How Does Artificial Intelligence Application Enable Sustainable Breakthrough Innovation? Evidence from Chinese Enterprises

Abstract

As the core driving force of the new generation of industrial revolution, artificial intelligence technology has brought new opportunities for empowering enterprise innovation and advancing sustainability. Focusing on Chinese A-share listed enterprises and based on the unbalanced panel data from 2011 to 2023, this study systematically examines the relationship mechanism between artificial intelligence (AI) application and enterprise breakthrough innovation, and further explores the mediating effect of knowledge recombination and the moderating role of market competition. The empirical results show that AI application has a significant promoting effect on the enterprise breakthrough innovation. Knowledge recombination creation and knowledge recombination reuse play mediating roles in the relationship between the AI application and enterprise breakthrough innovation, forming the key transmission path for empowering breakthrough innovation with AI. In addition, market competition positively moderates the relationship between knowledge recombination and enterprise breakthrough innovation and strengthens the driving effect of knowledge recombination on innovation output, thus fostering more sustainable competitive advantages.

1. Introduction

In the surging tide of the rapid development of the digital economy, artificial intelligence (AI), as the core driving force leading a new round of technological revolution and industrial transformation, is penetrating into various aspects of enterprise operation at an unprecedented speed and breadth, profoundly reshaping the innovation mode and competitive landscape of enterprises in ways that are increasingly intertwined with sustainability. For enterprises, under the dual pressures of the digital economy and globalized competition, breakthrough innovation has become a critical path to break industry monopolies and achieve corner overtaking and long-term sustainable development. Breakthrough innovation refers to an innovation model that, based on major technological breakthroughs, scientific discoveries, or cognitive leaps [1], fundamentally changes the core logic of products, services, processes, or technological systems, thereby achieving leapfrog improvements in performance, efficiency, cost, or user experience [2,3], and exerting profound impacts on the existing industry structure, market rules, or user demands [4,5]. Breakthrough innovation differs from both disruptive innovation and radical innovation. Breakthrough innovation focuses on the existing mainstream market and breaks through bottlenecks with technological or performance leaps (e.g., 5G compared with 4G) [6,7]. It targets the clear high-end needs of mainstream customers. Initially, products are often expensive due to high technological investment [8], but they can directly compete head-on with existing products, helping leading enterprises with core resources further consolidate their market advantages [9,10]. Disruptive innovation starts from the market edge, targeting low-end needs or new scenarios [11]. Its initial products are insufficient in mainstream performance but are cheap and convenient (e.g., early PCs) [12]. It opens up new technological trajectories, and after gradual iteration and upgrading may eventually disrupt the existing market pattern [13]. Radical innovation breaks away from the original trajectory and builds a brand-new technological system (e.g., from internal combustion engine technology of fuel vehicles to electric drive technology of electric vehicles) [14,15]. In the early stage, it faces immature technology and low market acceptance, but can reconstruct industry standards and even create new markets, mostly driven by startups or cross-industry players [16,17].

Breakthrough innovation, as a key path to break existing technological paradigms and market rules and promote leapfrog development of enterprises, has always been a focus of attention in academia and practice in terms of its generation mechanism and influencing factors. Some scholars focus on the enterprise itself, arguing that entrepreneurs [18,19], entrepreneurial spirit [20,21] and corporate culture [22] constitute the primary drivers of breakthrough innovation. Other studies highlight that internal resources and capability constraints—such as funding, knowledge, and talent—serve as the fundamental conditions for enterprises to achieve sustainable breakthrough innovation [23,24]. Meanwhile, additional research emphasizes that external environmental factors are pivotal to enterprises’ realization of breakthrough innovation, noting that elements like the market environment [25], connections with external organizations [26,27], innovation policies [28,29] and innovation systems [30] can all exert a significant impact on enterprise sustainable breakthrough innovation.

