Artificial Intelligence and the New Quality Productive Forces of Enterprises: Digital Intelligence Empowerment Paths and Spatial Spillover Effects

Abstract

The 20th CPC Central Committee stressed that the key to high-quality economic development is to cultivate new quality productive forces, and AI plays a key role in cultivating new quality productive forces. This paper takes A-share listed enterprises in China from 2013 to 2022 as a sample, constructs comprehensive level indicators of AI from the strategic side, application side, and innovation side of enterprises’ AI, and empirically examines the impact, mechanism, and spatial spillover effect of AI development on enterprises’ new quality productive forces from the perspective of digital intelligence empowerment and the spatial perspective. The results of this study show that AI can significantly promote the development of new productivity, and the development of AI within enterprises can promote the improvement of new productivity levels of neighboring enterprises or regions. At the same time, the role of AI in promoting the development of new quality productive forces is more obvious when the enterprise is a private enterprise, the managers have a digital background, and the enterprise is located in an industry with fierce market competition or a strategic industry. The purpose of this paper is to reveal the mechanism and spatial spillover effect of AI in promoting the new quality productive forces of enterprises and to provide a new theoretical basis and research perspective for enterprises to cultivate new quality productive forces.

1. Introduction

The Third Plenary Session of the 20th CPC Central Committee emphasized that “achieving high-quality development is the primary task of building a modern socialist country in an all-round way”. To promote high-quality development, it is necessary to base efforts on the new development stage and implement the new development concept, and the development of new quality productive forces is a major deployment based on the new development stage, as well as an inherent requirement and an important focus of high-quality development. Different from the traditional productive forces, the new quality of productive forces is driven by science and technology innovation, which plays a leading role in the productive forces. The “new” aspect lies in the realization of key disruptive technological breakthroughs, through innovation-driven approaches, optimizing the allocation of resources, improving production organization, etc. These advancements enhance productivity levels and are in line with the new development concept of advanced productive forces in terms of quality [1]. Meanwhile, as a cutting-edge technology with an innovation-driven kernel, AI is highly compatible with the innovation concept of new quality productive forces, and AI, by virtue of its powerful computation and prediction capabilities, high-precision automation characteristics, and deep-learning-driven innovation capabilities, is rapidly changing the operation mode and organizational structure of each economic entity [2], promoting a new wave of industrial revolution and scientific and technological progress, which will inevitably be able to be used in the corporate world. It is bound to play a key role in the process of accelerating the cultivation of new quality productive forces and help enterprises realize high-quality development. Therefore, researching the role of artificial intelligence in promoting the cultivation of new quality productive forces of enterprises, revealing the mechanism paths involved, and exploring the influence effects of the two in the spatial plane have important theoretical and practical value for in-depth understanding and acceleration of the cultivation of the new quality productive forces of enterprises and the formulation of regional digitalization policies.

Artificial intelligence (AI) refers to the technology that enriches computer programs or algorithms through algorithm optimization, machine learning, deep learning, and other technological means, thus simulating or enhancing human intelligence. At present, there are a lot of studies focusing on AI to help cultivate new quality productive forces, and a large number of studies have shown that the application of AI can empower the cultivation of new quality productive forces, but most of them are focused on the theoretical level. For example, Qi et al. [3] argued that AI, by virtue of its numerical and intellectual attributes, creates a new value chain by promoting the formation of the enterprise’s human–machine collaborative production mode of data elements with robotic applications, and by connecting various links in the value chain, creating new value chain links, creating a new ecology of the value chain, and promoting the cross-border integration of organizational platforms to empower the development of new quality productive forces. Guo [4], from the perspective of new quality production factors, argues that AI can transform data into new quality labor objects, improve the ability to apply new quality labor skills, and thus promote the systematic transformation of the new quality industrial system. Jiang [5] also argues that AI, as a new type of labor tool, can improve the intelligence, greenness, and innovation in the production process, and in improving total factor productivity, it can breed new patterns of industry and thus broaden the development path of new quality productive forces. Sun [6], by deconstructing the traditional production function, argues that AI technology empowers the data elements to change the traditional production function, promotes technological innovation, and improves the quality of labor, promoting the development of new quality productive forces. Lu [7], at the industrial level, also believes that AI promotes the deep integration of traditional industries with digital and real systems, empowers the digitization of strategic emerging industries to achieve multiplier growth, and drives the scale of the application of reconstruction of the intelligent industrial chain and innovation chain to promote industrial transformation and upgrading, achieving the intelligent advancement of the industrial structure and then promoting the development of new quality productive forces. The studies above discuss the artificial industry from various levels.

Based on the various levels discussed above, artificial intelligence empowers the cultivation of new quality productive forces. This paper provides useful research ideas, and a review of the relevant literature reveals that there is no literature based on the patent data from listed companies or empirical research on whether artificial intelligence can promote the integration of digital and real systems, facilitate the flow of knowledge, and empower the cultivation of new quality enterprise productivity through digital intelligence. At the same time, the characteristics of artificial intelligence determine the impact not only on the enterprise itself, but also across broader dimensions. At the same time, the characteristics of AI determine that it can not only have an impact on enterprises themselves, but also have an impact on other enterprises across spatial limitations. However, there is no research that focuses on the spatial spillover effect of AI on the new quality productive forces of firms, and the shortcomings mentioned above provide space for this paper to make a breakthrough. Therefore, based on the data of China’s A-share listed companies from 2013 to 2022, this paper analyzes the mechanism and mechanism path of AI-enabled cultivation of enterprises’ new quality productive forces by establishing a theoretical framework; constructs the comprehensive level indicators of enterprises’ AI including the strategic level, the application level, and the innovation level in the empirical test, and then applies the two-way fixed effect model to empirically verify the impact effect and its mechanism path of AI-enabled cultivation of enterprises’ new quality productive forces. The impact effect and its mechanism path are empirically tested, focusing on the mechanism from the perspective of digital intelligence empowerment; at the same time, based on the theory of spatial economics, the spatial weight matrix at the enterprise level is established based on the latitude and longitude of the enterprise by utilizing Python software (version: Python 3.11), and the spatial spillover effect of the AI-enabled new quality productive forces is innovatively studied based on the spatial econometric model.

In summary, this paper focuses on the impact of AI on firms’ new qualitative productivity and its spatial spillover effects, and the possible marginal contributions of this paper are as follows: first, by centering on the micro enterprise level, we empirically test the facilitating role of AI empowerment in fostering new quality productive forces in enterprises, providing a new empirical basis and research ideas for the subsequent research in this field. Second, drawing upon the strategic positioning, practical applications, and innovative development of AI in enterprises, we innovatively establish comprehensive indicators for AI, which address the shortcomings of existing methods. Third, from the perspectives of collaborative innovation theory and sustainable development, we reveal the mechanism path through which AI enables the cultivation of enterprises’ new quality productive forces, with a view to providing a theoretical basis for subsequent research. Fourthly, from the spatial dimension, we introduce spatial measurement methods into the study of AI on the development of new productivity of enterprises, establish a spatial weight matrix of enterprises based on the latitude and longitude data, and construct a spatial Dubin model to empirically test the spatial spillover effect of AI-enabled cultivation of new productivity, so as to provide research methods for subsequent research and a theoretical basis for the policy formulation.

2. Theoretical Foundation and Research Hypotheses

2.1. A Theoretical Analysis of Artificial Intelligence for New Quality Productive Forces in Firms

New quality productive forces refer to a novel form of production capacity characterized by intelligent, networked and sustainable features, which is formed through technological innovation and resource integration, with emerging digital technology acting as the core driving force and data, knowledge, and technology serving as the main production factors. The essence of new quality productive forces lies in breaking the dependencies on capital, labor, and land, reconfiguring production modes and value creation, and achieving a holistic improvement in resource allocation efficiency and production effectiveness. New quality productive forces can not only promote the optimization and upgrading of industrial structure, but also facilitate the realization of a green economy and high-quality development, and will soon become an important engine of economic growth and social progress in the new era.

New quality productive forces are distinguished by their focus on innovation, technological advancement, higher quality, efficiency, and sustainability compared to traditional productivity. Here, “new” refers to the fact that the essence of new quality productive forces is innovation, and the attribute of innovation is the starting point of all its connotations; while “quality” highlights that scientific and technological innovation serves as the engine of development, prioritizing the quality of the production process over mere quantity [8]. Artificial intelligence plays a significant role in empowering the cultivation of new quality productive forces of enterprises by influencing multiple factors of enterprises.

(1) First, we consider changing the traditional production function, realizing the enterprise productivity leap, and empowering the cultivation of new quality productive forces by improving the total factor productivity of enterprises. As a collection of technological advancements and disruptive innovations, AI is a cutting-edge technology that simulates and extends the technology and application system of human intelligence [8]. Looking back at history, productivity leaps have been driven by scientific and technological innovations, and artificial intelligence, with its technical characteristics and the innovation nature of the new quality productive forces being highly compatible, has become a key promoter in cultivating new quality productive forces in enterprises [9].

