Based on the resource-based theory (Wernerfelt, 1984), firms’ innovation activities are characterized by long input-output cycles, high risk, and the non-recoverability of initial investments. These activities typically require substantial amounts of capital, human resources, experimental equipment, and other resources to maintain a competitive advantage (Barney, 1991; Peteraf, 1993; Bharadwaj, 2000). In particular, compared with traditional green innovation, BGI emphasizes exploration and breakthroughs in new technological fields. They typically entail higher levels of risk and uncertainty, and the implementation of its R&D activities faces greater constraints in terms of innovation resources, such as capital and talent. However, IM has the potential to transform firms’ traditional production models and innovation processes (Ribeiro-Navarrete et al., 2021) and to invigorate green innovation of firms (Jin and Chen, 2024). This will crowd in corporate R&D resources, thereby providing ample resource support for enterprises to break the path dependence of technology choice and foster BGI.

First, IM can drive enterprise’s digital transformation (Zhou et al., 2018), aiding in the stimulation of their green innovation power potential. On the one hand, IM has created a lot of data and changed the innovation model of enterprises (Gao et al., 2020; Kusiak, 2017). In the industrial economy era, the corporate green innovation process primarily relies on a linear chain model of knowledge accumulation, research, and application. With the support of intelligent technology, big data, as a new production factor, participates in the innovation process (Chatterjee et al., 2022). This can not only promote the digitalization of enterprise production processes but also facilitate the digitalization of consumer information. New data from the production department, as well as new green ideas and demands from consumers, will be transmitted to the R&D department in a more convenient way. Under these circumstances, the boundaries between R&D and production within enterprises, as well as between manufacturers and consumers, will become increasingly blurred, and the process of innovation will gradually integrate (Nambisan et al., 2017). The data generated by IM have revolutionized the traditional innovation mode, creating a notable “multiplier effect” in green innovation. These shifts will drive companies to boost their investment in green innovation, helping them to acquire a competitive advantage by advancing BGI.

On the other hand, IM can also enhance enterprises’ big data mining capabilities and stimulate their innovation potential (Mikalef et al., 2019). In the IM mode, powerful computing power and accurate algorithms have become an important support for enterprises to quickly analyse massive data. In the face of scattered and complex data with different structures, big data can strengthen enterprises’ process ability for unstructured and non-standardized data, and fully mine high-value data elements through coding and integration of data (Akter et al., 2019). On the production side, IM brings an organic combination of digital technology and traditional means of production, enabling production data within the enterprise to be fully mined and helping the enterprise to discover the hidden information among data elements, thus providing more original data elements for green innovation activities. On the consumer side, intelligent technology can mine customers’ green data using machine learning and data mining (Hamilton and Sodeman, 2020), realize real-time interaction between enterprises and customers (Cappa et al., 2021), form a customer-centered design network and design thinking, and depict the ‘user portraits’ of consumers from multiple angles. By processing and analysing valuable user information that supports product innovation (Shukla et al., 2019), enterprises can promptly transform it into efficient and accurate green innovation behaviour, thereby identify the optimal plan during the green innovation process. This will stimulate their green innovation vitality, and enhance their investment in green innovation, thereby promote enterprises’ BGI.

Second, IM can optimize the labor structure of enterprises and contribute to enhancing the quality of human capital. IM has a substitution effect on the production personnel who do repetitive and programmed work (Acemoglu and Restrepo, 2019; Acemoglu and Restrepo, 2020; Bessen, 2019), but it has a complementary effect on non-repetitive and non-programmed R&D personnel (Furman and Seamans, 2019). The combined influence of these two factors has significantly promoted the upgrading of the labor skill structure in enterprises. According to the theory of human capital (Becker, 2009), a highly skilled labor force with stronger knowledge reserves and professionalism is not only facilitates innovation (Diebolt and Hippe, 2019), which can promote breakthroughs in core green technologies, but also enhances enterprises’ understanding of sustainable development concepts, such as green and environmental protection, thereby promoting enterprises to pursue long-term green development. Therefore, the upgrading of labor structure will help enterprises increase their investment in green innovation and promote BGI.