In recent years, as AI technology has continued to make strides in data processing, pattern recognition, and self-learning, there has been a growing number of cases where enterprises leverage AI for product R&D, process optimization, and business model innovation [31,32]. AI not only permeates various aspects of enterprise R&D, production, and management as an emerging technology [33], but also has a breakthrough impact on entrepreneurial decision-making patterns, resource allocation efficiency, innovative network connection methods, and even the implementation path of innovation policies with its data processing capabilities, self-learning capabilities, and scenario adaptation capabilities [34,35,36]. Existing research generally believes that the promotion effect of AI application on enterprise innovation is mainly reflected in the four core dimensions of technology, resources, organization and market. At the technical level, AI improves R&D efficiency with deep learning and big data analysis, breaks through human cognitive limitations, expands technical possibilities, and helps enterprises realize systematic improvement of product technology or functions [37,38,39]. At the resource level, AI can integrate internal and external fragmented knowledge, release human resources for high-level innovation tasks, and optimize R&D funding with predictive models [40,41]. At the organizational level, AI can break cross-department and cross-enterprise boundaries to achieve collaborative innovation, stimulate organizational learning ability through feedback mechanism, cultivate innovation culture of all employees through flat communication, and enhance flexibility of enterprise management [42,43]. At the market level, AI can spawn cross-border innovation scenarios, accurately tap into user needs, capture market opportunities, and accelerate the commercialization of outcomes by relying on marketing tools, thus injecting all-round impetus into enterprise innovation [44,45]. Existing literature mostly focuses on the direct impact of AI on innovation output, while research on how it affects the special form of breakthrough innovation is relatively scarce, especially the exploration of intermediate transmission mechanisms.

Knowledge, as the core resource of enterprise innovation, plays a crucial role in the flow, integration, and reconstruction of the innovation process [46]. The Knowledge Based View points out that a company’s competitive advantage comes from its accumulation, integration, and application of knowledge [47,48]. Knowledge Recombination refers to the recombination or secondary utilization of existing knowledge elements by enterprises to generate new knowledge outcomes or application scenarios [49]. Based on differences in recombination methods, knowledge recombination can be further divided into knowledge recombination innovation (generating new knowledge combinations through creative integration) and knowledge recombination reuse (adapting existing knowledge outcomes to multiple scenarios) [50]. In the era of AI, fundamental changes have taken place in the ways knowledge is generated, disseminated, and applied. AI technology can not only quickly process massive amounts of data and convert it into structured knowledge, but also simulate the association and combination of knowledge through algorithms, thereby providing efficient tools for knowledge recombination [51,52]. Knowledge recombination provides enterprises with the core cognitive foundation and source of ideas for sustainable breakthrough innovation by breaking existing knowledge boundaries, integrating heterogeneous information, and forming novel knowledge connections [53,54]. Incorporating knowledge recombination as an intermediary variable into the analytical framework can not only unravel the black box of “AI application empowering sustainable breakthrough innovation” and reveal how technology applications are transmitted to innovation results through knowledge level changes, but it can also provide a key pivot for accurately grasping the mechanism of technology empowering innovation and supporting enterprises’ sustainable development, thereby deepening the understanding of the relationship between AI application and enterprise breakthrough innovation.

Moreover, enterprises’ innovative behaviors do not exist in isolation, instead they are profoundly influenced by the external market environment. As a core indicator for measuring the market environment, market competition intensity reflects the degree of fierceness in the competition among enterprises for resources and market shares [55]. Due to its high risk, sustainable breakthrough innovation is far more sensitive to the competitive environment than incremental innovation. In a fiercely competitive market, to maintain their competitive edge or achieve a corner overtaking, enterprises tend to break away from path dependence and explore new innovation directions through knowledge recombination [56]. In markets with moderate competition, companies may rely more on incremental innovation to maintain stable returns. Existing studies have largely overlooked the dynamics of how the innovative effects of knowledge recombination are influenced by the competitive environment. Clarifying the moderating effect of market competition can provide precise guidance for enterprises to optimize their strategies according to competitive dynamics and consolidate the foundation for sustainable breakthrough innovation development.