Firstly, AI alters the traditional production function by integrating data elements. Traditional production functions typically involve capital, labor, and other factor inputs, ignoring the potential of data as a new quality production factor. The data-driven core of AI introduces data as a new production factor into the production function and drives its transformation. Unlike traditional production factors, data have the unique characteristics of non-consumption, high mobility, and strong externality, which can expand the value-added mode and application scope of factors of production through the multiplier effect. Data analytics technology can help enterprises accurately identify bottlenecks in the production chain, optimize production processes, and enhance production technology, thereby improving production efficiency and quality. Moreover, AI allows enterprises to analyze market demand in real time and dynamically adjust production strategies, significantly improving the input–output ratio of enterprises and transforming the traditional production function from linear efficiency growth to a non-linear leap [9,10]. Secondly, AI itself is integrated into the production function as a new production factor. Artificial intelligence is not only a technological tool, but also an independent production factor, which is able to achieve the overall improvement of production efficiency by combining with traditional factors. Generative artificial intelligence, as a powerful production support tool, empowers users with its deep learning capabilities for continuous innovation and provides enterprises with a new production support method. Specifically, generative AI can quickly generate multiple design solutions in the product design stage, providing innovative inspiration for R&D personnel; in the production process, generative AI generates multiple design solutions in the product design stage, simulates complex process conditions, and provides a reliable basis for optimizing the production process. The application of this technology not only improves labor output but significantly enhances production efficiency [11]. In addition, generative AI plays an important role in enterprise management. By providing intelligent prediction and optimization functions, generative AI aids enterprise management by providing intelligent prediction and optimization functions, improving resource allocation, and enhancing capital returns [12,13].

Finally, artificial intelligence promotes institutional innovation and indirectly promotes productivity improvement. The improvement of enterprise productivity is intrinsically linked to the support of advanced systems, and the powerful simulation and prediction capabilities of artificial intelligence enable institutional innovation. Through a digital simulation system, enterprises can evaluate various management modes and production programs, identifying institutional frameworks best suited to their development needs. By simulating the impact of different performance management mechanisms on employee productivity, enterprises can design optimal incentive policies. Additionally, in supply chain management, AI empowers enterprises to establish agile supply chain systems by predicting market dynamics and real time demand analysis, thus enhancing overall productivity [14].

(2) Second, we consider promoting technological progress and exploring the availability of technology and innovation-driven cultivation of new quality productive forces. Technological innovation is the core driving force of enterprise development. Research based on the theoretical model of technological progress shows that, all other things being equal, introducing technologies like AI can not only trigger technological change but also significantly increase output growth rates by improving factor efficiency and facilitating innovation diffusion [15]. AI plays an unparalleled role in advancing technological progress by reshaping innovation models. Firstly, traditional innovation is usually confined to the R&D department within the enterprise and lacks the idea of open innovation and co-creation of intelligent platforms. The emergence of AI has enabled the creation of intelligent collaboration platforms, facilitated the flow of data and experience sharing among innovation stakeholders, removed information barriers, optimized the efficiency of information exchange, and thus enhanced the synergy and interaction of all parties, laying a solid foundation for technological progress. Secondly, AI offers a unique form of availability, referring to the potential actions that AI technology can enable users to achieve their desired outcomes [16]. In the context of product development and optimization, for instance, AI possesses significant data storage and advanced model-building capabilities, facilitating precise simulations of various experimental conditions throughout the product development process, as well as in post-launch market scenarios. Through technologies such as virtual simulation, automated research and development, and others, AI enhances both the efficiency and quality of enterprise product development. In the optimization phase, AI-driven tools like Generative Adversarial Networks (GANs) can automatically generate new product solutions. These networks continuously create new samples by drawing on data from training libraries—leveraging existing design data and methods—while integrating multi-domain knowledge to generate innovative product designs and address the shortcomings of current designs. For instance, GANs have found extensive application in industrial design, intelligent manufacturing, and other sectors, providing both innovation-driven inspiration and decision support, enabling companies to make breakthroughs in product design and optimization. This innovation-driven approach enables enterprises to remain competitive in rapidly evolving markets, ultimately achieving new levels of quality productivity [17].

(3) Third, we consider reinventing the business model, optimizing the management structure, and fostering collaboration to cultivate new quality productive forces. Under the impact of the artificial intelligence era, traditional business models have shown numerous limitations, particularly in adapting to rapidly changing market demands, improving efficiency, and navigating complex competitive environments. Therefore, enterprises are gradually integrating AI technology into all aspects of their business models to achieve comprehensive transformation, from core functions to extended capabilities. AI not only reshapes the core elements of business models but also provides technical support and innovation to optimize management structures [18,19].

First, AI drives the transformation of business models to demand orientation. Traditional business models have typically been supply-driven, with product and service designs often based on the capabilities and limitations of production processes, neglecting the increasingly diverse needs of consumers. The rise of AI, on the other hand, enables companies to transform from supply orientation to demand orientation through frameworks such as Industry 4.0. Industry 4.0 helps companies delve deeper into consumer behavior and latent needs through big data analytics and machine learning technologies. With real-time data collection and analysis tools, companies can dynamically predict market trends and provide highly personalized and customized services for different groups of consumers, thus enhancing the customer experience. This covers not only the customized design of products, but also the precise matching of service delivery and follow-up maintenance. In the manufacturing industry, leading companies have already realized the application of flexible production lines through AI technology, which adjusts the production process based on real-time changes in customer orders, improving the efficiency and responsiveness of resource allocation. Furthermore, AI enhances the adaptability of the business model, enabling enterprises to swiftly respond to environmental changes and seize market opportunities, so as to take advantage of the competition [20,21].

Second, AI plays a key role in management architecture optimization. Traditional enterprise management structures often suffer from inefficiencies such as information silos, prolonged decision-making processes, and an inability to respond quickly to complex, high-frequency market changes. The introduction of artificial intelligence has fundamentally addressed these challenges. By leveraging deep learning and big data analytics, AI is able to handle highly complex and multidimensional datasets, generate accurate predictions and insights, and provide scientific, real-time decision support for enterprises. In supply chain management, AI systems monitor operations in real time, predict potential disruptions, and proactively adjust supply plans through intelligent decision-making systems, minimizing losses due to unforeseen events. Additionally, the widespread application of AI-driven Robotic Process Automation (RPA) technology within enterprises enhances the efficiency and streamlines management processes. RPA technology is particularly suitable for manufacturing, logistics, finance, and other areas requiring high process standardization. By automating routine tasks, RPA reduces errors and uncertainties associated with manual operations, while freeing up human resources for higher-value strategic tasks, dramatically improving process efficiency and accuracy [22,23].

Finally, the synergistic effects of AI facilitate the deeper integration of business models and management structures. AI not only independently affects the business model and management structure, but also creates greater value through the synergy between the two. By combining business model innovation and management structure optimization through an intelligent collaboration platform, enterprises can achieve digital management and intelligent operation throughout the whole process. In the retail sector, AI-powered decision-making systems have enabled companies to analyze consumer shopping behaviors in real time, seamlessly linking marketing strategies with inventory management. This integration enhances customer satisfaction and optimizes operating costs. Under this framework of synergistic development, enterprises have gradually formed a dynamic, efficient, and intelligent management system, laying a solid foundation for the cultivation of new quality productive forces in a globalized and digitally competitive environment.

Hypothesis 1:

Artificial intelligence has a significant contribution to empowering the cultivation of new quality productive forces in enterprises.

2.2. Digital Intelligence Empowerment Influence Channels

2.2.1. Digital–Real Fusion Effect Influence Channel

Artificial intelligence drives the creation and development of new quality productive forces by facilitating the integration of the digital and real world, promoting the deep convergence of the real economy with digital technologies. Digital–real integration is a process of intelligent transformation of the real economy based on digital technology, with the aim of breaking down the boundaries between virtual and real, digital and physical, and making data the core production factor. The process empowers all aspects of production, management and service through advanced technology. As enterprises undergo digital and intellectual transformation, the application scenarios of AI are continuously expanding, with intelligent technologies becoming increasingly embedded in every facet of operations. Within the framework of digital–real integration, AI has evolved from being merely a tool for data processing and analysis to a driving force in the digital and intelligent transformation of the real economy [6]. A prominent example of such a technology is the Internet of Things (IoT). The principle behind the IoT is to embed sensors and smart devices into physical assets, equipment, products, supply chains, and other elements of the real economy. This interconnectedness enables the intelligent integration of information, logistics, and data flows, facilitating automated control and optimization through data recovery and analysis [24,25]. Consequently, enterprises are empowered to analyze multidimensional data, including logistics, inventory, and market demand, thereby enabling the development of more effective purchasing, production, and distribution strategies. This reduces inventory backlogs, enhances logistics efficiency, and enables the high-efficiency and flexible operation of assets. Digital–real integration is the process of transitioning from traditional physical productivity to a new form of productivity that is data-driven, algorithmic, and intelligent-system-based. This intelligent decision-making capability, which is driven by data, enhances enterprises’ ability to navigate complex market environments and fosters the development of new quality productive forces. Based on this, we propose hypothesis 2:

Hypothesis 2:

Artificial intelligence can promote the digital–real integration of an enterprise, which in turn empowers the cultivation of enterprise new quality productive forces.