BGI places a strong emphasis on the development and creation of new technologies, which require the continual search, acquisition, and accumulation of new knowledge. This process aims to achieve transformations from quantitative to qualitative levels and ultimately yield breakthrough outcomes. It is challenging for enterprises to rely solely on their traditional closed green innovation approaches to achieve these objectives; therefore, they often need to engage in GOPI with multi-agent cooperation. According to the open innovation theory (Chesbrough, 2003), effectively integrating internal and external innovation resources (Teece, 2007) and promoting the exchange of diverse knowledge and ideas across organizational boundaries (Chesbrough and Bogers, 2014) are crucial for enhancing breakthrough innovation within enterprises (Wang et al., 2024b). IM can facilitate GOPI and enrich the diversity of external knowledge sources. With the assistance of intelligent platforms, the inflow of a large number of heterogeneous resources helps expand the boundaries of enterprise knowledge search and continuously optimizes the existing knowledge infrastructure. This process enhances companies’ capabilities to tackle complex green technologies, thereby fostering the emergence of BGI.

First, IM can reduce the cost of cooperation among various green innovation entities. IM can use digital technology to convert vast amounts of information into binary digital resources, facilitating the liquefaction of these resources. Liquefied resources can accelerate the speed and frequency of knowledge and information flowing among various green innovative entities, break the spatial limitations of knowledge spillover in the traditional sense, and promote the deep integration and complementary advantages of green innovative resources between enterprises, as well as between enterprises and universities and relevant research institutes (Fernández-Mesa et al., 2014). Diversified knowledge, technology, and information are interconnected among innovation entities (Perez-Arostegui et al., 2012), further expanding the network boundary of GOPI and strengthening the connections among green innovation entities. IM reduces the cost of cooperation between enterprises and other green innovative entities, and facilitates the development of BGI.

Second, IM will reduce the cost of searching for green innovative partners. According to the search friction theory, due to the heterogeneity of demand among innovation entities (Tödtling et al., 2009), there are a large number of demand and supply entities in the green innovation market, and it is often necessary to pay a non-negligible search cost to achieve cooperation between innovation entities (Salge et al., 2013). However, IM can store, organize and mine the production and R&D data of green innovative entities, and establish a data pool. Further, through intelligent technologies such as matching algorithms and cluster analysis, IM can find the best partner for enterprise green innovation activities. IM will reduce the search cost among the innovative partners and improve the matching efficiency of innovation cooperation (Van Alstyne et al., 2016). This will help enterprises establish an open and efficient GOPI network, and then optimize green innovation resources and enhance BGI’s capabilities.

IM will reduce the internal information asymmetry and alleviate the agency conflict. The “hidden action” caused by moral hazard is an important obstacle to enterprise innovation (Holmstrom, 1989; Manso, 2011). Unlike traditional green innovation, BGI involves the exploration and breakthrough of high-quality innovations, making the innovation process more complex. This may lead to severe moral hazards. On the one hand, the “Type I agency conflict” between shareholders and management is more pronounced (Zardkoohi et al., 2017). Although BGI is crucial for corporate green development, the long-term nature and high level of difficulty in its R&D process may lead management to adopt short-term strategies that ultimately hinder BGI. On the other hand, the agency conflict between management and the R&D department is more salient. BGI not only involves communication and collaboration among several departments within the enterprise but also necessitates deep knowledge of market dynamics and user needs, characterized by stronger information asymmetry. Based on performance evaluations, R&D departments may exhibit caution and reluctance in taking initiatives within BGI. However, IM can optimize internal management processes (Kretschmer and Khashabi, 2020), effectively mitigate agent conflicts (Ortega-Argiles et al., 2005), and create a favorable internal environment for enhancing BGI.

First, IM can enhance the oversight of management behaviours and help mitigate “Type I agency conflict” between shareholders and management. The separation of ownership and control in enterprises results in serious information asymmetry between shareholders and management. Since management has more information during the innovation process, they may prefer short-term actions that benefit performance appraisal, but have limited motivation to pursue BGI, which would enhance corporate long-term competitiveness. However, IM can collect and monitor the behaviour and decision-making information of management in real time with the help of a big data platform, thereby to a certain extent strengthen the scope and intensity of supervision over management. This process will effectively restrain the opportunistic behaviours of enterprise management by violating rules and discipline and excessively pursuing short-term economic benefits (Menon, 2018), thus alleviating the “Type I agency conflict,” which will promote BGI.