Based on this, this study focuses on the impact of AI application on enterprise breakthrough innovation, delves into the mediating effects of knowledge recombination (including knowledge recombination innovation and knowledge recombination reuse), and further analyzes the moderating role of market competition intensity in the relationships between them. Using the unbalanced panel data of Chinese A-share listed companies from 2011 to 2023, this study empirically examines and reveals the inherent connections between variables through fixed-effects regression analysis, aiming to provide theoretical references and practical insights for enterprises to better utilize AI technology, optimize knowledge recombination strategies, and enhance sustainable breakthrough innovation. The marginal contribution of this study is as follows: firstly, it deepens the understanding of the relationship between AI application and enterprise breakthrough innovation, injecting long-term impetus into enterprise sustainable development. Existing studies mainly explore the direct impact of AI on the quantity of innovation outputs or innovation efficiency, with insufficient attention to the breakthrough nature of innovation. However, sustainable breakthrough innovation is precisely the core engine for enterprises to break through growth bottlenecks and achieve sustainable development. This study conducts an in-depth analysis of the impact of AI application on enterprise sustainable breakthrough innovation, providing a more precise theoretical perspective and empirical evidence for understanding the unique value of AI in driving enterprises to achieve leapfrog innovation breakthroughs and consolidate the foundation for sustainable development. Secondly, it expands the application scenarios of the knowledge recombination theory, offering a new path for the development of enterprise sustainable innovation capabilities. Knowledge recombination is a key mechanism for enterprises to continuously generate innovation momentum, directly related to the achievement of their sustainable development goals. This study focuses on the internal transmission mechanism of the impact of AI application on enterprise sustainable breakthrough innovation, revealing the microscopic pathway through which AI application drives breakthrough innovation by empowering knowledge recombination. It addresses the relative lack of exploration into the intermediate links between the two in previous studies, providing theoretical support for enterprises to strengthen their sustainable innovation capabilities by optimizing the knowledge recombination process. Thirdly, it reveals the moderating effect of market competition intensity, providing contextualized guidance for enterprises to achieve sustainable development in a dynamic environment. The sustainable innovation development of enterprises is inseparable from accurate adaptation to the external competitive environment, and the linkage mechanism between knowledge recombination and market dynamics constitutes the core bond for balancing innovation input and risk control. This study incorporates market competition intensity into the analytical framework, uncovers the linkage mechanism between external market dynamics and internal knowledge recombination, and thus offers a more contextually adaptive theoretical explanation for understanding the relationship between knowledge recombination and breakthrough innovation. This helps enterprises formulate appropriate innovation strategies under different competitive situations, enabling the steady enhancement of their sustainable development capabilities.

The remainder of the study is presented below. Section 2 presents the theoretical analysis and the research hypotheses. Section 3 presents the research methodology and data. Section 4 gives the empirical results. A discussion is presented in Section 5. Finally, it is summarized in Section 6 and concludes with relevant recommendations.

2. Theoretical Analysis and Research Hypothesis

2.1. AI Application and Breakthrough Innovation

Artificial intelligence, as a general technology, boasts a strong technological spillover effect [57]. This technological characteristic forms an organic synergy with the massive information storage of big data, accelerating the diffusion, promotion, and in-depth application of technologies. It provides core momentum for enterprises to break through traditional technological paths and reconstruct innovation paradigms, and can effectively drive the realization of enterprise sustainable breakthrough innovation, and injecting lasting momentum into the long-term sustainable innovation development of enterprises [58]. Firstly, AI application provides key support for enterprise sustainable breakthrough innovation by efficiently processing massive amounts of heterogeneous data. Compared with the limited search scope caused by the constraints of human cognition, the application of AI technology can intelligently explore more options and significantly accelerate the speed of problem-solving [59]. By virtue of advanced technologies such as natural language processing, image recognition, and deep learning, enterprises can break down data barriers, deeply integrate structured information and unstructured content scattered across different channels, and build multi-dimensional data models covering the entire business chain [60]. This integration not only enables centralized data management but also, through algorithms to mine hidden connections between data, allows originally isolated information to generate synergistic value, providing comprehensive data support for enterprise sustainable breakthrough innovation decisions [42]. Secondly, AI can drive enterprises’ business model innovation and build a new system for value creation and acquisition [45]. In terms of demand insight, AI can conduct real-time monitoring and intelligent analysis of large-scale, dynamically changing user-related data [61]. Through continuous tracking and in-depth analysis of users’ behavioral patterns, preference characteristics, and potential demands, enterprises can break through the limitations of traditional research methods, accurately capture unmet hidden needs in the market, and even predict the evolutionary trends of demands [62]. This forward-looking grasp of demand can guide enterprises to jump out of existing product or service frameworks, explore innovative directions, and provide clear R&D orientation for sustainable breakthrough solutions. Finally, AI application can optimize enterprises’ organizational management models and create an internal environment conducive to sustainable breakthrough innovation. At the organizational structure level, AI technology breaks down the information barriers of traditional hierarchical systems. By building an intelligent information sharing platform, it realizes real-time circulation and collaborative processing of cross-departmental and cross-level data [63]. This flat and agile structure can respond more quickly to innovation needs, providing organizational support for cross-domain collaborative innovation. In terms of decision-making mechanisms, by integrating multi-dimensional data from both internal and external sources of the enterprise, AI constructs intelligent analysis models to provide managers with data-driven decision-making suggestions [64]. This changes the traditional decision-making model that relies on empirical judgments and reduces the impact of subjective biases on the direction of sustainable breakthrough innovation.

Based on this, this study proposes the following research hypotheses:

H1. 

AI application can significantly promote enterprise breakthrough innovation.

