an approximate magnitude rather than a precise accounting decomposition. Nevertheless, the direction and relative magnitude of the estimates consistently indicate that the emission increment from scale expansion fully offsets the intensity reduction and causes a net increase in total emissions, providing supportive quantitative evidence for the scale rebound mechanism.

To verify the robustness of the conclusion on declining pollution intensity, we separately test the revenue-based emission intensity of four pollutants: sulfur dioxide, nitrogen oxides, ammonia nitrogen and chemical oxygen demand. The results are detailed in Supplementary Table A2. The regression results show that the emission intensity of all four pollutants declines to varying degrees. Among them, the decline of sulfur dioxide and nitrogen oxides is significant at the 5% and 10% levels respectively, while the decline of ammonia nitrogen and chemical oxygen demand fails to pass the significance test. This indicates that IIC delivers a more prominent intensity reduction effect on waste gas pollutants and a relatively limited effect on wastewater pollutants, which overall supports the basic judgment of declining pollution intensity per unit of output.

5.4 Heterogeneity analysis

5.4.1 Heterogeneity by ownership

Ownership nature is a core factor affecting firms’ policy responsiveness and operational decision-making logic. We divide the sample firms into state-owned enterprises (SOEs) and non-state-owned enterprises, and employ the interaction term to identify the heterogeneous moderating effect of ownership. The regression results are presented in columns (1) and (2) of Table 7. The coefficient of the interaction term in column (1) is 0.039 and statistically significant at the 1% level, indicating that the GTFP improvement effect of IIC is significantly stronger among SOEs. The interaction coefficient in column (2) is 0.080 and significant at the 1% level, which demonstrates that SOEs achieve greater productivity growth while suffering a more prominent emission rebound effect.

TABLE 7

VariablesOwnershipManufacturingTech-level
(1) (2) (3) (4) (5) (6)
0.0080.030**−0.0040.0050.053***0.151***
(0.005)(0.015)(0.007)(0.021)(0.010)(0.030)
0.039***0.080***​​​​
(0.007)(0.021)​​​​
​​0.043***0.088***​​
​​(0.008)(0.024)​​
​​​​−0.028**−0.091***
​​​​(0.011)(0.032)
0.00513.101***0.00113.092***−0.00213.084***
(0.018)(0.051)(0.018)(0.052)(0.018)(0.052)
YESYESYESYESYESYES
YESYESYESYESYESYES
YESYESYESYESYESYES
172511725117251172511725117251
0.84560.79230.84560.79230.84550.7923

Heterogeneity analysis results.

***, **, * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Standard errors in parentheses.

SOEs usually have sufficient capital reserves, sound IT infrastructure and standardized management systems, which create favorable initial conditions for IIC platform deployment and data-driven process reengineering, and allow firms to fully absorb efficiency dividends from digital technologies (Zhang, 2024). Yue et al. (2025) study the impact of digital transformation on energy efficiency and find that although digital transformation significantly reduces energy consumption, its effects vary substantially across different corporate governance structures.

5.4.2 Heterogeneity by industrial sector

Manufacturing is the primary application scenario of IIC and the dominant sector of energy consumption and pollutant emissions. To test industrial heterogeneity, we construct an interaction term between the IIC pilot dummy variable and the manufacturing dummy variable for regression, and the results are shown in columns (3) and (4) of Table 7. The interaction coefficient in column (3) is 0.043 and significant at the 1% level, which suggests that the GTFP-promoting effect of IIC is more pronounced for manufacturing firms. The interaction coefficient in column (4) is 0.088 with 1% statistical significance, revealing that manufacturing enterprises also experience a stronger emission rebound effect.

Manufacturing production systems are characterized by diverse equipment, long process chains, and intensive material and energy flows, which provide abundant connection nodes and optimization space for IIC (Tao et al., 2019). However, manufacturing production features strong scalability, and capacity released by efficiency gains is more likely to be converted into output growth rather than idle capacity, so the scale rebound effect is more prominent in the manufacturing sector.