Second, IM will change the management process and green innovation chain, which will aiding in diminishing the information asymmetry between the management and the R&D department. In the traditional mode, the enterprise organization structure is hierarchical and linear. But IM can promote the transformation of the enterprise management structure to networking and flattening (Kretschmer and Khashabi, 2020). The reduction of organizational hierarchy decreases the internal information asymmetry, promotes the efficient communication between management and the R&D department, and helps enhance the breakthrough of green technology. Moreover, IM’s dynamic tracking of the green innovation chain will further strengthen the supervision of the R&D department and reduce the strategic green innovation behaviours adopted in response to assessments. This will guide the R&D department to take more substantive innovative activities, thereby promoting BGI.

In summary, IM can promote BGI, which involves higher risk, greater requirements for knowledge integration, and stronger information asymmetry. Therefore, four research hypotheses are proposed as follows.

Hypothesis 1

IM can positively promote BGI.

Hypothesis 2

IM can promote BGI by crowding-in R&D resources.

Hypothesis 3

IM can promote BGI by strengthening GOPI.

Hypothesis 4

IM can promote BGI by alleviating agency conflict.

Figure 1 displays the theoretical framework of IM’s on BGI.

BGI is the explained variable. Based on two streams of studies on green innovation and breakthrough innovation (Yan et al., 2024; Capponi et al., 2022), we use the technology categories and the number of patents for new entrants in green invention patents to construct the BGI (Balsmeier et al., 2017; Gao et al., 2021). The basic logic behind the selection of BGI is that IPC subclass codes are frequently used to identify firms’ capabilities to engage in or explore new technological areas (Gao et al., 2021). According to Balsmeier et al. (2017), the categories and numbers of new green patent entries represent critical breakthroughs in existing technological capabilities and can indicate a firm’s level of BGI. On the one hand, the more newly green patent categories an enterprise enters, the broader its exploration and expansion in new technology areas. This indicates that firms have broken through existing technological paradigms, resulting in higher novelty of technological innovations (Lin et al., 2021). On the other hand, a greater number of green patent applications under the category of new entrants by firms implies that firms are more abundant in green innovations within new technologies and products. This suggests that firms have invested more resources in innovation and have more diverse sources of external knowledge. Moreover, it is worth noting that we select green invention patents as the base data for constructing the BGI. Compared with green utility model patents, the application and examination processes of green invention patents are more complex and stringent (Jia et al., 2019), and the information asymmetry within the innovation process is higher. Therefore, the selection of green invention patents as the base data for the BGI can better reflect the complexity of the innovation processes of enterprises. In summary, the higher the novelty, the more diverse the knowledge sources, and the more complex the innovation process of an enterprise’s green technology, the higher level of BGI.

Specifically, based on the information of the IPC main group (Gao et al., 2021), we constructed BGI by following these steps. To begin with, we take the number of newly entered categories of green invention patents over the past 5 years as the primary indicator of BGI, named as IPC5. IPC5 is distinguished by the revolutionary ability of green knowledge or technical methodologies into emerging domains (Lin et al., 2021). Additionally, based on the calculation of the index of IPC5, we calculate the number of patents applications of green invention entering new fields in the past 5 years as the second indicator of BGI, regarded as Newpatent5. Newpatent5 is featured by the quantity of existing green invention patents to break through the boundaries of technology and knowledge.