2.2.2. Knowledge Flow Effect Affects the Channel

In today’s era of the Internet of Everything, no enterprise can achieve technological progress and continuous innovation in isolation. By collaborating with external organizations, enterprises can acquire advanced technology and management expertise to bridge gaps in their knowledge and skills, which are often disseminated through inter-firm knowledge exchanges [26]. Artificial intelligence plays a critical role in constructing and optimizing the architecture of knowledge flows among enterprises. On the one hand, the AI-driven open innovation model provides a new path for knowledge flow between enterprises and external partners. Through data sharing, joint research and development, and open platforms, enterprises can engage in joint innovation projects with external technology teams, research institutions, and other partners, breaking the knowledge boundaries of enterprises [27]; on the other hand, data analysis and intelligent reasoning empower management to systematically classify and organize the large amount of knowledge accumulated within an enterprise, consolidating it into a centralized database. Intelligent analytical tools can monitor and analyze real-time data on production, market dynamics, and customer feedback, yielding accurate insights and suggestions to support decision-making for more scientific resource allocation and strategic planning [28]. Through effective knowledge sharing and flow, enterprises can quickly assimilate new technologies, theories and management concepts, enhancing resource allocation efficiency, promoting the transformation and application of knowledge, and ultimately forming new quality productive forces. Based on this, we propose hypothesis 3:

Hypothesis 3:

Artificial intelligence can improve the enterprise knowledge flow effect, which in turn empowers the cultivation of enterprise new quality productive forces.

2.3. Other Channels

2.3.1. Sustainable Development Impact Channels

The “new” aspect of new quality productive forces is also reflected in their environment and socio-economic sustainability, emphasizing the harmonious coexistence between humanity and nature while promoting economic greening and decarbonization and ultimately achieving high-quality development [1]. By enhancing resource allocation efficiency and optimizing energy use, AI enables enterprises to reduce dependence on natural resources and mitigate environmental impact, ensuring that economic growth aligns with ecological sustainability. In this way, AI serves as a critical driver in fostering the development of new quality productive forces [29]. In terms of environmental governance and energy saving, AI enhances resource efficiency by monitoring and analyzing resource consumption throughout production processes, minimizing energy waste while maintaining operational stability. Additionally, AI strengthens energy-efficiency control through intelligent energy management systems and AI-driven smart grids, which dynamically adjust energy distribution based on real-time demand, thereby improving overall energy utilization. This facilitates the green transformation of enterprises, mitigating their environmental footprint and guiding them toward environmentally responsible production models [30,31]. From the perspective of social responsibility and corporate governance, AI reshapes workforce structures by automating repetitive tasks, redefining roles, and fostering high-quality employment in the digital era. Through AI-driven digital skills training, enterprises can cultivate employees proficient in new technological capabilities such as data analysis, machine learning, and intelligent system operation. This not only strengthens workforce innovation and competitiveness but also ensures that employees remain adaptable in evolving digital workplaces, enriching the talent pool necessary for the advancement of new quality productive forces. Furthermore, the theory of resource acquisition suggests that a strong corporate reputation and social image enhance firms’ ability to secure essential development resources. With the help of artificial intelligence, enterprises can track and evaluate their environmental and social impacts in real time, ensuring alignment with sustainable development goals. This enables enterprises to play a more significant role in balancing social value creation and economic benefits, thereby enhancing their reputation and competitive advantage, and therefore establishing a strong social foundation for sustainable development and the cultivation of new quality productive forces [32]. Based on this, we propose hypothesis 4:

Hypothesis 4:

Artificial intelligence can enhance the level of enterprise sustainable development and then empower the cultivation of enterprise new quality productive forces.

2.3.2. Capital Market Synchronization Channel

Enterprises’ production and operation activities as well as scientific and technological innovation require adequate financial support, and the capital market is the essential arena in which enterprises secure financing. The efficient market hypothesis states that a company’s stock price fully reflects all available information about the company’s operation [33]. However, in reality, the capital market is not perfectly efficient, and there are still problems such as information asymmetry. This asymmetry may make a company’s perceived value have some inconsistency with its actual value or compared with companies in the same industry, which affects the efficiency of the capital market and may lead to the loss of investor confidence, which may further lead to financing constraints. In this context, the application of AI technology offers new opportunities for companies to mitigate information asymmetry in the capital market and enhance market synchronization. Through data analysis, predictive modeling, and related techniques, AI can significantly improve enterprises’ perception of the capital market, enabling them to better align their performance and needs with the capital market. Specifically, AI empowers corporate investment departments to conduct real-time market monitoring, capturing critical information such as stock market volatility and shifts in investor sentiment. Armed with these insights, companies are able to adopt more flexible investment and financing strategies in the capital market. Moreover, AI can help companies gain a deeper understanding of investor needs and preferences through big data analytics and social network mining tools, thereby enhancing interactions with investors. Through precise investor relationship management, companies can provide investors with customized data reports and tailored communication solutions, improve information transparency and communication efficiency, and boost investor confidence. This precise interaction not only increases investor awareness of the company, but also consolidates the company’s position in the capital market, resulting in more stable and consistent performance. The role of AI in improving capital market synchronization is also reflected in its ability to mitigate information asymmetry. With the help of AI, companies can more efficiently integrate internal operational data with external market information, establishing dynamic feedback mechanisms that ensure corporate decisions remain aligned with market demand. This synchronization not only enhances the company’s adaptability to the capital market, but also significantly expands financing channels, reduces financing costs, and further helps the company obtain the funds needed for development. Based on this, we propose hypothesis 5:

Hypothesis 5:

Artificial intelligence can enhance the synchronization of the enterprise capital market, which in turn empowers the cultivation of enterprise new quality productive forces.

2.4. The Spatial Spillover Effect of Artificial Intelligence Empowers the Cultivation of Enterprise New Quality Productive Forces

Artificial intelligence possesses a high degree of fluidity and inter-temporal dissemination. In the process of empowering enterprises with new quality productive forces, it can not only exert a direct effect on individual enterprises, but also generate a broad, far-reaching impact on regions and even the entire economic system through spatial spillover effects. The spatial spillover effect refers to the phenomenon whereby production factors—such as technology, knowledge, and innovation—originating from local enterprises not only benefit those enterprises but also extend to neighboring regions and firms via spatial dissemination [34,35]. As a generalized technology with extensive adaptability and diverse application scenarios, AI fosters a diffusion effect through mutual learning, imitation, and synergy among enterprises [36]. When an enterprise takes the lead in adopting AI technology to optimize production processes, enhance management efficiency, or improve product innovation, other enterprises or industries can learn from its success and technological achievements, subsequently disseminating these advances through the supply chain and across the broader industrial network [37], which in turn promotes the enhancement of the new quality productive forces in other enterprises. In addition, in regions with robust scientific and technological innovation capacity, the research, development, and application of AI technology are often concentrated within specific innovation clusters. These regions, characterized by high concentration of talent, capital, and technological resources, are able to continuously carry out technological research, development, and innovation to generate knowledge spillover effects that enhance the productivity of surrounding enterprises; innovations are then disseminated to neighboring enterprises through mechanisms such as startup incubation, the promotion of technological services, and so on, thereby guiding them in cultivating new quality productive forces. Furthermore, the driving effect of AI-enabled new quality productive force cultivation is not only confined solely within clusters; its innovative practices and technological advancements can also extend to enterprises or regions beyond the cluster. By providing technical services or product support in other regions, AI promotes the intelligent development of enterprises in other regions and exerts a significant impact on their new quality productive forces, based on which we propose hypothesis 6:

Hypothesis 6:

Artificial intelligence not only has a promoting effect on an enterprise’s own new quality productive force development, but also affects neighboring enterprises through spatial spillover effects.

In summary, this paper argues that AI can significantly promote the development of new productivity in enterprises and play a role in promoting digital–real integration, accelerating the flow of knowledge between enterprises, improving the level of sustainable development of enterprises, and enhancing the synchronization of the capital market of enterprises, among other channels. At the same time, this paper also argues that the promotion of AI on the new quality productive forces of enterprises has a certain spatial spillover effect. The theoretical analysis framework of this article is shown in Figure 1.

3. Research Design

3.1. Model Setting

In order to verify the impact of artificial intelligence on the new quality productive forces of enterprises, this paper adopts a two-way fixed-effects model controlling for industry and year at the same time, which is set as follows:

$$Npr{o}_{i,t}={\alpha }_0+{\alpha }_{1}A{I}_{i,t}+{\alpha }_{2}Control{s}_{i,t}+Year+Industry+{\epsilon }_{i,t}$$

(1)

where $Npro$ is the explanatory variable of enterprise new quality productive force level, $AI$ is the core explanatory variable of artificial intelligence comprehensive level, ${\alpha }_0,{\alpha }_1,{\alpha }_2$ are the regression coefficients of concern in this paper, Controls is a set of control variables, Year and Industry are year and industry fixed effects, respectively, and $\epsilon$ is a random disturbance term; $i,t$ are the individual and time differences, respectively, while the regression in this paper adopts clustering to the firm-level robust standard errors.

In order to test the mechanism effect of AI on the new quality productive forces of firms, following the suggestion of Jiang [38] on the mediation effect test, a two-step approach is taken to study the mechanism effect, and this paper establishes a mechanism test model as shown in (2):

$$M{V}_{i,t}={\beta }_0+{\beta }_{1}A{I}_{i,t}+{\beta }_{2}Control{s}_{i,t}+Year+Industry+{\gamma }_{i,t}$$

(2)

where $MV$ is the mechanism variable, $\gamma$ is the random disturbance term, ${\beta }_0,{\beta }_1,{\beta }_2$ are the regression coefficients, and other variables are defined as above.