5.4.3 Heterogeneity by technological intensity

High-tech and non-high-tech industries differ greatly in technological absorptive capacity, R&D investment intensity and material intensity of production processes, which may lead to heterogeneous effects of the IIC policy. We construct the interaction term to examine the moderating effect of technological intensity, and the estimation results are reported in columns (5) and (6) of Table 7. The interaction coefficient in column (5) is −0.028 and significant at the 5% level, which implies that the GTFP growth of high-tech industries is significantly weaker than that of non-high-tech industries. The interaction coefficient in column (6) is −0.091 and significant at the 1% level, indicating that the incremental pollutant emissions of high-tech industries are also markedly lower.

This pattern can be explained by the law of diminishing marginal returns. High-tech industries already possess advanced technological foundations and high digitalization levels, so the marginal productivity improvement brought by IIC is relatively limited. Brynjolfsson et al. (2018) argue that the productivity effects of digital technologies are unevenly distributed and exhibit obvious diminishing marginal returns. Dedrick et al. (2003) also confirm in their research on IT investment and productivity that returns to IT investment differ drastically across industries, and diminishing marginal returns are more evident in technology-intensive sectors.

6 Extended discussion

6.1 Heterogeneous moderating effect of market power

Market power determines firms’ pricing capacity and output adjustment flexibility in product markets, which further shapes the allocation of efficiency gains and the magnitude of pollution scale rebound. Following the approach of Elzinga and Mills (2011), we use the Lerner index to measure firm-level market power, and add the interaction term of the policy dummy and market power into regression models to identify its moderating mechanism. Columns (1) and (2) of Table 8 present the test results for the moderating effect of market power. The interaction term yields significantly negative coefficients in both the GTFP and pollution emission regressions. This implies that stronger market power weakens the GTFP-promoting effect of IIC, and simultaneously reduces the incremental pollutant emissions induced by the policy.

TABLE 8

VariablesMarket powerManagerial myopia
​(1) GTFP(2) Pollution(3) Pollution(4) Pollution intensity
0.043***0.096***0.109***0.085***
(0.006)(0.017)(0.016)(0.027)
−0.130***−0.214*​​
(0.039)(0.110)​​
​​−0.007***−0.011***
​​(0.002)(0.003)
YESYESYESYES
YESYESYESYES
YESYESYESYES
17251172511446414464
0.84540.79220.79450.8423

Market power and managerial myopia.

***, **, * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Standard errors in parentheses.

Such heterogeneity stems from different operational incentives across market structures. Firms in highly competitive markets face stronger pressure to improve efficiency and expand output. They make fuller use of IIC to optimize production processes, which leads to larger GTFP gains. They also convert efficiency gains more rapidly into capacity expansion to seize market share, resulting in a more pronounced emission rebound. In contrast, firms with stronger market power face less competitive pressure. They have weaker incentives to pursue efficiency breakthroughs or rapid capacity expansion. Both the productivity gain and the emission rebound are thus relatively smaller. This finding aligns with the escape competition logic that competitive pressure drives faster efficiency improvement and capacity expansion (Aghion et al., 2005). Liang and Sun (2024) also find that digital transformation delivers a stronger green innovation effect for firms in more competitive industries.

6.2 Catalytic effect of managerial myopia

Managers’ decision-making horizons directly shape firms’ long-term investment preferences and environmental strategic arrangements. Myopic managers tend to prioritize allocating efficiency dividends generated by IIC to expanding current output, thereby intensifying the emission rebound effect. We use the share of R&D expenditure in operating revenue as an inverse indicator of firms’ focus on short-term performance. Lower R&D intensity generally reflects greater emphasis on near-term operational outcomes, which correlates with higher managerial myopia. We construct its interaction term with the policy dummy to examine heterogeneous effects across different levels of long-term orientation. Columns (3) and (4) of Table 8 present the corresponding regression results. The interaction term yields significantly negative coefficients in both the total pollution emission and emission intensity regressions. This indicates that firms with lower R&D intensity, or weaker long-term orientation, see larger emission increments and higher pollution intensity after the policy shock. The result suggests that short-term-focused operational tendencies may amplify the environmental costs of IIC. Li et al. (2024) show that managerial myopia substantially curbs corporate GTFP, and one key transmission channel lies in the reduction of both the quantity and quality of green innovation. Wang and Feng (2025) also find that the promotional effect of market competition on firms’ digital transformation varies sharply between state-owned and non-state-owned enterprises, as well as high-tech and non-high-tech firms, which verifies that firms’ internal micro characteristics act as critical moderators for the exertion of digital policy effects.