Simultaneously, taking into account that the dimension of time may influence the accuracy of BGI, we also calculate the number of newly entered categories and corresponding applications of green invention patents over the past four and 3 years, which we consider additional indices for conducting robustness tests on the BGI. To more effectively evaluate the characteristics of BGI’s knowledge diversity, we utilize text analysis techniques and mine the abstract information of green invention patent to construct the cosine similarity (Arts et al., 2019). We consider this index as an additional metric for measuring the BGI. The lower the similarity between the summary information of enterprises’ green invention patents, the richer the knowledge base involved in green innovation, and thus the wider the boundary of expansion and the higher the level of enterprises’ BGI. Furthermore, we also examined the technical novel characteristics of BGI. The frequency of patent citations is frequently considered as an index to measure the influence of patents (Hall et al., 2001), and commonly used to measure breakthrough innovation (Verhoeven et al., 2016). The frequency with which a green patent is cited correlates with the breadth of its technological spillover, which, to a certain extent, reflects the enterprise’s departure from the original technological paradigm and the novelty of green technology. From this perspective, we employ a proxy indicator by calculating the top 20% of highly cited green invention patents to gauge BGI, which is further regarded as an indicator for a robustness assessment.

We take the IMPP released by the Ministry of Industry and Information Technology (MIIT) as a quasi-natural experiment to investigate the impact of IM on corporate BGI (Wei et al., 2024; Lyu et al., 2023). To advance the creation of a powerful industrial nation, the MIIT selected 305 enterprises to implement IMPP from 2015 to 2018. The specific situation of enterprises implementing the IMPP in terms of time and quantity is as follows: 46 in 2015, 63 in 2016, 97 in 2017, and 99 in 2018. IMPP provides valuable practical experience for the country to explore IM models and drive the intelligent advancement of the manufacturing sector. We regard the policy of annually launching and publishing the list of pilot enterprises as an obvious exogenous event. This method can overcome the endogenous problem in selecting the IM index to some extent, and provide an opportunity to accurately identify the green innovation impacts of IM.

When it comes to determining the list of IMPP enterprises, we consulted the research conducted by Wei et al. (2024) and extracted information on firms that have adopted IMPP by analysing relevant documents ² . First, we referred to the list of IMPP released by the MIIT between 2015 and 2018. Second, we cross-referenced using tools like Qichacha and Cninfo to verify the pilot companies, thereby compiling a comprehensive list of IMPP companies. Lastly, we ensured the accuracy of the list by conducting meticulous manual searches and individual verifications. As a result, we identified a final roster of 105 companies that have successfully implemented IMPP. For enterprises that were repeatedly selected for IMPP, this paper defines the dummy variable for IM based on the year they were first selected. In addition, we also employ text analysis methods to identify IM-related keywords in the annual reports of companies and regard them as proxy variables for IM to conduct robustness checks.

Apart from the variables previously mentioned, we also consider the following enterprise characteristics that might influence BGI: company size (Size), company age (Age), net profit rate of total assets (Roa), book-to-market value ratio (Btm), cash flow status (Cfp), total compensation of the top three managers (Tec3), dual roles of CEO and chairman (Dual), and proportion of fixed assets (Pfa). Table 1 represents descriptions of these variables.

The basic results show that after the implementation of IMPP, the BGI of enterprises has been impacted. However, an important premise for the conclusion is the parallel trend test, which requires that in the absence of exogenous impacts, the changes in BGI between the treatment group and the control group are similar. Based on this, we employ a dynamic difference-in-differences model ³ to explore the dynamic influence of IM on BGI. B G I i t = α + β S p r e c u t D i × I t − T D < − 2 + ∑ s = − 2 − 1 β s p r e D i × I t − T D = s + ∑ s = 0 2 β s p o s t D i × I t − T D = s + β S p o s t c u t D i × I t − T D > 2 + γ X i t + θ i + δ t + ε i t   (2)

Based on Equation 2, we re-estimate the model and plot the dynamic graphs of the parallel trend. Supplementary Figures SA3, A4 represent the results of parallel trend test. These findings indicate that prior to the implementation of IMPP, there is no systematic difference in BGI among enterprises. After the implementation of IMPP, the treatment group exhibits a considerable rise in BGI compared to the control group, supporting the parallel trend test.

Although the aforementioned research found that IM positively impacts BGI, the results may be affected by other factors. To ensure that the impact of IM on BGI is not affected by other random variables, we use a placebo test to examine the random nature of the IM effect. Based on the distribution of the IM variable in the benchmark regression, we performed 500 random samples and created a “pseudo-policy dummy variable,” and then regressed it again using Equation 1 to test the coefficient value and P - value distribution. From the test results depicted in Supplementary Figures SA5, A6, we observe that the influence of IM on BGI is not due to other random factors. These findings bolster the credibility of our estimation results.