Simultaneously, to verify the spatial spillover effect of the enterprise AI level on new quality productive forces, a series of spatial econometric tests were performed. Based on these tests, the spatial Dubin model is constructed according to (3) for analysis, and the model is set as follows [39]:

$$\begin{array}{l}Npr{o}_{i,t}={\eta }_0+\rho {\displaystyle \sum _{j=1}^{n}WNpr{o}_{i,t}}+{\eta }_{1}A{I}_{i,t}+{\eta }_{2}Control{s}_{i,t}+Year+Industry+\\ {\sigma }_{1}WA{I}_{i,t}+{\sigma }_{2}WControl{s}_{i,t}+{\epsilon }_{i,t}\end{array}$$

(3)

where $\rho$ is the coefficient of the spatial lag term of the explanatory variable new quality productive forces, $W$ is the spatial weight matrix, ${\eta }_0,{\eta }_1,{\eta }_2$ are the regression coefficients, ${\sigma }_1,{\sigma }_2$ are the spatial regression coefficients, and the other variables have the same meaning as above.

3.2. Research Process

Figure 2 shows the research process of this paper. First, a benchmark regression is performed to test the core hypothesis, which is hypothesis 1 of this paper (artificial intelligence makes a significant contribution to the cultivation of new quality productive forces in enterprises). Second, the results of the benchmark regression are tested for robustness and endogeneity, respectively, to ensure their robustness and increase the scientificity and reliability of the conclusions. Third, the mechanism test is conducted to test the mechanism path of AI affecting new quality productive forces in enterprises and to verify hypotheses 2–5. Fourth, a heterogeneity analysis is conducted to verify the heterogeneity of the effect of AI on new quality productive forces in enterprises when the property rights of enterprises, the background of executives, the degree of competition in the industry, and the nature of the industry are different. Fifth, spatial econometric tests are conducted to verify the spatial spillover effect of AI on firms’ new quality productive forces and to test hypothesis 6.

3.3. Sample Selection and Data Sources

In this paper, the A-share listed companies in Shanghai and Shenzhen from 2013 to 2022 were selected as samples to empirically test the impact of artificial intelligence on the new quality productive forces of enterprises. The annual report data of the listed companies were obtained from the official websites of SSE and SZSE, the patent data were obtained from the CNRDS platform and the State Intellectual Property Office of China, the other financial data were obtained from the CSMAR database, and the industry employment data were obtained from the China Labour Statistics Yearbook. The acquired data were processed as follows: ① As the spatial econometric analysis method requires highly balanced panel data, enterprises with missing data for any year during 2013–2022 were excluded. ② Enterprises with important indicators seriously missing were excluded. ③ Enterprises with ST/*ST/PT situations during the survival of the sample period were excluded. ④ All continuous variables were shrunk by Winor2 at the upper and lower 1% quartiles. ⑤ The moving average method was used to fill in some missing values. The final sample consisted of 769 companies and 7690 samples.

3.4. Explained Variable: New Quality Productive Forces (Npro)

Drawing on the methods of Song [40] and Wang [41], we construct the indicator system for the new quality productive forces of enterprises across three dimensions: laborers, labor materials, and labor objects, and we use the entropy method to synthesize the indicators. Specifically, the labor level includes R&D personnel salary, headcount ratio, and the ratio of highly educated personnel; the labor material level includes the ratio of manufacturing cost, the ratio of R&D depreciation and amortization, the ratio of leasing expense, and the ratio of direct input; and the labor object includes the ratio of intangible assets, the total asset turnover ratio, and the derivative of the equity multiplier. Specific indicator descriptions as well as indicator weights are shown in

Table 1. New quality productive force indicator system.

	Variable Name	Variable Description	Weight
Laborers	R&D staff salary ratio	(R&D Expense—Salary and Wages)/Operating Income	28%
	Percentage of R&D personnel	Number of R&D Personnel/Number of Employees	4%
	Percentage of highly educated personnel	Number of Bachelor’s Degree or Above/Number of Employees	3%
Labor Materials	R&D depreciation and amortization ratio	(R&D Expense—Depreciation and Amortization)/Operating Income	28%
	Percentage of R&D lease fee	(R&D Expenses − Lease Fee)/Revenue	2%
	Percentage of R&D direct investment	(R&D Expenses − Direct Inputs)/Revenue	28%
Labor Object	Intangible assets	Intangible Assets/Total Assets	7%

Source: own work.

.

3.5. Explanatory Variables: The Comprehensive Level of Artificial Intelligence (AI)

Currently, there are various opinions in the academic community regarding how to measure the level of AI. Some approaches include the text analysis method that counts word frequency in annual reports [42], calculating the ratio of AI-related intangible assets to total assets from the notes of annual reports [43], and measuring the number of industrial robots deployed by enterprises [44,45]. While each of these methods offers certain advantages, they also present shortcomings: The text analysis method may produce statistical distortions of the indicators due to deliberate exaggeration and misrepresentation in the annual report by the management in an offer to attract investors’ attention and bolster shareholders’ confidence [46]; using the ratio of intangible assets can be problematic because differences in accounting standards may prevent it from accurately reflecting a company’s overall level of AI. Moreover, indicators such as the number of AI patents or the number of industrial robots are insufficient to capture the full extent of AI application within enterprises [43]. As a result, the existing literature on AI measurement exhibits notable limitations, highlighting the need to develop a more comprehensive indicator that encompasses the application, strategy, and innovation aspects of AI.

In this paper, the comprehensive dimensions of AI are categorized into three levels: strategic position, practical application, and innovation development. First, the annual report of a company reflects its current and future strategic direction and development plans to a certain extent, so the frequency of AI-related keywords in the annual report can indicate the company’s attention to AI and its strategic intentions, so the strategic position dimension is measured by employing text analysis to calculate the frequency of AI-related terms in the annual report, following the research method of Yao et al. [42]. An AI dictionary is constructed by crawling the annual reports of listed companies and processing the text using the Jieba tokenizer function of Python software (Version: Python3.11) to extract the words in annual reports that match the AI dictionary; these words are statistically counted and summarized into word frequencies. Given that the data have obvious right-skewed characteristics, the word frequency is transformed by adding one before taking the logarithm.

Second, given that most enterprises currently apply AI technology to industrial robots, and drawing from existing studies, the micro-level AI penetration is used as a proxy variable for the AI application level [12], with the measurement method following Wang [45]. To calculate the industry-level industrial robot penetration index, first, the stock of industrial robots in the industry for a given year is divided by the total number of employment in the industry, and then the ratio of the enterprise’s production sector employees in that year to the industry’s median number of production sector employees of all enterprises is multiplied by the industry’s robot penetration to construct the enterprise-level industrial robot penetration index.

Again, patents are a direct manifestation of a company’s R&D and technological innovation achievements, particularly in the digital economy and artificial intelligence, where companies safeguard their innovations through patents. Before a patent is granted, a company typically engages in substantial research and development activities focused on creating technological solutions and innovations. However, as some patents obtained by enterprises may not be directly related to AI, and given that AI is highly associated with the digital economy driven by digitization [47], this study only considers the digital economy patents obtained by enterprises as a measure of AI innovation development, and the number of these patents is transformed by applying the logarithm to the patent count, which serves as a proxy variable for AI innovation development.

Finally, the entropy weighting method is used to synthesize the three to obtain the indicator (AI) that measures the comprehensive level of artificial intelligence of enterprises.

3.6. Control Variables

To control the impact of omitted variables on the results of this paper, several control variables are selected based on the existing literature. These include Tobin’s q-value (tobin), the profitability of total assets (roa), the growth rate of operating income (growth), the liquidity ratio (liqui), the percentage of independent directors (indep), two positions in one (both), the gearing ratio (lev), and other variables selected as control variables. The definitions and descriptions of the variables are shown in

Table 2. Definition and description of variables.

Variable Name	Variable Symbol	Variable Description
New quality productive forces	Npro	Evaluation index of new quality productive forces constructed by entropy weight method
Artificial intelligence level	AI	Comprehensive level of artificial intelligence measured at the strategic, application, and innovation levels
Tobin’s q-value	tobin	Market value of the firm/replacement cost of assets
Profitability of total assets	roa	Firm’s annual earnings divided by the value of total assets
Growth rate of operating income	growth	Previous year’s revenue/current year’s revenue
Current ratio	liqui	Ratio of total current assets to total current liabilities
Independent directors	indep	Number of independent directors/number of directors
Both	both	Whether the chairman and general manager are the same person; if yes, take 1; otherwise, take 0
Gearing ratio	lev	Total assets/total liabilities of the company

Source: own work.

, and the descriptive statistics of the variables are shown in

Table 3. Descriptive statistics of main variables.

Variables	Sample Size	Mean	Standard Deviation	Minimum	Median	Maximum
Npro	7690	5.611	2.352	1.589	5.251	14.611
AI	7690	0.629	0.515	0.009	0.452	2.214
lev	7690	0.415	0.198	0.052	0.408	0.927
roa	7690	0.030	0.068	−0.302	0.032	0.192
tobin	7690	2.113	1.358	0.818	1.675	8.733
growth	7690	0.239	0.573	−0.680	0.107	3.565
liqui	7690	2.351	2.392	0.288	1.615	15.896
both	7690	0.269	0.444	0.000	0.000	1.000
indep	7690	0.375	0.053	0.333	0.357	0.571

Source: own work.

.

4. Empirical Tests

4.1. Benchmark Regression

Table 4. Benchmark regression results.