2.2. AI Application and Knowledge Recombination

AI application provides sustainable breakthrough innovation tools for enterprise knowledge recombination creation. By breaking down knowledge barriers, activating tacit knowledge, and fostering cross-boundary integration, it promotes the reconstruction of knowledge systems from fragmentation to systematization and from static to dynamic. Firstly, at the level of knowledge integration, AI technology can conduct comprehensive scanning and structured processing of the scattered explicit knowledge and tacit knowledge within the enterprise, weaving isolated knowledge points into an interconnected knowledge network [65]. This integration not only eliminates knowledge silos between departments but also identifies intersections of knowledge across different fields, laying the foundation for interdisciplinary innovation. Secondly, in terms of knowledge activation, AI simulates human cognitive processes through deep learning algorithms, enabling in-depth excavation and reinterpretation of existing knowledge. By analyzing the R&D team’s historical solutions and market feedback data, AI can extract hidden success patterns and failure lessons, converting them into reusable knowledge modules [66]. Meanwhile, AI-driven intelligent question-answering systems can respond to employees’ knowledge needs in real time, and continuously optimize the form of knowledge expression during interactions, transforming accumulated knowledge from static storage into dynamically flowing innovation resources [67]. Finally, in terms of cross-boundary integration, AI can integrate external industry-leading knowledge, technological trends with internal core capabilities of enterprises to generate brand-new knowledge combinations [68]. By monitoring cutting-edge academic achievements and market dynamics worldwide, AI can identify connections between emerging technologies and industries. Combined with an enterprise’s own technological accumulation, it can propose cross-domain innovation concepts. This kind of knowledge recombination creation is no longer limited to the existing boundaries of the enterprises, but promotes enterprises to construct more forward-looking knowledge systems through open knowledge interaction and lays a cognitive foundation for sustainable breakthrough innovation, and helping enterprises achieve sustainable development goals while achieving technological breakthroughs.

Based on this, this study proposes the following research hypotheses:

H2a. 

AI application has a significant positive impact on enterprise knowledge recombination creation.

AI application enhances the efficiency and depth of enterprise knowledge recombination reuse by building an intelligent knowledge management system, transforming accumulated knowledge resources into momentum for the continuous generation of innovative value. Firstly, at the level of knowledge retrieval and matching, AI technology breaks through the limitations of traditional keyword-based retrieval, enabling accurate knowledge positioning based on semantic understanding [65]. Natural language processing algorithms can parse users’ demand intentions, and by combining the hierarchical relationships and association strengths of concepts in the knowledge graph, quickly screen out the most relevant knowledge modules. For example, in the product R&D process, the technical requirements put forward by R&D personnel can be converted by AI into knowledge retrieval criteria, which automatically match solutions, technical parameters and lessons learned from historical projects. This enables the instant reuse of knowledge and reduces resource waste caused by redundant work. Secondly, in terms of knowledge adaptation and reconstruction, AI can reorganize and optimize existing knowledge according to specific scenarios. By analyzing the goals, constraints, and environmental characteristics of the current task, AI can extract relevant knowledge elements from the knowledge base and reconstruct an adapted solution framework [69]. For instance, in supply chain optimization, AI can integrate historical logistics data, inventory management experience, and real-time market demand to reorganize a dynamically adjustable supply chain strategy, enabling existing knowledge to generate new value in new scenarios. Finally, at the level of knowledge iteration and inheritance, AI promotes the cyclic upgrading of knowledge recombination reuse through a continuous learning mechanism. After intelligent systems record feedback data during the process of knowledge application, they will automatically update knowledge association relationships and applicable conditions, enabling the knowledge system to continuously evolve with business development [67]. Meanwhile, AI-driven virtual mentor systems can transform the tacit knowledge of senior employees into structured teaching content and deliver it to new employees through personalized learning paths. This ensures the effective inheritance and reuse of core knowledge, avoids knowledge loss caused by personnel turnover, and provides stable knowledge support for the sustainable innovation and development of enterprises.

Based on this, this study proposes the following research hypotheses:

H2b. 

AI application has a significant positive impact on enterprise knowledge recombination reuse.