Given the potential sample selection issues during the implementation of IMPP, we utilize the propensity score matching method (PSM-DID) for verification. We regard the control variables as covariates. Furthermore, we employ two matching methods, namely, kernel matching and nearest neighbor matching, to ensure robustness in our analysis. To more evaluate the net effect of IM on corporate BGI, we re-estimate the regression using Equation 1, excluding unmatched samples. Table 4 indicates that IM can enhance BGI.

According to Goodman-Bacon (2021), the two-way fixed-effects estimator in the estimation of the TDID model is equivalent to the weighted average of all possible two-period DID estimators in the sample. This approach may make the heterogeneity of treatment effects unrobust, for instance, due to issues such as negative weights. As each firm was selected for IMPP at a distinct point in time, the following three controls will be included in our study: ①New entrants to the IMPP (Treatment) and those who have never been selected for the IMPP (Never Treated); ②New entrants to the IMPP (Earlier Group Treatment) and those who have not yet joined the IMPP (Later Group Comparison); ③New entrants to the IMPP (Later Group Treatment) and those who are already participating in the IMPP (Earlier Group Comparison). Among these three controls, the first two serve as ‘good controls’ in that they assess the impact of whether or not to enter the IMPP on firms’ BGI. However, the treatment effect in the third control is not necessarily homogeneous. If the weights assigned to the control mean estimator of the third are greater, this can influence the results of the two-way fixed effects estimator, thereby biasing the results (Goodman-Bacon, 2021; Sun and Abraham, 2021).

Based on this observation, we employed the Goodman-Bacon decomposition to examine whether the third control was over-represented. Supplementary Figures SA7, A8 present the TDID Goodman-Bacon decomposition results for IPC5 and Newpatent5, respectively. These results demonstrate that the percentage of the third control is smaller in this study, implying that the effect of IM on BGI is not attributable to the third control. Additionally, we also report decomposition results specifically for IPC5 and Newpatent5, and these results are presented in Table 5. Based on the Goodman-Bacon decomposition results of IPC5, it is evident that the primary contributor to the overall DID estimation results in this study is the estimation results from the sample that was never exposed to the IMPP, serving as the control group, with a weight as high as 98.9% (Treatment vs. Never Treated). In contrast, the sample with earlier treated individuals as the control group carries a weight of only 0.3% (Later Group Treatment vs. Earlier Group Comparison), which has a minimal impact on the overall estimation results and does not lead to an overestimation of the effect of IM on BGI. Newpatent5 follows a similar judgment process. The above analyses demonstrate that the results of the TDID estimation we employed are reliable.

The relationship between IM and BGI may be affected by endogeneity issues. To address this, we employed the instrumental variable approach to eliminate potential endogeneity. Specifically, we selected the interaction term between the number of landline telephones in each city in 1984 and the time variable (IV1), as well as the interaction term between the number of post offices in each city in 1984 and the time variable (IV2), as instrumental variables for IM (Nunn and Qian, 2014). The selection of these variables is justified on two grounds. First, regional landline telephones and post offices can influence the spread and development of information by shaping the technological and usage habits of Internet use, thereby satisfying the relevance requirement of the instrumental variable. Second, the number of landline telephones and post offices in 1984 does not directly contribute to firms’ current BGI, thereby satisfying the exogeneity assumption of the instrumental variable.

Table 6 presents the estimation results of the Two-Stage Least Squares (2SLS) method, with Columns (1)-(3) indicating the results for IV1 and Columns (4)-(6) indicating the results for IV2. The results of the LM test show that IV1 and IV2 exhibit a significant correlation with IM; the F-values for both IV1 and IV2 are significantly greater than 10, thereby rejecting the original hypothesis of weak instruments. Our central conclusions remain robust.