Variables	(1) Npro	(2) Npro
AI	0.686 *** (3.55)	0.696 *** (3.65)
Control variables	NO	YES
Year and industry fixed effects	YES	YES
Constant	5.170 *** (41.50)	4.654 *** (10.57)
Sample size	7690	7690

T-values in parentheses, *** p < 0.01, source: own calculations.

reports the results of the benchmark regression of this paper. Models 1 and 2 display the regression coefficients of the level of artificial intelligence on firms’ new quality productive forces before and after the inclusion of control variables, respectively. The regression coefficients of artificial intelligence in both models are significantly positive at the 1% level, which initially proves that hypothesis 1 (artificial intelligence has a significant contribution to empowering the cultivation of new quality productive forces in enterprises) is valid.

4.2. Robustness Tests

In order to verify the robustness of the core conclusions of this study and to eliminate potential research errors resulting from omitted variables, a series of robustness tests were conducted as follows.

4.2.1. Endogeneity Test

1.

Instrumental variable method

There may exist a degree of reverse causality between new quality productive forces and artificial intelligence; to address possible endogeneity problems, this study employs the instrumental variable method for estimation. The first instrumental variable utilized is the Bartik instrumental variable (Bartik_IV). The Bartik instrumental variable is constructed based on the product of the initial share of the unit of analysis and the overall growth rate. This method essentially leverages the initial share composition of an individual unit and the overall growth rate to estimate values over time. The resulting estimates exhibit a high correlation with actual values while remaining uncorrelated with other residual terms [48,49]. The specific construction process is as follows: First, the growth rate of AI levels in each subindustry for each year is calculated as the overall growth rate; Second, the overall growth rate is multiplied by the AI level of each enterprise at the end of the previous fiscal year to derive the Bartik instrumental variable. In terms of correlation, the enterprise AI level is closely linked to the industry AI level, in terms of exclusivity, the Bartik variable does not directly affect the new quality productive forces of enterprises.

Furthermore, drawing on Chen [50], this study also uses the mean AI level of other firms within the same industry (mean_AI) as an instrumental variable. This variable meets the criteria of relevance and exclusivity: in terms of relevance, firms’ AI development is inevitably affected by other firms in the industry, while in terms of exclusivity, it is difficult for firms’ new quality productive forces to be directly affected by the AI level of other firms in the industry.

Table 5. Instrumental variable method.

Variables	AI Phase I	Npro Phase II	AI Phase I	Npro Phase II
Bartik_IV	2.463 *** (0.373)
AI		1.110 *** (0.280)		1.785 *** (0.353)
mean_AI			9.811 *** (0.555)
Kleibergen–Paap rk LM statistic		28.372 ***		135.609 ***
Kleibergen–Paap rk Wald F statistic		43.606 [16.38]		313.005 [16.38]
Year and Industry Fixed Effects	YES	YES	YES	YES
Control variables	YES	YES	YES	YES
Number of samples	7690	7690	7690	7690

T-values in parentheses, *** p < 0.01, source: own calculations.

reports the regression results of the instrumental variables. The results of the Kleibergen–Paap rk LM statistic and Kleibergen–Paap rk Wald F statistic indicate that both instrumental variables pass the weak instrumental variable test and the unidentifiable test, indicating that the instrumental variables selected in this study are reasonable; meanwhile, in the second-stage regression results, AI’s regression coefficients remain significantly positive at the 1% level, confirming that the core conclusions of this study still hold even after addressing the endogeneity problem by mutual causality.

2.

Heckman two-stage model

Another reason for the endogeneity problem may stem from the existence of a certain sample self-selection bias in the model, i.e., due to the selection of the sample being not random or representative enough, resulting in a bias in the final results of a study; therefore, this paper uses the Heckman two-stage model to test in order to solve the sample self-selection problem. The key to this method is to select exclusive dummy variables and dummy variables for whether the explanatory variables in the original regression equation are observed or taken values. Firstly, for the exclusionary dummy variables, referring to the study of Zhang [51], this paper selects the AI_mean of other enterprises in the same industry in the same year as mentioned above as the exclusionary constraint variable; secondly, in terms of the dummy variables, when the value of the explanatory variable AI is too small or even close to 0, the corresponding sample enterprises are often ignored by the researcher and thus lead to the problem of self-selection, so this paper takes the dummy variable of whether the explained variable is observed or taken in the original regression equation to solve the sample self-selection problem. To solve the selection bias problem, this paper takes the average value of the industry AI level as the boundary. When the enterprise AI level is greater than the industry average, it will be assigned as 1; otherwise, it will be 0. Based on this, the AI dummy variable (AI_dummy) is constructed.

Subsequently, based on the above variables, the first step is to regress the artificial intelligence dummy variable (AI_dummy), the artificial intelligence mean (AI_mean) of other enterprises in the same industry in the same year, and the control variables in the baseline regression with the Probit model and to calculate the inverse Mills ratio (Imr) of the model. In the second step, the inverse Mills ratio is substituted into the benchmark regression model, thus enabling the correction of the model.

Table 6. Heckman two-stage model.

Variables	AI_dummy	Npro
AI		0.686 *** (0.000)
Imr		125.70 * (0.060)
AI_mean	0.237 *** (0.000)
Constant	−0.318 *** (0.000)	−117.389 * (0.068)
Control variables	YES	YES
Industry/year fixed effects	YES	YES
Sample size	7690	7690

T-values in parentheses, * p < 0.10, *** p < 0.01, source: own calculations.

reports the results of the test of Heckman’s two-stage model; in the first step, the regression coefficient of AI_mean is significantly positive at the 1% level, and the estimated coefficient of Imr in the second step is significantly positive at the 10% level, while the regression coefficient of AI remains positive at the 1% level. The above test illustrates that there is a certain sample self-selection bias in the sample selection of this paper, and the core conclusions of this paper still hold after the correction of the model.

4.2.2. Robustness Test

1.

Extended time observation window

The impact of AI on firms’ new quality productive forces may have certain unobserved omitted variables at other time points, so this paper uses an extended time observation window for robustness testing. This is done by generating F.Npro, F2.Npro, and F3.Npro variables by front-loading the explanatory variables by one to three periods and generating L.AI, L2.AI, and L3.AI variables by lagging the explanatory variables by one to three periods.

Table 7. Extended time observation window.

Variables	F.Npro	F2.Npro	F3.Npro	(1) Npro	(2) Npro	(3) Npro
AI	0.758 *** (3.68)	0.827 *** (3.81)	0.872 *** (3.96)
L.AI				7.548 *** (3.62)
L2.AI					8.201 *** (3.69)
L3.AI						8.519 *** (3.73)
Year and industry fixed effects	YES	YES	YES	YES	YES	YES
Control variables	YES	YES	YES	YES	YES	YES
Constant	4.605 *** (9.91)	4.370 *** (8.59)	4.340 *** (8.25)	4.629 *** (9.99)	4.627 *** (9.55)	4.647 *** (9.38)
Sample size	6921	6152	5381	6921	6149	5383

T-values in parentheses, *** p < 0.01, source: own calculations.

reports the results of the robustness test for the extended time observation window, and it can be seen that the regression coefficients of AI, L.AI, L2.AI, L3.AI are always significantly positive at the 1% level, which is consistent with the benchmark regression, indicating that the benchmark regression results of this paper are still valid and the core conclusions are robust after taking into account the omitted variables at other time points.

2.

Other robustness tests

In addition to solving the endogeneity problem and extending the time observation window, this paper also conducts the following robustness tests: 1. Excluding the stock market crash years: The catastrophic performance of the stock market may be transmitted to the real economy through the financial system, which may have certain impacts on the real enterprises, so the years when a stock market crash occurs are excluded from the robustness test. 2. Excluding the epidemic years: The global public health event that started in early 2020 had a huge impact on China’s economy, and the daily business activities of enterprises in the subsequent years have been affected unprecedentedly, so the epidemic that occurred in 2020–2022 is proposed in the model. 3. Replacement of explanatory variables: This test refers to Yao [42], using the word frequency statistics of artificial intelligence fuzzy words (AI_1) as a substitute variable for the regression. 4. High-dimensional fixed effects: The baseline regression in this paper controls for year and industry fixed effects, but there may be unobserved omitted variables at the city level, so city fixed effects are added and tested using the year–industry–city fixed-effects model.

Table 8. Other robustness tests.

Variables	Excluding the Stock Market Crash	Excluding Outbreaks	Substitution of Explanatory Variables	Higher-Dimensional Fixed Effect
AI	0.687 *** (3.59)	0.832 *** (4.16)		0.722 *** (3.92)
AI_1			0.020 *** (5.68)
Year and industry fixed effects	Yes	Yes	Yes	Yes
Control variables	Yes	Yes	Yes	Yes
Constant	4.649 *** (10.41)	4.804 *** (10.36)	4.888 *** (11.62)	4.263 *** (10.36)
Sample size	6921	4612	7690	7690

T-values in parentheses, *** p < 0.01, source: own calculations.

reports the remaining robustness test regression results of this paper, and it can be seen that the regression coefficients of AI with stock market crash excluded, epidemic effect excluded, explanatory variables replaced, and high-dimensional fixed effects controlled are all significantly positive at the 1% level, indicating that the core findings of this paper are robust.

4.3. Mechanism Testing

4.3.1. Digital–Reality Integration Effect Influence Channels

The construction of digital–real integration (TechConv) indicators refers to the study of Huang [52]; based on the main classification, the patent IPC classification of digital–real industry technology integration is defined as if the patent IPC main classification belongs to the non-digital industry technology, and at least one of the patents cited in the patent classification pertains to digital industry technology, the patent is defined as a digital–real industry technology fusion behavior of the enterprise. Finally, the counts are summed up at the firm-year level and logarithmically transformed to obtain a proxy variable for digital–real integration.