6.3 Factor bias of capital-labor structure

As a typical carrier of capital-biased technological progress, IIC not only changes firms’ overall production efficiency, but also profoundly reshapes the factor input mix and relative proportions in production processes. Columns (1) and (2) of Table 9 report the estimation results regarding capital-labor structure. The IIC exerts an insignificant impact on firm employment scale, while its coefficient on the capital-labor ratio is 0.138 and significantly positive at the 1% statistical level. IIC promotes capital deepening while keeping total employment roughly stable, and its technological bias is closer to a labor-augmenting pattern rather than a labor-replacing one (Acemoglu and Restrepo, 2020). Although capital deepening cuts energy consumption intensity per unit output, expanded capital stock unlocks greater production potential and lays a material foundation for the scale rebound effect. This result differs from cross-country evidence on the employment impacts of automation documented by Autor and Salomons (2018), reflecting the unique features of IIC as a networked collaborative technology. Tao et al. (2024) also find that corporate digital transformation significantly stimulates green innovation via three channels: improved environmental management, human capital upgrading, and higher resource allocation efficiency.

TABLE 9

VariablesCapital-labor structureGreen innovation
​(1) Employment(2) Capital-labor ratio(3) Green patents(4) Green patents
−0.0100.138***0.390***0.427***
(0.030)(0.038)(0.066)(0.069)
​​​−0.377*
​​​(0.228)
YESYESYESYES
YESYESYESYES
YESYESYESYES
17100171001725117251
0.45780.44370.22870.2289

Capital-labor structure and green innovation.

***, **, * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Standard errors in parentheses.

6.4 Crowding-out effect on green innovation

The previous mechanism analysis shows that GTFP growth is mainly driven by efficiency catch-up rather than technological frontier breakthrough. This leads us to examine how output growth pressure shapes the green innovation effect of IIC. We adopt the total number of green patent applications as the explained variable for empirical testing, and columns (3) and (4) of Table 9 present the regression results for green innovation. Column (3) shows that the coefficient of the IIC on green patent applications is significantly positive, demonstrating that IIC generally fosters corporate green innovation. In column (4), the interaction term of revenue growth rate and the policy dummy is significantly negative. This shows that the green innovation promotion effect of IIC is weakened for firms with faster revenue growth. One plausible explanation is that rapid output expansion may divert limited corporate resources away from long-term green innovation activities under resource constraints. Despite this heterogeneous moderation, the overall promoting effect of IIC on green innovation remains statistically significant and economically meaningful. We treat this as suggestive heterogeneity evidence rather than strict proof of R&D crowding out. In their research on financing constraints and innovation, Hottenrott and Peters (2012) argue that firms prioritize resource allocation to capacity expansion when market opportunities emerge. Sun et al. (2025) further verify that digital transformation drives corporate green innovation through three channels: optimized environmental management, upgraded human capital, and improved resource allocation efficiency. This finding also provides supplementary interpretative logic for the “green paradox” identified earlier.

7 Conclusions and policy implications

7.1 Conclusions

Taking the IIC pilot policy as a quasi-natural experiment, this paper adopts the multi-period DID model to identify the dual effects of IIC construction on corporate GTFP and total pollutant emissions. We further decompose the internal channels driving new quality productivity growth and the formation mechanism of rising total pollution emissions. Overall, the empirical results reveal that IIC construction significantly boosts enterprises’ new quality productivity, yet it is accompanied by a surge in total pollutant emissions. This pattern constitutes a typical backfire effect and verifies a firm-level green paradox phenomenon driven by scale expansion.