To investigate the correlation between IM and BGI, it is critical to exclude the influence of other policies. Given that other pilot cities and regional policies may interfere with the identification of IMPP effects, we construct dummy variables to exclude the influence of other related policies within the same period. These policies encompass the Made in China 2025 IM Pilot Project (Li and Branstetter, 2024), the National Big Data Comprehensive Pilot Zone (Wang X. et al., 2024), the Broadband China Policy (Wang and Wang, 2024), and the Smart City Policy (Qi et al., 2024). We label them as Policy1, Policy2, Policy3, and Policy4, respectively. Subsequently, we systematically incorporate the aforementioned four policies as control variables in Equation 1, and the Columns (1)-(8) of Table 7 report the estimation results. The results indicate that the estimated coefficients of IM are significantly positive, following the incorporation of the aforementioned policies as control variables. Moreover, we consider the joint influence of Policy1 - Policy4, and Columns (9) and (10) present the regression results. The results show that the regression coefficients of IM on IPC5 and Newpatent5 remain positive and statistically significant. These results align with our findings.

(1) Replacing the explained variables. First, referring to Lin et al. (2023), we gauge the novelty of innovation outcomes by calculating the cosine similarity of green invention patent abstracts ⁴ , name it BGI1, and regard it as the first indicator of BGI, Second, referring to Aghion et al. (2019) and Chang et al. (2019), we exploit the top 20% of green invention patent citations as the second indicator for measuring BGI, and name it BGI2. Last, following Makri and Scandura (2010), we apply the dispersion of green invention patent technology ⁵ as the third indicator to measure BGI, and name it BGI3. We also calculate the number of newly entered categories of green invention patents and their applications by the company over the past four and 3 years. We employ them as alternative indicators for BGI in the robustness test and name them IPC4, Newpatent4, IPC3, and Newpatent3, respectively. Table 8 represents the estimation results.

(2) Replacing the explanatory variable. Drawing upon Verhoef et al. (2021), we exploit intelligent transformation (IM_R) based on text analysis methods to replace IM ⁶ . Among them, IM_R1 denotes the total number of words related to IM_R, and IM_R2 represents the ratio of the total number of words related to IM_R to the total number of words in the annual report. The results of the alternative explanatory variables are displayed in Table 9. These findings demonstrate that whether substituting the explained variables or explanatory variables, IM can significantly enhance BGI.

To ensure the reliability of the regression conclusions, referring to Jin and Chen (2024), we also selected the fixed effects model for the robustness test. Table 10 displays the regression results after replacing the econometric model. The coefficients of the IM remain significantly positive, indicating that the conclusions of our study are robust.

IM will crowd in the R&D resources, then fully stimulate the BGI vitality of enterprises. Based on this, we employ the ratio of a company’s R&D expenditure to its total assets to quantify R&D resources. A higher proportion of investment in R&D will lead to a larger pool of green innovative resources, ultimately facilitating BGI. Table 11 presents the results of the mechanism test for R&D resources. Columns (1)-(3) display the regression results for the mediation effect model. Column (1) demonstrates that the coefficient of IM is significantly positive, indicating that IM contributes to the enhancement of R&D resource. In Columns (2) and (3), the coefficients of IM exhibit a notably positive correlation, and the coefficients of R&D also show a significantly positive relationship. These findings imply that IM has the ability to enhance BGI by facilitating a growth in R&D resources.

To ensure the reliability of the regression results for R&D resources, we constructed an interaction term of IM and R&D (IM × R&D) and incorporated IM × R&D, IM, and R&D into Equation 3 for the moderating effect test. The main logic of this approach is that if R&D resources serve as the channel through which IM influences BGI, then the effect of IM on BGI should increase with R&D resources, meaning that the coefficient of IM × R&D should be significantly positive. Columns (4)-(5) of Table 11 present the regression results. As shown in Columns (4)-(5), the coefficients of IM × R&D are significantly positive at the 1% level, indicating that R&D resources indeed function as an important channel through which IM enhances BGI. These findings suggest that R&D resources serve as a crucial channel through which IM fosters BGI. In summary, hypothesis 2 is verified.