Column 2 of

Table 9. Mechanism tests.

Variables	TechConv	Flow	Sus	Syn
AI	0.134 ** (2.03)	2.879 *** (2.69)	0.189 *** (3.93)	0.012 ** (2.40)
Year and industry fixed effects	YES	YES	YES	YES
Control variables	YES	YES	YES	YES
Constant	0.085 (0.64)	−1.261 * (−1.83)	4.085 *** (23.40)	0.442 *** (20.02)
Sample size	7690	7690	7690	7690

T-values in parentheses, * p < 0.10, ** p < 0.05, *** p < 0.01, source: own calculations.

reports the digital–real integration channel of AI-enabled new quality productive force development, and the results show that the regression coefficient of AI, which is 0.134, is significantly positive at the 5% level, indicating that one of the key channels of AI-enabled cultivation of new quality productive forces is the promotion of its digital–real integration, which forms the “artificial intelligence development–digital–real integration–new quality productive force enhancement” channel, and hypothesis 2 (artificial intelligence can promote the digital–real integration of an enterprise, which in turn empowers the cultivation of enterprise new quality productive forces) is proved.

4.3.2. Knowledge Flow Effect Influence Channels

The citation relationship of firms’ patents can be viewed as the construction of a knowledge flow network [53], where the number of citations of firms’ patents is summed based on the firm-year level, and the number of citations of firms’ patents is derived by adding one to the number of times to take the logarithm, thus measuring the level of firms’ knowledge flow.

Column 3 of Table 9 reports the knowledge flow channels of AI-enabled new quality productive force development, and the results show that the regression coefficient of AI, which is 2.879, is significantly positive at the 1% level, indicating that AI can effectively promote the level of knowledge flow among enterprises and promote the cultivation of new quality productive forces by strengthening knowledge flow and knowledge effect, forming the channel of “AI development–knowledge flow effect–new quality productive force enhancement”. Hypothesis 3: Artificial intelligence can improve the knowledge flow effect of enterprises, which in turn promotes the cultivation of enterprises’ new quality productive forces.

4.3.3. Sustainable Development Channels

Sustainable development is a theory about the coordinated development of nature, science and technology, and economy and society, and the ESG (Environmental, Social, and Governance) index is largely consistent with the concept of sustainable development, and it can well measure the performance of enterprises in environmental and social responsibility and corporate governance, so this paper uses the CSI ESG index to measure the sustainability level (Sus) of enterprises.

Column 1 of Table 9 reports the sustainable development channel of AI-enabled new quality productive force development, and the results show that the regression coefficient of AI, which is 0.189, is significantly positive at the 1% level, indicating that AI can effectively improve the level of sustainable development of the enterprise and promote the cultivation of new quality productive forces through the improvement of the level of sustainable development, forming the channel of “artificial intelligence development–sustainable development–enhancement of new quality productive forces”, and hypothesis 4 of this paper is proved. Hypothesis 4 (artificial intelligence can improve the level of enterprise sustainable development and then empower the cultivation of enterprise new quality productive forces) is proved by the channel of “AI development–sustainable development–new quality productive force enhancement”.

4.3.4. Capital Market Synchronization Impact Channels

The construction of stock price synchronization (Syn) in this paper is based on the weighted average of market capitalization outstanding [54], which is first modeled in order to compute the goodness of fit $R^2$ of weekly stock returns:

$$R_{i,w,t}={\phi }_0+{\phi }_1{R}_{M,w,t}+{\phi }_2{R}_{M,w-1,t}+{\phi }_3{R}_{I,w,t}+{\phi }_4{R}_{M,w-1,t}+{\nu }_{i,w,t}$$

(4)

$R_{i,w,t}$$R_{M,w,t}$$R_{I,w,t}$ denote the return on stock i in week w of year t taking into account the reinvestment of cash dividends, the liquid market capitalization-weighted average return on all A-shares of the market in week w of year t; and the liquid market capitalization-weighted average return on industry I in week w of year t with stock i excluded. $R^2$ is then logarithmized, which in turn yields an indicator of stock price synchronicity.

$$Sy{n}_{i,t}=\ln ({R_{i,t}}^2/(1-R_{i,t}^2))$$

(5)

Column 4 of Table 9 reports the sustainable development channels of AI-enabled new quality productive force development, and the results show that the regression coefficient of AI, which is 0.012, is significantly positive at the 5% level, indicating that AI can effectively improve the level of capital market synchronization of enterprises and promote the cultivation of new quality productive forces through the enhancement of capital market synchronization, forming the channel of “artificial intelligence development–enhancing capital market synchronization–new quality productive force enhancement”. With the development of the “artificial intelligence–improve capital market synchronization–new quality productive force enhancement” channel, hypothesis 5 (artificial intelligence can enhance the synchronization of the enterprise capital market, which in turn empowers the cultivation of enterprise new quality productive forces) has been proved.

4.4. Heterogeneity Analysis

4.4.1. Heterogeneity of Property Rights

Differences in property rights backgrounds may lead to the ability of AI empowerment to cultivate new quality productive forces. State-owned enterprises (SOEs) are not solely profit-oriented due to their state-owned background, and they need to take on certain social responsibilities and political tasks under government guidance. In contrast, non-SOEs often need to rely on themselves more, facing greater market competition and risks. As a result, they tend to develop and apply AI more intensively in order to enhance productivity and secure their position in the competitive market. Columns 1 and 2 of

Table 10. Heterogeneity analysis.

Variables	Property Right Heterogeneity		Industry Competition Heterogeneity		Executives Digitalization Background Heterogeneity		Industry Nature Heterogeneity
	Nationalized Business	Non-State Enterprise	High Level	Low Level	Digital Background	No Digital Background	Strategic Industry	Other Industries
AI	0.335 (1.19)	0.971 *** (4.96)	0.739 *** (3.33)	0.340 * (1.74)	1.539 *** (5.52)	0.388 ** (2.16)	0.799 *** (3.23)	0.408 ** (2.00)
Year and industry fixed effects	YES	YES	YES	YES	YES	YES	YES	YES
Control variables	YES	YES	YES	YES	YES	YES	YES	YES
Constant	5.83 *** (7.37)	4.078 *** (8.60)	4.127 *** (7.14)	5.639 *** (12.46)	2.574 ** (2.27)	5.092 *** (11.27)	3.853 *** (5.92)	5.635 *** (9.79)
Sample size	4665	3025	5410	2280	1112	6578	4172	3518

T-values in parentheses, * p <0.10, ** p < 0.05, *** p < 0.01, source: own calculations.

report the heterogeneity of AI for new quality productive forces, and the results show that, for non-state-owned enterprises, the regression coefficients are significantly positive at the 1% level compared to the insignificant coefficients for state-owned enterprises. This suggests that the role of AI in promoting new quality productive forces varies significantly depending on the ownership structure of the firm.

4.4.2. Heterogeneity of Industry Competition

In highly competitive industries, firms often strive to enhance their core competitiveness through technological innovation and organizational change [55] to avoid being squeezed out of the market and to secure their place within it. In contrast, in relatively less competitive industries, this sense of competition may not be as pronounced, diminishing the significant role of AI in promoting new quality productive forces. In this paper, we use the Herfindahl–Hirschman Index (HHI), calculated based on the total assets of the companies, to measure the degree of competition within an industry. Generally speaking, an HHI between 0.15 and 0.25 indicates a relatively balanced distribution of market shares, reflecting an average level of competition. Given the distribution of the sample data, this study establishes that a value greater than 0.18 indicates a relatively centralized distribution of market shares and a low degree of competition, while a value greater than 0.18 indicates a relatively decentralized distribution of market shares and a high degree of competition. Columns 3 and 4 of Table 10 report the heterogeneity of AI’s impact on new quality productive forces across industries. The results show that the effect of AI-enabled new quality productive forces is more significant for firms in highly competitive industries compared to firms with a low degree of competition, suggesting that there is a certain degree of industry competitive heterogeneity in this effect.

4.4.3. Heterogeneity in the Digital Background of Executives

In modern corporate governance structures, company executives are responsible for the day-to-day operations and the implementation of the board’s strategy. This structure means that the qualities of the executives can influence the company’s business planning and development philosophy to a significant extent. For instance, executives with information technology backgrounds are more likely to plan corporate digital strategies over the long term compared to their counterparts, thereby reducing the impact of short-term thinking and boosting the output of corporate technological innovations [56]. Consequently, whether or not executives have digital backgrounds may contribute to the heterogeneity in the development of AI-enabled new quality productive forces. In this paper, we construct a dummy variable for executive digital background, assigning a value of 1 to executives with an information technology education or relevant tenure experience and 0 otherwise. Columns 5 and 6 of Table 10 report the heterogeneity of executive characteristics in AI-enabled fostering of new quality productive forces. The results show that the regression coefficients for firms where executives have a digital background are more significant, suggesting that executives’ backgrounds indeed introduce some heterogeneity in the effect of AI empowerment.