2.3. The Mediating Role of Knowledge Recombination

AI application promotes enterprise knowledge recombination creation through the integration of internal and external knowledge, cross-domain matching, and dynamic updates, providing a sustainable breakthrough knowledge foundation and continuous momentum for enterprises to achieve long-term sustainable innovation and development. Firstly, knowledge recombination creation can break enterprises’ path dependence and open up new innovation directions. By integrating knowledge elements across fields and disciplines, knowledge recombination creation helps enterprises break through the limitations of existing technical frameworks and business models [70,71]. When scattered knowledge nodes form new interconnected networks, they can give rise to disruptive technical concepts or product ideas, enabling enterprises to break free from reliance on their traditional advantageous fields and enter entirely new market tracks [68]. Secondly, knowledge recombination creation can help break through industry bottlenecks and reshape industry competition rules. The new knowledge systems formed through recombination and creation often involve breakthroughs in existing cognition, which can directly tackle long-standing technical difficulties or market pain points within the industry [72]. This sustainable breakthrough reconstruction of existing cognition enables enterprises to bypass the gradual development path of traditional technological iteration and create disruptive value through a “dimension-reduction strike”. It may not only redefine product forms but also completely reshape the competitive rules of the industry. Finally, knowledge recombination creation can enhance strategic foresight and help seize potential market opportunities. The absorption of external cutting-edge knowledge during the process of knowledge recombination creation enables enterprises to gain earlier insights into the evolution trends of technologies and changes in market demands [73]. Strategic decisions formulated based on new knowledge systems enable enterprises to proactively lay out plans for future markets, build insurmountable competitive barriers, seize the opportunities during the window period of sustainable breakthrough innovation, and help achieve long-term sustainable development.

Based on this, this study proposes the following research hypotheses:

H3a. 

Knowledge recombination creation plays a mediating role between AI application and enterprise breakthrough innovation.

AI application promotes enterprise knowledge recombination reuse, injecting core momentum into sustainable breakthrough innovation through the integration and activation of internal and external knowledge, cross-domain matching and reuse, and dynamic cyclic utilization. Firstly, knowledge recombination reuse can reduce innovation costs and increase the possibility of continuous breakthroughs. By redeveloping existing knowledge resources, knowledge recombination reuse reduces repeated investment in the innovation process. Enterprises can carry out modular recombination based on existing technical accumulation and empirical data, which not only reduces the cost of trial and error but also accelerates the speed of innovation iteration, providing resource guarantee for continuous breakthrough innovation [74]. Secondly, knowledge recombination reuse can strengthen knowledge synergy effects and improve innovation efficiency. The re-adaptation and integration of knowledge in the process of reuse can promote knowledge sharing among different departments and projects, and activate the tacit knowledge deposited [72]. This synergy enables scattered innovation elements to form a cohesive force, shortening the cycle from the generation of ideas to the implementation of outcomes, and allowing sustainable breakthrough innovation concepts to be transformed into actual productivity more quickly [75]. Finally, knowledge recombination reuse can ensure the continuity of innovation and support the implementation of long-term strategies. The dynamic knowledge management system built through knowledge recombination reuse ensures the inheritance and evolution of enterprises’ core knowledge assets [76]. Even in the face of personnel turnover or changes in the external environment, the sustainability and stability of innovation capabilities can still be maintained through knowledge reuse mechanisms. This allows sustainable breakthrough innovation strategies to be advanced over the long term and prevents innovation disruptions caused by knowledge gaps.

Based on this, this study proposes the following research hypotheses:

H3b. 

Knowledge recombination reuse plays a mediating role between AI application and enterprise breakthrough innovation.

2.4. The Moderating Role of Market Competition

Market competition intensity can positively moderate the relationship between knowledge recombination creation and enterprise breakthrough innovation by accelerating the transformation process from knowledge recombination creation to sustainable breakthrough innovation, enhancing the resource investment and risk tolerance of enterprises in knowledge recombination, strengthening the barrier protection of innovation achievements. Firstly, market competition pressure can accelerate the transformation from knowledge recombination creation to sustainable breakthrough innovation. Under high-intensity market competition, enterprises face a sharp increase in pressure for survival and development, and the market life cycle of existing technologies and products is significantly shortened [77]. At this point, knowledge recombination creation is no longer an “alternative option” for enterprises, but becomes a core means to break through competitive dilemmas. Enterprises will promote the integration and reconstruction of cross-domain knowledge at a faster pace, and quickly transform new knowledge systems into sustainable breakthrough innovative achievements with market impact [78]. This acceleration in the transformation speed significantly amplifies the driving effect of knowledge recombination creation on sustainable breakthrough innovation amid fierce competition, helping enterprises quickly establish differentiated advantages to cope with competitive challenges. Secondly, market competition intensity enhances the impact of knowledge recombination creation on sustainable breakthrough innovation by increasing the level of resource input and risk-bearing capacity for knowledge recombination creation. In a low-competition environment, enterprises may be more conservative in their investment in knowledge recombination creation, tending to focus on incremental knowledge optimization. However, under high competitive pressure, enterprises are more willing to bear the uncertainties of cross-domain knowledge recombination and increase resource allocation to the acquisition of cutting-edge knowledge and the establishment of cross-departmental collaboration mechanisms [79]. This greater willingness to bear risks and higher resource density enable knowledge recombination creation to achieve deeper cognitive breakthroughs and generate more sustainable breakthrough innovation concepts, thereby exerting a stronger moderating effect in its relationship with breakthrough innovation. Finally, competitive intensity strengthens the protective effect of the innovation barriers formed by knowledge recombination creation. Fierce competition compresses the living space for imitators, making the unique knowledge combinations generated by knowledge recombination creation harder to replicate. At this point, sustainable breakthrough innovation based on new knowledge systems can quickly form exclusive advantages, which in turn incentivize enterprises to further deepen knowledge recombination creation. This forms a positive cycle of “knowledge recombination—breakthrough innovation—competitive advantage”, thereby strengthening the positive relationship between the knowledge recombination creation and breakthrough innovation under high competitive intensity.