IM can foster the establishment of GOPI network by connecting enterprises with other enterprises, universities, and research institutes. Generally speaking, the greater the extent to which enterprises engage in green collaboration, the more abundant the resources of knowledge and technology required for green innovation become. Within the framework of the GOPI network, the communication cost among enterprises will be effectively reduced, thereby release the full potential of BGI (Laursen and Salter, 2006). Hence, we aim to study whether IM impacts BGI via GOPI. Drawn on Hong and Su (2013), we calculate the number of innovation collaborative partners involved in green invention patents to measure GOPI ⁷ . The greater the GOPI, the more universities and research institutes engage in green collaborative innovation with the enterprise, indicating a more robust and comprehensive GOPI network. Table 12 presents the results of the mechanism test for GOPI. Columns (1)-(3) display the regression results for the mediation effect model. Specifically, Column (1) reveals that the coefficient of IM is saliently positive, suggesting that IM facilitates the expansion of the green innovation partner network and significantly reinforces GOPI. Furthermore, in Columns (2) and (3), the coefficients of GOPI are significantly positive. These findings imply that IM has the capability to promote BGI by enhancing GOPI.

To ensure the reliability of the regression results for GOPI, we constructed an interaction term of IM and GOPI (IM × GOPI) and added IM × GOPI, IM, and GOPI to Equation 3 for analysis. The core rationale of this approach is that if GOPI mediates the relationship between IM and BGI, then the effect of IM on BGI should increase with higher levels of GOPI, implying that the coefficient of IM × GOPI should be significantly positive. The results of Columns (4)-(5) reveal that the coefficients of IM × GOPI are significantly positive at the 1% level, indicating that GOPI serves as a critical channel through which IM enhances BGI. These findings suggest that GOPI play a pivotal role in facilitating the promotion of BGI by IM. Accordingly, hypothesis 3 is confirmed.

IM has the ability to enhance resource allocation, internal management efficiency, and alleviate information asymmetry. These advantages can assist enterprises minimize agency conflict, resulting in enhance BGI. Therefore. we investigate the possibility of agency conflict acting as a mediator between IM and BGI. Following Ang et al. (2000), we select the management expense ratio to gauge agency conflict. It is determined by dividing management expenses by operating income. A higher ratio indicates larger agency conflict. Table 13 presents the results of the mechanism test for agency conflict. Columns (1)-(3) display the regression results for the mediation effect model. Specifically, Column (1) indicates that the coefficient of IM is notably negative, suggesting that IM contributes to reducing the management cost of enterprises. In Columns (2) and (3), the coefficients of IM are notably positive, whereas the coefficients of agency conflict are significantly negative. These findings indicate that IM can enhance BGI by alleviating agency conflict.

To ensure the reliability of the regression results for agency conflict, we constructed an interaction term of IM and agency conflict (IM × AC) and added IM × AC, IM, and agency conflict to Equation 3 for testing. The primary rationale of this approach is that if agency conflict mediates the relationship between IM and BGI, then the impact of IM on BGI should increase as agency conflict decreases, implying that the coefficient of IM×AC is significantly negative. Columns (4)-(5) of Table 13 present the regression results. In Columns (4) and (5), the coefficients of IM × AC are significantly negative at the 1% level. These findings suggest that agency conflict serves as a crucial channel through which IM affects BGI. In other words, IM can enhance BGI by alleviating agency conflict. Therefore, hypothesis 4 is confirmed.

Manufacturing, especially high-pollution and high-energy consumption industries, exhibit a strong path dependence concerning technological innovation. The existing inventory of polluting technologies within enterprises may impede the development of clean technologies (Aghion et al., 2016). In particular, China, as the representative of emerging manufacturing countries, relies heavily on high-pollution and high-energy consumption industries, which act as both the cornerstone of national economic growth and a major source of pollution discharge. Therefore, it is of great significance for emerging manufacturing countries to expedite the achievement of green development by finding ways to overcome the path dependence of these industries. In this context, we examine the influence of IM on BGI within high-pollution and high-energy consumption industries.

In this part, based on the industry’s pollution intensity and energy consumption levels, we divide the total sample into four parts: high-pollution (HP) industries and non-high-pollution industries (NHP) ⁸ , high-energy consumption (HEC) industries and non-high-energy consumption industries (NHEC) ⁹ . Table 14 displays that the coefficients of IM are significant positive in Columns (1), (2), (5), and (6) for the HP and HEC sub-samples, whereas the coefficients of IM in the NHP and NHEC sub-samples are not significant. Therefore, we infer that IM can disrupt the path dependence of technological innovation in HP and HEC industries, thereby promote BGI of these industries.