4.4.4. Heterogeneity in the Nature of the Industry

There are notable differences between strategic industries and traditional industries in terms of technological suitability, innovation drive, policy support, and market competition. In the State Council Decision on Accelerating the Cultivation and Development of Strategic Emerging Industries, China has identified sectors such as energy-saving and environmental protection, information, biology, high-end equipment manufacturing, new energy, new materials, and new energy automobiles as key areas for development. Strategic emerging industries, being more forward-thinking and innovative, are better positioned for future growth compared to traditional industries. As a result, the impact of AI in enabling the development of new quality productive forces may differ between the two. To examine this, this paper divides industries into two categories: strategic emerging industries, which include sectors like automobile manufacturing, pharmaceutical manufacturing, special equipment manufacturing, and other industries identified for vigorous development in the “14th Five-Year Plan”, and traditional industries. The regression results in Table 10, Columns 7 and 8, show that the AI effect is more significant in strategic emerging industries, indicating a stronger promotional effect and highlighting the heterogeneity of AI’s impact across different industry types.

5. Spatial Measurement Analysis

5.1. Spatial Correlation Tests

The prerequisite for spatial econometric analysis is that the object under study has some spatial relevance, so the first step is to measure the global Moran’s I of the explanatory variable firms’ new quality productive forces, and Moran’s index is calculated as follows:

$$I={\displaystyle \frac{{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^n{W}_{ij}}}(Npr{o}_i-\stackrel{\_\_\_\_\_\_\_}{Npro})(Npr{o}_j-\stackrel{\_\_\_\_\_\_\_}{Npro})}{S^2{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^n{W}_{ij}}}}}$$

(6)

where $S^2$ is the sample variance; $n$ is the number of samples; the mean value of firms’ new quality productive forces is denoted, and the other variables have the same meanings as above. The value range of Moran’s I index is (−1, 1), with I > 0 denoting a positive spatial correlation, i.e., that the values of similarity are inclined to be clustered together, or in other words, that there is spatial agglomeration; I < 0 denotes a negative spatial correlation, i.e., that the values of similarity are inclined to be scattered, which is spatially expressed as spatial disaggregation.

Table 11. Spatial Moran index.

Year	Spatial Geographic Distance Matrix			Spatial Economic Geography Nested Matrix
	Moran’s I	Z-Value	p-Value	Moran’s I	Z-Value	p-Value
2013	0.096 ***	3.531	0.000	0.119 ***	4.511	0.000
2014	0.081 ***	4.641	0.000	0.109 ***	4.133	0.000
2015	0.091 ***	3.372	0.000	0.101 ***	3.820	0.000
2016	0.125 ***	4.641	0.000	0.111 ***	4.242	0.000
2017	0.108 ***	4.009	0.000	0.110 ***	4.210	0.000
2018	0.101 ***	3.764	0.000	0.109 ***	4.169	0.000
2019	0.105 ***	3.897	0.000	0.111 ***	4.238	0.000
2020	0.125 ***	4.640	0.000	0.127 ***	4.830	0.000
2021	0.116 ***	4.311	0.000	0.120 ***	4.576	0.000
2022	0.123 ***	4.579	0.000	0.125 ***	4.750	0.000

*** p < 0.01, source: own calculations.

reports the spatial Moran’s index of the explanatory variables for the years 2013–2022 under the spatial geographic distance matrix and the spatial economic geographic nesting matrix, and the results show that the Moran’s index of Npro is significantly positive at the 1% level for either matrix, suggesting that the explanatory variables in this paper have significant spatial correlation, hence the need for a spatial econometric analysis.

5.2. Spatial Weighting Matrix

The basis of a spatial econometric analysis is the construction of a spatial weight matrix; this paper selects the spatial geographic distance matrix, which is more widely used in academia, for carrying out spatial econometric benchmarking regression, and at the same time, it applies the spatial economic geographic nested matrix to the robustness test. The matrix construction is described below.

5.2.1. Spatial Geographic Distance Matrix

The spillover effect of AI on firms’ new qualitative productivity levels may be significant from a geographical perspective. To investigate this, this paper constructs a geographical distance matrix for benchmarking regression studies. The geographic distance matrix is an N × N-dimensional matrix, which is created using the square of the inverse of the geographic distance between firms, in accordance with common practice. The geographic distance between firms is calculated based on the latitude and longitude where their offices are located using the geopy function of Python software, which is then arranged one-to-one in a matrix. The formula for calculating the geographical distance matrix is shown below, where location denotes the coordinates of a firm and i and j denote the firms.

$$\left\{\begin{array}{l}{(\frac{1}{{\mathit{location}}_i - {\mathit{location}}_j})}^2\hspace{1em}i\ne j\hfill \\ 0\hspace{1em}i\ne j\hfill \end{array}\right.$$

(7)

5.2.2. Spatial Economic Geography Nested Matrix

In addition to geographic distance, the economic strength of firms can also lead to spatial spillover effects. Building on the concept of spatial economic geography weight matrix in macro spatial econometric analysis [57], this paper constructs a nested spatial economic geography by integrating a spatial financial matrix with a spatial geographic distance matrix. The spatial financial matrix is derived from the logarithm of the asset size of enterprises. The matrix construction formula is as follows, where lnSize denotes the logarithm of the enterprise’s asset size. This matrix will be used for the robustness test of spatial measurement in this study.

$$\left\{\begin{array}{l}{(\frac{\ln {\mathit{size}}_i - \ln {\mathit{size}}_j}{{\mathit{location}}_i - {\mathit{location}}_j})}^2\hspace{1em}i\ne j\hfill \\ 0\hspace{1em}i\ne j\hfill \end{array}\right.$$

(8)

5.3. Selection of Spatial Econometric Regression Models

The next step is to determine which model to use for spatial econometric regression analysis; the mainstream econometric models are the spatial Dubin model (SDM), spatial error model (SEM), spatial autoregressive model (SAR), etc. Spatial Dubin models (SDMs), spatial autoregressive models (SARs), and spatial error models (SEMs) each have unique advantages and applicable scenarios. The spatial Dubin model (SDM) takes into account the spatial spillover effects of both dependent and independent variables and is suitable for analyzing complex spatial correlation mechanisms and multi-channel spatial influences. The spatial autoregressive model (SAR) focuses mainly on the spatial dependence of the dependent variable and is suitable for investigating how the results of one region are affected by the results of neighboring regions, but it assumes that there is no spatial spillover of the independent variable. Spatial error models (SEMs), on the other hand, are used to capture spatial error dependence in the model and are suitable for dealing with the effects of random disturbances between the dependent variable and the spatial structure, thus correcting for estimation bias. The testing of the model is divided into the Lagrange Multiplier Test (LM test), Wald test, likelihood ratio test (likelihood ratio, LR test), etc., and the results of each test are shown in the table.

Table 12. Test results of the spatial measurement model.

Wald Test		LR Test (Which Model to Choose)		LR Test (What Fixed Effects to Choose)		Hausman Test
Wald-SEM	Wald-SAR	LR for SEM	LR for SEM	LR-ind	LR-time.	/
46.19 ***	40.22 ***	40.16 ***	46.05 ***	580.30 ***	10016.72 ***	5026.65 ***

*** p < 0.01, source: own calculations.

reports the results of the tests, where the Wald test is used to determine whether the SDM model can be degraded to an SEM model or an SAR model, and the LR test is used to identify which fixed-effects model to choose. The results show that the LR test (SAR), LR test (SEM), Wald (SAR), and Wald (SEM) statistics are all significant at the 1% level, indicating that the spatial Dubin model (SDM) should be chosen and that the SDM model cannot be degraded into the SEM model or the SAR model; therefore, combining the above tests, this paper chooses the SDM model for econometric analysis.

After determining what kind of spatial panel model to use, the mode of the spatial econometric model is to be determined, and the Hausman test indicates that the choice of fixed-effects model is reasonable. Generally speaking, spatial econometric models exist in several forms, with no fixed effects, time or spatial fixed effects, and time and spatial double fixed effects. The LR test was used for screening, and the test results are shown in the table; the results show that the time and spatial fixed-effects model has joint significance, so the model is determined to be a time and space double-fixed spatial Dubin model.

5.4. Spatial Econometric Regression Analysis

Column 1 of

Table 13. Spatial measurement regression results.

Variables	Npro
	Spatial Geographic Distance Matrix	Spatial Economic Geography Nested Matrix
AI	0.209 *** (4.31)	0.207 *** (4.26)
WAI	0.279 *** (3.49)	0.175 *** (2.05)
Control variables	YES	YES
Time and space fixed effects	YES	YES
ρ	0.060 *** (3.96)	0.053 *** (3.57)
Brochure	7690	7690
Sigma2_e	1.304 *** (61.98)	1.102 *** (59.15)

T-values in parentheses, *** p < 0.01, source: own calculations.

reports the results of the benchmark regression of spatial measures based on the SDM model, which shows that the spatial term coefficient of AI is significantly positive, once again illustrating the need to test the spatial spillover effect of AI on the level of firms’ new quality productive forces. Moreover, the regression coefficient of AI is still significantly positive at the 1% level after adding the spatial term coefficient, which further proves that the core conclusions of this paper are valid, and at the same time, the spatial term coefficient of AI (WAI) is significantly positive at the 1% level, which indicates that with the further development of enterprise AI, the degree of its empowerment to cultivate the level of new quality productive forces will be even stronger, and the new quality productive forces of the enterprises in the vicinity will be subject to spatial spillovers. Hypothesis 6 (artificial intelligence not only has a promoting effect on an enterprise’s own new quality productive force development, but also affects neighboring enterprises through spatial spillover effects) is established.

Meanwhile, in order to further enhance the robustness of the conclusions of this paper and avoid the omission of variables brought about by a single matrix test, the spatial economic geography nested matrix is used as the basis for the robustness test, and the results are shown in Table 13, Column 2. The coefficients of AI and its spatial terms are consistent with the baseline regression after the replacement of the matrix, which suggests that the baseline regression of spatial measurement in this paper is robust.