Based on this, this study proposes the following research hypotheses:

H4a. 

Market competition can positively moderate the relationship between knowledge recombination creation and breakthrough innovation.

Market competition intensity positively moderates the facilitating effect of knowledge recombination reuse on sustainable breakthrough innovation by accelerating transformation, enhancing input and depth, and strengthening sustainability. Firstly, high market competition intensity accelerates the transformation from knowledge recombination reuse to sustainable breakthrough innovation. Fierce market competition compresses enterprises’ innovation cycles, forcing them to activate their existing knowledge stock more efficiently [80]. At this juncture, knowledge recombination reuse is no longer a tool for incremental optimization, but a core means to quickly respond to market changes. By reconfiguring and adapting existing knowledge modules, enterprises can formulate innovative solutions in a short period of time. Its supporting role in sustainable breakthrough innovation is significantly amplified, helping enterprises seize the initiative in competition. Secondly, market competition pressure enhances the resource allocation and depth of knowledge recombination reuse. The fiercer the market competition, the more enterprises tend to increase investment in knowledge management systems, driving knowledge recombination reuse to develop in greater depth [81]. It shifts from simple knowledge reuse to cross-scenario creative recombination, tapping into the hidden innovative value within accumulated knowledge. Such increased resource input and expanded application depth make knowledge recombination reuse more likely to spawn sustainable breakthrough innovative outcomes, thereby enhancing enterprises’ competitive resilience. Finally, a highly competitive environment reinforces the sustainability of innovation through knowledge recombination reuse. Under intense market competition, enterprises need to maintain their advantages through continuous innovation, and knowledge recombination reuse provides a guarantee for such continuity. By constantly iterating the recombination models of existing knowledge, enterprises can stably deliver innovative outcomes, avoiding innovation disruptions caused by knowledge gaps. Competitive pressure compels enterprises to establish dynamic knowledge reuse mechanisms, thereby strengthening their long-term driving effect on sustainable breakthrough innovation.

Based on this, this study proposes the following research hypotheses:

H4b. 

Market competition can positively moderate the relationship between knowledge recombination reuse and breakthrough innovation.

In summary, the theoretical model of this study is shown in Figure 1.

3. Research Design

3.1. Data Source and Sample Selection

To test the relationship between the AI application, knowledge recombination and enterprise breakthrough innovation, this study selects the unbalanced panel data of A-share manufacturing listed companies from 2011 to 2023 as the sample, for two reasons. On the one hand, as a major part of the real economy, the sustainable breakthrough innovation capability of the manufacturing industry is of great significance for promoting high-quality development of China’s economy. On the other hand, AI technology has achieved rapid iteration and widespread application in the past decade, and enterprises have mainly systematically integrated it into their production and operation processes during this period. Therefore, this study sets the sample period as 2011–2023. In addition, considering the lag in measuring the knowledge recombination variable, this study selects the patent data of A-share listed companies from 2005 to 2023. Among them, enterprise data are sourced from the CSMAR database, and enterprise annual reports are from Juchao Information Network. This study also processes the data as follows: (1) Only counting patent data with valid legal status. (2) Excluding samples with severe missing data. (3) Excluding listed companies in the financial and insurance sectors. (4) Excluding ST and *ST samples. (5) Winsorizing all continuous variables at the 1% upper and lower tails.

3.2. Variable Measurement

3.2.1. Dependent Variable

Breakthrough Innovation (HTI). Most of the existing literature uses patent data to measure enterprise breakthrough innovation. Referring to the research methods of Pang et al. [82], and this study uses the number of authorized invention patents obtained by enterprises during the sample period as an indicator to measure their sustainable breakthrough innovation output. The reasons are as follows: Compared with utility model patents and design patents, invention patents have stricter requirements on the novelty, inventiveness and practicality of technology. According to the relevant provisions of the Patent Law, invention patents need to have prominent substantive features and significant progress, which is highly consistent with the sustainable breakthrough characteristics emphasized by “breakthrough innovation” and can well reflect the significant progress of enterprises in technological innovation.