Mechanism tests indicate that the promotion of GTFP driven by IIC mainly stems from green technical efficiency catch-up via management optimization and improved resource allocation, while the contribution of breakthroughs in technological frontiers remains limited. The rise in total pollution emissions does not result from higher emission intensity per unit output, but from the scale rebound effect triggered by output expansion stimulated by efficiency gains. The additional emissions brought by expanded production scale outweigh the emission reduction benefits from lower pollution intensity per unit output.

Heterogeneity analysis confirms that ownership type, industrial sector and technological intensity moderate the dual policy effects. State-owned enterprises, manufacturing firms and non-high-tech industries experience more significant GTFP growth as well as stronger emission rebound.

Extended discussions further demonstrate that market power, managerial myopia and capital-labor factor structure shape the magnitude of both policy effects. Meanwhile, firms facing faster output growth exhibit a weaker green innovation response, suggestive of resource diversion under growth pressure.

7.2 Policy recommendations

First, governments should continuously promote the large-scale application of IIC to fully leverage its advantages in optimizing production processes and improving energy efficiency, so as to consolidate the digital foundation for the development of new quality productivity. In addition, policymakers should abandon the single efficiency-oriented evaluation framework and incorporate total pollution emission control into the assessment system of digital industrial policies. Total emission constraint indicators should be added to the evaluation criteria of IIC pilots, guiding enterprises to fulfill emission reduction targets simultaneously during digital upgrading.

Second, given that current efficiency gains from IIC mainly rely on technical efficiency catch-up rather than technological breakthroughs, governments should scale up support for green technology R&D under digital scenarios. Special subsidies and green tax incentives can be adopted to guide firms to allocate digital dividends to green innovation and accelerate breakthroughs in green technological frontiers. Differentiated environmental regulations should be implemented to address the scale rebound effect: strict standards for clean production and total pollutant discharge should be imposed on firms expanding capacity through digital transformation, preventing efficiency dividends from being fully converted into unregulated capacity expansion.

Third, for state-owned enterprises and traditional manufacturing firms, governments should strengthen environmental regulations while supporting their deepened IIC adoption, so that digital advantages can be more translated into emission reduction capacity and mitigate the adverse environmental impacts of scale rebound. For high-tech industries, policymakers should further tap into the potential of IIC to empower technological innovation and encourage integrated breakthroughs combining digital and green low-carbon technologies. Moreover, firms should be guided to establish long-term decision-making mechanisms to alleviate the negative impacts of managerial myopia on environmental performance.

7.3 Limitations and future research directions

This study has several limitations that offer avenues for future research. First, our sample only includes A-share listed firms on Shanghai and Shenzhen stock exchanges. A large number of unlisted small and medium-sized industrial enterprises are excluded. Meanwhile, the number of IIC pilot enterprises is relatively limited, resulting in a small share of treated observations in the full sample. The generalizability of our findings requires further verification with broader samples as the pilot scope expands. Second, restricted by the availability of firm-level refined pollution data, we convert city-level pollution data into firm-level emissions based on the proportion of corporate employees in urban employment. Although this conversion method is widely accepted in existing literature, measurement errors cannot be completely eliminated. Third, this paper only interprets the logic of the “green paradox” from the internal operational perspective of individual firms. Future research can extend to external channels such as industrial chain transmission and regional spatial spillovers, providing more comprehensive empirical evidence for resolving the green paradox and advancing the coordinated development of digitalization and green low-carbon transition.

Statements

Data availability statement

The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author.

Author contributions

WW: Resources, Writing – review and editing, Conceptualization, Methodology. CG: Data curation, Conceptualization, Methodology, Writing – original draft, Software.