For the regression results of the SDM model, a further decomposition of the regression results into direct and indirect effects is necessary to better understand the spatial spillover effects of the explanatory variables on the dependent variables. Columns 1 and 2 of

Table 14. Decomposition results for the spatial Dubin model.

	Geographic Distance Matrix			Economic Geography Nested Matrix
	Ratio	Z-Value	p-Value	Ratio	Z-Value	p-Value
aggregate effect	0.528 ***	5.55	0.000	0.413 ***	4.18	0.000
indirect effect	0.312 ***	3.83	0.000	0.202 **	2.34	0.019
direct effect	0.216 ***	4.33	0.000	0.211 ***	4.23	0.000

** p < 0.05, *** p < 0.01, source: own calculations.

report the further decomposition of the SDM model regression results under the two spatial weight matrices. The direct effect represents the impact of local explanatory variables on local explanatory variables, while the indirect effect represents the impact of local explanatory variables on neighboring explanatory variables, i.e., the spatial spillover effect. The total effect is the sum of the direct and indirect effects. The results show that the regression coefficients of the direct, indirect, and total effects are significant, further confirming the core conclusion of this paper that AI can significantly empower the cultivation of new quality productive forces of enterprises. Additionally, the findings verify the existence of a significant spatial spillover effect of AI on the new quality productive forces of enterprises, reinforcing that the core conclusions of this study remain robust.

6. Conclusions and Recommendations

6.1. Conclusions

The core issues focused on in this paper are whether AI can promote the development of new quality productive forces of enterprises, through what mechanism it promotes the development of new quality productive forces of enterprises, and whether AI has spatial spillover effects on new quality productive forces of enterprises; few studies have explored these issues in depth. To this end, this paper proposes hypotheses 1 to 6, takes Chinese A-share listed companies as samples from 2013 to 2022, combines patent data and the theory of numerical–realistic fusion, and empirically examines the impact of AI development empowering the cultivation of new quality productive forces, the mechanism of the role, and the spatial spillover effect from the panel regression and the spatial perspective. The results of this study show that hypotheses 1 to 6 are accepted, the core problems of this paper are solved, and the following conclusions are drawn: (i) Artificial intelligence can significantly promote the cultivation of new quality productive forces and enhance the level of new quality productive forces, and this conclusion still holds after a series of endogeneity and robustness tests. (ii) AI mainly empowers the cultivation of new quality productive forces through the channels of promoting sustainable development of enterprises, facilitating the integration of digital and real systems, promoting the flow of knowledge, and improving the synchronization of the stock market. (iii) The role of AI in promoting the development of new quality productive forces is more obvious when the enterprise is a private enterprise, its executives have a digital background, it is in an industry with fierce market competition, and it is a strategic industry. (iv) The spatial econometric analysis based on the spatial Dubin model illustrates that the impact of AI on new quality productive forces has a significant positive spatial spillover effect; i.e., the development of the level of AI in this enterprise can promote the enhancement of the level of new quality productive forces in the neighboring enterprises or regions.

6.2. Recommendations

First, we recommend formulating and enhancing AI-driven policies for digital intelligence empowerment and spatial layout and focusing on fostering new levels of productivity in enterprises by enhancing the development of AI empowerment. As the core driving force of the fourth industrial revolution, AI has already demonstrated significant disruptive potential, necessitating a systematic policy framework to maximize its economic and societal impact. Effective policy formulation must go beyond the technological innovation and application of AI; it must also incorporate regional economic dynamics, industrial structures, and the spatial distribution of technological resources to ensure that AI’s advantages translate into tangible drivers of economic growth and enterprise transformation. On the basis of digital intelligence empowerment, policies should promote the deep integration of artificial intelligence with traditional industries, leveraging intelligent transformation and digitalization to improve the productivity level of enterprises and reduce labor costs and operational barriers. Additionally, AI policies should account for the factors of spatial layout considerations, given that technological innovation spillovers exhibit significant regional characteristics. The agglomeration and diffusion of innovation resources determine the competitiveness gap between different regions. Therefore, in policy planning, a differentiated layout based on regional characteristics is needed. Supporting the establishment of AI innovation highlands, by encouraging synergistic cooperation among universities, research institutions, and enterprises, will promote the formation of an innovation ecosystem in the region, accelerate the transformation of innovation results, promote the coordinated development of the regional economy, and form a situation of complementary advantages. In addition, the policy should focus on solving the obstacles that enterprises may face in the process of applying AI technology. The promotion and application of AI technology, especially among small and medium-sized enterprises (SMEs), is often constrained by technological bottlenecks, financial pressures, and talent shortages. Therefore, a policy can provide multi-level services such as technical support, data sharing, and talent training by building a public service platform for the application of AI technology, thus helping enterprises to break through technological constraints and enhance their innovation capabilities.

Secondly, we recommend strengthening the role of AI development in empowering the cultivation of new quality productive forces and assisting the development of new quality productive forces with a data-driven core concept. The essence of AI technology lies in its ability to collect, analyze, and process large-scale data to uncover latent patterns and optimize decision-making. Data serve not only as the fuel for the development of AI technology, but also as a key resource for innovation in enterprise production and management. In the production process, through real-time monitoring and analysis of vast amounts of operational data, enterprises can dynamically adjust production plans, optimize resource allocation, and improve production efficiency. Relying on a data-driven production mode can not only reduce manual intervention and improve the level of automation, but also realize refined management, thus significantly improving the level of productivity. Moreover, a data-driven approach also provides a scientific basis for innovative decision-making. Through accurate analysis of market demand, customer behavior and industry trends, enterprises can quickly adapt to changes in the external environment and enhance market competitiveness. A data-driven approach can promote the formation of enterprise innovation ecology in the process of strengthening AI empowerment. Artificial intelligence technology not only relies on the input and processing of massive amounts of data, but also mines new knowledge and innovation points through in-depth learning and analysis of data. This innovation is reflected not only in the technical level, but also in the change of business model and management style. Through the data-driven intelligent system, it discovers inefficient links in the traditional management mode, optimizes organizational processes and business models, improves operational efficiency, prompts enterprises to form a more flexible innovation mechanism, breaks down information barriers, and promotes synergistic innovation in different departments and segments, which further promotes the enhancement of enterprises’ new quality productive forces.

Third, AI industrial layout policies should be tailored to the heterogeneity of industries and sectors, enabling the classification and precision of AI’s spatial distribution. From a spatial perspective, the application and development of AI technology usually have a significant agglomeration effect, and the spatial distribution of its impact is often closely related to the developed degree of regional innovation ecology. Therefore, its industrial layout policy should fully take into account the technological needs and development stages of different industries and sectors, and there are obvious differences in the depth and breadth of the application of AI technology in different industries and sectors. In formulating the AI industrial layout policy, it is necessary to differentiate the layout according to the technological needs and application scenarios of each industry, ensuring AI technologies align with sector-specific requirements, and to avoid wasting resources and repetitive construction. At the same time, the AI industrial layout policy should also reflect the principle of precise policy application to achieve the efficient allocation of policy resources and optimal results. Precise policy application entails the strategic allocation of policy resources according to the development potential of each industry and region. For industries and regions with strong innovation capacity and high development potential, policies should provide greater support and preferential treatment to foster cluster effects, enabling them to assume a leading role in AI-driven development. Conversely, for industries and regions with relatively weak foundations, policies should prioritize guidance and targeted support, assisting them in overcoming development bottlenecks by providing technical support and establishing cooperation platforms. This approach will gradually enhance their position within the AI industry supply chain, fostering balanced and sustainable AI-driven industrial growth.

6.3. Limitations and Prospects

Restricted by data availability and differences in regulatory and accounting standards, the scope of this paper is limited to Chinese A-share listed companies in analyzing the impact of AI development on firms’ new quality productive forces and its spatial spillovers. However, this limitation makes it difficult for this study to fully reflect the potential impact of AI on a global scale. We speculate that under the influence of different market environments, human backgrounds, economic growth patterns, and other factors, the effect of AI on firms’ new quality productive forces may vary, and in some cases even show the opposite results to the conclusions of this paper: in economically developed countries with mature markets, AI may further improve firms’ efficiency through highly optimized digitalization processes, while in economically less developed regions, its impact may be limited due to insufficient infrastructure or low technology acceptance.

We are currently in an era of rapid development of AI, a highly powerful technological tool that is not only changing the way businesses operate, but also driving leaps and bounds in academic research. The development of AI has opened up new possibilities for understanding economic phenomena in different market environments. By using AI for large-scale data mining, linguistic processing, and model optimization, we are expected to overcome the barriers of cultural, linguistic, and institutional differences and more comprehensively analyze the impact of AI on the new quality productive forces of enterprises under the influence of different market conditions, human customs, and social structures. Meanwhile, data collection and sharing on a global scale still face many challenges at present, including data protection, cross-border transfer restrictions, and significant differences in disclosure requirements between countries. However, with the gradual increase in data liquidity and the continued global pursuit of an open and transparent market environment, cross-border data research is expected to develop further in the future.

Based on this, we expect to broaden our research perspective in the future and extend the scope of analysis from China to the whole world to explore the impact of AI on firms’ new quality productive forces in different regions and its spatial spillover effects. This will not only provide a theoretical basis for the application and promotion of AI globally, but also provide practical guidance for different economies in formulating relevant policies.