3.2.2. Independent Variable

AI Application (AI). Drawing on the research of Yao et al. [83], we crawl the enterprise annual reports of 13 years from 2011 to 2023 from the CNINFO website, uses 73 high-frequency words in its AI dictionary, and applies the ‘jieba’ text analysis tool in Python 3.9 to analyze the annual reports of listed companies and calculate the frequency of occurrence of keywords in the AI dictionary, as shown in

Table 1. AI Dictionary.

Form	AI Keywords
Basic techniques and algorithms	Machine Learning, Artificial Intelligence, Deep Learning, Neural Networks, Speech Recognition, Image Recognition, Data Mining, Feature Recognition, Speech Synthesis, Knowledge Graph, Support Vector Machines (SVMs), Long Short-Term Memory (LSTMs), Recurrent Neural Networks, Reinforcement Learning, Pattern Recognition, Distributed Computing, Edge Computing, Smart Computing, Deep Neural Networks
Hardware and infrastructure	AI chips, smart sensors, smart chips, wearables, big data platforms, cloud computing, IoT
Areas of application	AI products, machine translation, computer vision, human–computer interaction, intelligent supervision, intelligent banking, intelligent insurance, human-computer collaboration, intelligent investment advisor, intelligent education, intelligent customer service, intelligent retail, intelligent agriculture, intelligent voice, augmented reality, virtual reality, intelligent medical care, intelligent speakers, voiceprint recognition, intelligent government, automatic driving, intelligent transportation, convolutional neural network, face recognition, feature extraction, Driverless, smart home, Q&A system, smart body, business intelligence, smart finance, big data processing, smart aging, big data marketing, big data risk control, big data analytics, smart voice, human-computer dialogue, big data operation, biometrics, natural language processing, robotic process automation

. After adding 1, the natural logarithm is taken as the indicator of the enterprise’s AI application level.

3.2.3. Mediating Variable

Knowledge Recombination (KR). Referring to the methods of Yu & Yu [84], the knowledge recombination is measured by using IPC classification numbers. By identifying the way of knowledge recombination, it can be divided into knowledge recombination creation (KRC) and knowledge recombination reuse (KRR). The specific method is as follows. (1) Construct a knowledge recombination judgment experimental group and a control group, where the patent data applied by the enterprise during the t-6, t-5, t-4 years are used as the control group, and the patent data applied by the enterprise during the t-3, t-2, t-1 years are used as the experimental group. (2) Exclude patent data containing only one classification number, and extract the elements before the “/” in the IPC classification number as representatives of knowledge elements for pairwise combination. (3) Compare the knowledge element combination situations of the experimental group and the control group to determine whether a certain knowledge element combination appears for the first time. (4) When a knowledge element combination first appears in the experimental group, it is recorded as knowledge recombination creation. When a knowledge element combination appears in both the control group and the experimental group, it is recorded as knowledge recombination reuse. (5) Calculate the number of times of knowledge recombination creation and knowledge recombination reuse of enterprise i in t-3, t-2, t-1 years, respectively, and use them as the recombination creation and recombination reuse of enterprise i in year t.

3.2.4. Moderating Variable

Market Competition (PCM). Market competition intensity is an important indicator used to measure the intensity of competition among enterprises. The higher the market competition intensity, the more intense the competition among enterprises in the market, and the weaker the control ability of a single enterprise over the market. Referring to the research of Guo et al. [85], we use the Herfindahl index (HHI) to measure the market competition intensity, which is represented by the sum of the squares of the percentage of the current main operating income of each enterprise in the industry to the total main operating income of the whole industry. Given that a higher HHI implies a lower level of market competition, in order to transform the original HHI into a positive measure of market competition intensity, we use 1 minus HHI to represent PCM.

3.2.5. Control Variables

Considering other factors that may affect enterprise breakthrough innovation, this study selects the following indicators as control variables to be included in the regression model. (1) Enterprise Age (Age), measured by the number of years since the enterprise was founded. (2) Enterprise Performance (ROA), represented by the return on assets of the enterprise. (3) Revenue Growth Rate (RGR), measured by the growth amount of enterprise revenue divided by the total revenue of the previous year’s enterprise. (4) Ownership Concentration (OC), measured by the shareholding proportion of the top ten shareholders. (5) Board Scale (BS), measured by the total number of directors constituting the board of directors.

The symbols and calculations for the above variables are shown in

Table 2. Variable description table.