Funding

The author(s) declared that financial support was not received for this work and/or its publication.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenvs.2026.1931075/full#supplementary-material

References

  • 1

    AcemogluD.RestrepoP. (2020). Robots and jobs: evidence from US labor markets. J. Polit. Econ.128, 2188–2244. 10.1086/705716

  • 2

    AghionP.BloomN.BlundellR.GriffithR.HowittP. (2005). Competition and innovation: an inverted-U relationship. Q. J. Econ.120, 701–728. 10.1093/qje/120.2.701

  • 3

    AutorD.SalomonsA. (2018). Is automation labor-displacing? Productivity growth, employment, and the labor share. Natl. Bur. Econ. Res. Work. Pap. (w24871). 10.3386/w24871

  • 4

    AzumahS. S.TettehF. K.ManuD.Owusu KwatengK. (2026). Circular supply chain practices and sustainable performance: the role of supply chain ethical leadership and green knowledge sharing. Bus. Strateg. Environ.35, 2418–2433. 10.1002/bse.70277

  • 5

    BerkhoutP. H.MuskensJ. C.VelthuijsenJ. W. (2000). Defining the rebound effect. Energy Policy28, 425–432. 10.1016/S0301-4215(00)00022-7

  • 6

    BibiS.KhanA.XuF. B.HuJ. F.HussainS. (2026). Integrating digitalization, environmental innovations, and green energy supply to ensure green production in China’s textile and fashion industry: environmental policy and laws optimization perspective. Environ. Dev. Sustain.28, 11767–11807. 10.1007/s10668-024-05417-4

  • 7

    BrynjolfssonE.RockD.SyversonC. (2018). “Artificial intelligence and the modern productivity paradox: a clash of expectations and statistics,” in The Economics of Artificial Intelligence: An Agenda (Chicago, IL, USA: University of Chicago Press), 23–57.

  • 8

    CallawayB.Sant’AnnaP. H. C. (2021). Difference-in-differences with multiple time periods. J. Econom.225, 200–230. 10.1016/j.jeconom.2020.12.001

  • 9

  • 10

    ChaoX. J.ZhouW. H.LiuY. Y. (2024). Industrial internet and total factor productivity of manufacturing enterprises. Bus. Manag. J.46 (7), 5–19. 10.19616/j.cnki.bmj.2024.07.001

  • 11

    ChenQ. (2024). Impact of sentiment tendency of media coverage on corporate green total factor productivity: the moderating role of environmental uncertainty. Financ. Res. Lett.65, 105637. 10.1016/j.frl.2024.105637

  • 12

    ChenY.LiM. (2024). How does the digital transformation of agriculture affect carbon emissions? Evidence from China’s provincial panel data. Humanit. Soc. Sci. Commun.11, 1–17. 10.1057/s41599-024-03223-x

  • 13

    ChenH.YuZ.HuS. (2025). Digital economy, human capital accumulation, and corporate green total factor productivity: based on strategic emerging industries. Int. Rev. Financ. Anal.103, 104152. 10.1016/j.irfa.2025.104152

  • 14

    ChengJ. J. (2023). Industrial internet promoting industrial chain modernization: theoretical logic and breakthrough paths. Mod. Econ. Res.1 (1), 93–102. 10.13891/j.cnki.mer.2023.01.006

  • 15

    ChettyR.LooneyA.KroftK. (2009). Salience and taxation: theory and evidence. Am. Econ. Rev.99, 1145–1177. 10.1257/aer.99.4.1145

  • 16

    ChoukseyA.RajanA. K.GurjarV.TiwariR.MishraP. K. (2026). The green paradox: the climate, environmental, and sustainability implications of artificial intelligence. Glob. Environ. Change Adv.6, 100029. 10.1016/j.gecadv.2025.100029

  • 17

    CobbinahJ.OseiA.AmoahJ. O. (2026). Innovating for a greener future: do digital transformation and innovation capacity drive enterprise green total factor productivity in the knowledge economy?J. Knowl. Econ.17, 325–363. 10.1007/s13132-025-02673-1

  • 18

    DedrickJ.GurbaxaniV.KraemerK. L. (2003). Information technology and economic performance: a critical review of the empirical evidence. ACM Comput. Surv.35, 1–28. 10.1145/641865.641866