There are several plausible explanations for these results. First, IM highlights the integration of artificial intelligence technology into the production process, which helps break the inertia of green innovation. What’s more, the characteristics of IM also can facilitate precise control and management throughout the green innovation process, thereby providing a new production paradigm for HP and HEC industries. Second, the production networks of enterprise based on artificial intelligence technology are beneficial for breaking the ‘locking effect’ of traditional production networks on traditional technology paths. Third, the transformation and upgrading of machinery equipment facilitated by IM could alleviate the “sunk cost” barrier to adopting green technology to some extent. Therefore, IM can disrupt the path dependence of enterprise technology selection by altering the production paradigm of the HP and HEC industries, and drive these industries to promote BGI.

The correlation between environmental regulation and green innovation has been a topic of academia. For a considerable time, two opposing views have existed regarding the effect of environmental regulation on green innovation. Based on neoclassical ‘cost-following’ theory, some research argues that environmental regulation raises pollution control costs and decreases R&D inputs (Conrad and Wastl, 1995). From the perspective of Porter’s theory, some scholars assert that environmental regulation significantly raises the opportunity cost of green innovation, compelling enterprises to proactively pursue green innovation initiatives (Du et al., 2021; Porter and Linde, 1995). What is the impact of environmental regulation on the relationship between IM and BGI? Currently, there is little academic discussion on this issue.

Based on this, we explore the influence of IM on BGI under different levels of environmental regulations. First, from the viewpoint of environmental governance, we calculate the proportion of pollution control investment for each province to measure the intensity of environmental regulation at the regional level (Liu et al., 2019). Based on the median, this paper divides the samples into high environmental regulation (HER) and low environmental regulation (LER). Second, we regard the evaluation standard of the environmental protection model city as another measure of environmental regulation. The underlying rationale is that the areas selected as model cities for environmental protection are subject to more rigorous environmental evaluations. Within this constraint, these cities intensify their commitment to environmental regulation. From this perspective, using information from the official website of the Ministry of Ecology and Environment of the People’s Republic of China ¹⁰ , we categorise the samples into two groups based on strong environmental protection (SEP) and weak environmental protection (WEP). A value of 1 is assigned when the company is situated in a model city for environmental protection and 0 otherwise. Table 15 displays the results. For the sub-sample of HER and SEP in Columns (1), (2), (5), and (6), the coefficients of IM exhibit a significant positive relationship. However, for the sub-sample of LER and WEP in Columns (3), (4), (7), and (8), the coefficients of IM do not exhibit a significant relationship. This empirical evidence suggests that as regional environmental regulations become stricter, the promoting effect of IM on BGI becomes more pronounced.

The above conclusions may be attributed to the following two factors. First, IM can decrease enterprise costs and form a joint force with environmental regulations to enhance BGI. On the one hand, environmental regulations could potentially increase the opportunity costs associated with pollution, thereby creating an incentive for green innovation within the enterprise. However, environmental regulations may also raise the costs of corporate governance and reduce the resources available for green innovation within enterprises. In the above two situations, the dominant cost factor under the aforementioned situations is also the subject of debate between the cost-following theory and Porter theory. Nevertheless, IM enhances efficiency of green innovation and contributes to lowering the cost of environmental governance to some extent. In areas with strict environmental regulations, as enterprises face higher opportunity costs, the impact of IM on boosting BGI becomes more pronounced. Second, environmental regulations can drive the formation of GOPI, which is beneficial for enhancing the mechanism of IM in promoting BGI. As a crucial method for environmental improvement, BGI requires advanced, extensive, and fundamental technical support. Universities and research institutes often have innovative and advanced green technologies, in which case, the growing stringency of regional environmental regulations compels enterprises to bolster their collaboration with universities and research institutes, thereby enhancing the GOPI network formation. In some respects, the above analysis also indicates that the stricter the regional environmental regulations, the more perfect the GOPI network, and the more effectively the IM promotes BGI.