  • 19

    DengX. J.YuanM. Q.LuoC. Y. (2024). Corporate digital transformation, market competition, and the environmental performance — microevidence from Chinese manufacturing. Bus. Strateg. Environ.33, 3279–3298. 10.1002/bse.3660

  • 20

    DreberA.JohannessonM.YangY. (2024). Selective reporting of placebo tests in top economics journals. Econ. Inq.62, 921–932. 10.1111/ecin.13217

  • 21

    DuX.CaiZ.ZengQ. (2025). The backfired effect of bonus depreciation policy on corporate environmental performance: evidence from a quasi-natural experiment. Span. J. Financ. Acc.54, 526–557. 10.1080/02102412.2025.2557057

  • 22

    ElzingaK. G.MillsD. E. (2011). The lerner index of monopoly power: origins and uses. Am. Econ. Rev.101, 558–564. 10.1257/aer.101.3.558

  • 23

    Font VivancoD.Freire-GonzálezJ.GalvinR.SantariusT.WalnumH. J.MakovT.et al (2022). Rebound effect and sustainability science: a review. J. Ind. Ecol.26, 1543–1563. 10.1111/jiec.13295

  • 24

    GaoX.LiS. (2025). A dynamic evolution and spatiotemporal convergence analysis of the coordinated development between new quality productive forces and China’s carbon total factor productivity. Sustainability17, 3137. 10.3390/su17073137

  • 25

    GedaM. W.KwongC. K. (2018). Forecasting of Used Product Returns for Remanufacturing. 2018 IEEE International Conference on Industrial Engineering and Engineering Management IEEM, 889–893. 10.1109/IEEM.2018.8607362

  • 26

    GreeningL. A.GreeneD. L.DifiglioC. (2000). Energy efficiency and Consumption—The rebound Effect—A survey. Energy Policy28, 389–401. 10.1016/S0301-4215(00)00021-5

  • 27

    GuoD.YangD. (2025). How does the integration of the digital economy and the real economy mitigate overcapacity: evidence from China. Environ. Dev. Sustain., 1–29. 10.1007/s10668-025-06732-0

  • 28

    HanD. R.LvX. L.DongH. Z.WangH. Y. (2023). Threshold effect of digital technology on the synergy of carbon reduction and pollution reduction. Resour. Sci.45 (11), 2130–2143. 10.18402/resci.2023.11.03

  • 29

    HottenrottH.PetersB. (2012). Innovative capability and financing constraints for innovation: more money, more innovation?Rev. Econ. Stat.94, 1126–1142. 10.1162/REST_a_00227

  • 30

    JianW.HuangW.YamakaW.LiuJ. (2023). Internet development and green total factor productivity: new evidence of mediation and threshold effects. Sustainability15, 12438. 10.3390/su151612438

  • 31

    JingP.LiS.WangM. (2025). Digital empowerment, industry chain integration and corporate energy efficiency. Energy Econ.145, 108446. 10.1016/j.eneco.2025.108446

  • 32

    KongL.MuX.HuG.TuC. (2023). Will energy efficiency improvements reduce energy consumption? Perspective of rebound effect and evidence from beijing. Energy263, 125665. 10.1016/j.energy.2022.125665

  • 33

    LiN.LiR.YuS. (2024). Managerial myopia and enterprise green total factor productivity: perspectives on the supervisory effect and incentive effect. Sustainability16, 7144. 10.3390/su16167144

  • 34

    LiangP.SunX. (2024). Does digital transformation promote the green innovation of China’s listed companies?Environ. Dev. Sustain.26, 22199–22235. 10.1007/s10668-024-05059-6

  • 35

    LiaoF.HuY.SunY.YeS. (2024). Does digital empowerment affect corporate green investment efficiency?Environ. Dev. Sustain.26, 23085–23111. 10.1007/s10668-023-03591-5

  • 36

    LinB.TengY. (2023). Digital revolution: does industrial chain digitalization lead the energy-saving wave?Sustain. Energy Technol. Assess.60, 103516. 10.1016/j.seta.2023.103516

  • 37

    LiuH.LiX. (2025). How digital technology can improve new quality productive forces? — perspective of total factor agricultural carbon productivity. J. Asian Econ.98, 101921. 10.1016/j.asieco.2025.101921

  • 38

    LiuY.LiuN.HuoY. (2025). Impact of digital technology innovation on carbon emission reduction and energy rebound: evidence from the Chinese firm level. Energy320, 135187. 10.1016/j.energy.2025.135187

  • 39

    LuM.WangK. (2024). The direct energy rebound effects for manufacturing and service sectors in China: evidence from firm-level estimations. J. Clean. Prod.474, 143617. 10.1016/j.jclepro.2024.143617

  • 40

    MaoJ.GuoY. Q.CaoJ.XuJ. W. (2022). Local government financing vehicle debt and environmental pollution control. J. Manag. World38 (1), 96–118. 10.19744/j.cnki.11-1235/f.2022.0147

  • 41

    PingW. Y.ShiH. A.LuoL. Q. (2025). Impact of industrial internet pilot policies on green transformation of manufacturing enterprises. J. Stat. Inf.40, 29–45. 10.20207/j.cnki.1007-3116.2025.0040

  • 42

    RambachanA.RothJ. (2023). A more credible approach to parallel trends. Rev. Econ. Stud.90, 2555–2591. 10.1093/restud/rdad056

  • 43

    RothJ.Sant’AnnaP. H.BilinskiA.PoeJ. (2023). What’s trending in difference-in-differences? A synthesis of the recent econometrics literature. J. Econom.235, 2218–2244. 10.1016/j.jeconom.2023.03.008

  • 44

    ShenY.ZhangX. (2024). Towards a low-carbon and beautiful world: assessing the impact of digital technology on the common benefits of pollution reduction and carbon reduction. Environ. Monit. Assess.196, 695. 10.1007/s10661-024-12860-3

  • 45

    SunZ.ZhaoL.MehrotraA.SalamM. A.YaqubM. Z. (2025). Digital transformation and corporate green innovation: an affordance theory perspective. Bus. Strateg. Environ.34, 433–449. 10.1002/bse.3991

  • 46

    TaoF.QiQ.WangL.NeeA. Y. C. (2019). Digital twins and cyber–physical systems toward smart manufacturing and industry 4.0: correlation and comparison. Engineering5, 653–661. 10.1016/j.eng.2019.01.014

  • 47

    TaoA.WangC.ZhangS.KuaiP. (2024). Does enterprise digital transformation contribute to green innovation? micro-Level evidence from China. J. Environ. Manag.370, 122609. 10.1016/j.jenvman.2024.122609

  • 48

    WangY. (2026). Deconstructing the green productivity paradox: how AI-enabled green innovation shapes efficiency and technological change. Technol. Forecast. Soc. Chang.230, 124757. 10.1016/j.techfore.2026.124757

  • 49

    WangM.FengL. (2025). Market competition and corporate digital transformation: the promoting role and mechanism of government subsidies. Financ. Res. Lett.80, 107354. 10.1016/j.frl.2025.107354

  • 50

    WangY. S.WuS. L. (2025). A study on configuration pathways: industrial internet policies boosting regional innovation resource allocation efficiency: from a regional alliance perspective. J. Ind. Technol. Econ.44 (6), 17–26. 10.3969/j.issn.1004-910X.2025.06.002

  • 51

    WangH. J.YangY.TangK. (2026). Green government procurement and firms’ green total factor productivity: causal inference based on double machine learning. J. Manag.5, 1–17. 10.19808/j.cnki.41-1408/F.2026.0029

  • 52

    WenH.ZhongQ.LeeC. C. (2022). Digitalization, competition strategy and corporate innovation: evidence from Chinese manufacturing listed companies. Int. Rev. Financ. Anal.82, 102166. 10.1016/j.irfa.2022.102166

  • 53

    WuJ. Y.ZhaoH. W. (2026). The impact of digital technology innovation on firms' green total factor productivity: evidence from corporate digital patents. Commer. Res.3 (3), 1–12. 10.13902/j.cnki.syyj.20260312.001

  • 54

    XiaoY.FuL.XuY.ShiX. (2026). Industrial internet and green total factor productivity: evidence from a quasi-natural experiment in China's renewable energy industry. Appl. Econ., 1–18. 10.1080/00036846.2026.2701325

  • 55

    XuR.RenL. (2025). Natural disaster, tax avoidance, and corporate pollution emissions: evidence from China. J. Bus. Ethics197, 195–217. 10.1007/s10551-024-05716-w

  • 56

    XuY.ZhaoF.YangL. (2026). How data factor marketisation promotes green and clean energy adoption: the influence of fossil energy use and green technological innovation. J. Innov. Knowl.16, 101025. 10.1016/j.jik.2026.101025

  • 57

    YiX.LuS.LiD.LiuW. (2024). Manufacturing enterprises digital collaboration empowered by industrial internet platform: a multi-agent stochastic evolutionary game. Comput. Ind. Eng.194, 110415. 10.1016/j.cie.2024.110415

  • 58

    YinS.LuY.SongH.LiaoY.GuoS. (2025). Measuring green total factor productivity in China’s power industry based on super-efficiency SBM and GML index models. Sustainability18, 38. 10.3390/su18010038

  • 59

    YueS.YeG.KhalidF.WangT. (2025). From digital transformation to energy efficiency: ownership concentration’s hidden role in driving climate solutions. Humanit. Soc. Sci. Commun.12, 1–14. 10.1057/s41599-025-05911-8

  • 60

    ZhangS. (2024). The impact of digital transformation on ESG performance and the moderation of mixed-ownership reform: the evidence from Chinese state-owned enterprises. Corp. Soc. Responsib. Environ. Manag.31, 2195–2210. 10.1002/csr.2656

  • 61

    ZhangQ.MaoZ. (2026). The impact of corporate digital transformation on green total factor productivity — the role of environmental regulation. Sustainability18, 2048. 10.3390/su18042048

  • 62

    ZhangY.HuS.ChenL. (2024). Internet technology adoption and firm energy efficiency: evidence from China. Technol. Forecast. Soc. Chang.201, 123214. 10.1016/j.techfore.2024.123214

  • 63

    ZhangY.ShiY.LiX.WangH. (2025). Environmental monitoring and enterprise reduction of pollution and carbon emission: a micro-level examination based on energy scale and intensity effects. Carbon Balance Manag.21, 1–21. 10.1186/s13021-025-00358-5

  • 64

    ZhangJ.ShanY.JiangS.XinB.MiaoY.ZhangY. (2026). Study on the coordinated development degree of new quality productivity and manufacturing carbon emission efficiency in provincial regions of China. Environ. Dev. Sustain.28, 8601–8635. 10.1007/s10668-024-05321-x

  • 65

    ZhuY.LanM. (2023). Digital economy and carbon rebound effect: evidence from Chinese cities. Energy Econ.126, 106957. 10.1016/j.eneco.2023.106957

Summary

Keywords

green total factor productivity, industrial internet construction, new quality productivity, pollution emissions, rebound effect

Citation

Wei W and Gao C (2026) Can industrial internet drive “new quality productivity” without fueling pollution? A paradox. Front. Environ. Sci. 14:1931075. doi: 10.3389/fenvs.2026.1931075

Received

07 July 2026

Revised

22 August 2026

Accepted

27 August 2026

Published

11 September 2026

Volume

14 - 2026

Edited by

Guifu Chen, Xiamen University, China

Reviewed by

Yuk Ming Tang, Hong Kong Polytechnic University, Hong Kong SAR, China

Tianren Li, Hunan University, China

Updates

Copyright

*Correspondence: Chen Gao,

Disclaimer

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.

Outline

Figures

Cite article

Copy to clipboard


Export citation file


Share article

Article metrics