Var Types	Vars	Description	Measurement
Dependent Variable	HIT	Breakthrough Innovation	The number of authorized invention patents by enterprises.
Independent Variable	AI	AI Application	The frequency of occurrence of keywords in the AI dictionary.
Mediating Variable	KRC	Knowledge Recombination Creation	The number of times of knowledge recombination creation of enterprise i in t-3, t-2, t-1 years.
	KRR	Knowledge Recombination Reuse	The number of times of knowledge recombination reuse of enterprise i in t-3, t-2, t-1 years.
Moderating Variable	PCM	Market Competition	1-HHI
Control Variables	Age	Enterprise Age	The number of years since the enterprise was founded.
	ROA	Enterprise Performance	The return on assets of the enterprise.
	RGR	Revenue Growth Rate	The growth amount of revenue divided by the total revenue of the previous year’s enterprise.
	OC	Ownership Concentration	The shareholding proportion of the top ten shareholders.
	BS	Board Scale	The total number of directors constituting the board of directors.

.

3.3. Model Selection

To verify the research hypotheses, we construct the following main effect model.

$${\mathrm{HIT}}_{\mathrm{it}}={\mathrm{\alpha }}_{01}+{\mathrm{\alpha }}_{11}{\mathrm{AI}}_{\mathrm{it}}+{\mathrm{\alpha }}_{21}{\mathrm{Controls}}_{\mathrm{it}}+{\mathrm{\mu }}_{\mathrm{i}}+{\mathrm{\delta }}_{\mathrm{t}}+{\mathrm{\epsilon }}_{\mathrm{it}}$$

(1)

In Equation (1), HIT_it is the dependent variable, representing the breakthrough innovation of enterprise i in year t. AI_it is the independent variable, representing the AI application level of enterprise i in year t. α is the coefficient of the variables. Controls_it is a series of control variables. μ_i and δ_t are individual and time fixed effects, respectively. ԑ_it is a random error term. $\mathrm{\alpha }$ is the coefficient of the variables.

Based on the above main effect, to explore the mediating effect of knowledge recombination in the relationship between the AI application and enterprise breakthrough innovation, the mediating effect models are constructed as follows:

$${\mathrm{KRC}}_{\mathrm{it}}={\mathrm{\beta }}_{01}+{\mathrm{\beta }}_{11}{\mathrm{AI}}_{\mathrm{it}}+{\mathrm{\beta }}_{21}{\mathrm{Controls}}_{\mathrm{it}}+{\mathrm{\mu }}_{\mathrm{i}}+{\mathrm{\delta }}_{\mathrm{t}}+{\mathrm{\epsilon }}_{\mathrm{it}}$$

(2)

$${\mathrm{HIT}}_{\mathrm{it}}={\mathrm{\theta }}_{01}+{\mathrm{\theta }}_{11}{\mathrm{AI}}_{\mathrm{it}}+{\mathrm{\theta }}_{21}{\mathrm{KRC}}_{\mathrm{it}}+{\mathrm{\theta }}_{31}{\mathrm{Controls}}_{\mathrm{it}}+{\mathrm{\mu }}_{\mathrm{i}}+{\mathrm{\delta }}_{\mathrm{t}}+{\mathrm{\epsilon }}_{\mathrm{it}}$$

(3)

$${\mathrm{KRR}}_{\mathrm{it}}={\mathrm{\beta }}_{02}+{\mathrm{\beta }}_{12}{\mathrm{AI}}_{\mathrm{it}}+{\mathrm{\beta }}_{22}{\mathrm{Controls}}_{\mathrm{it}}+{\mathrm{\mu }}_{\mathrm{i}}+{\mathrm{\delta }}_{\mathrm{t}}+{\mathrm{\epsilon }}_{\mathrm{it}}$$

(4)

$${\mathrm{HIT}}_{\mathrm{it}}={\mathrm{\theta }}_{02}+{\mathrm{\theta }}_{12}{\mathrm{AI}}_{\mathrm{it}}+{\mathrm{\theta }}_{22}{\mathrm{KRR}}_{\mathrm{it}}+{\mathrm{\theta }}_{32}{\mathrm{Controls}}_{\mathrm{it}}+{\mathrm{\mu }}_{\mathrm{i}}+{\mathrm{\delta }}_{\mathrm{t}}+{\mathrm{\epsilon }}_{\mathrm{it}}$$

(5)

Equations (2) and (3) are used to test the mediating effect of knowledge recombination creation in the relationship between the AI application and enterprise breakthrough innovation, where KRC_it represents the knowledge recombination creation of enterprise i in year t. Equations (4) and (5) are used to test the mediating effect of knowledge recombination reuse in the relationship between the AI application and enterprise breakthrough innovation, where KRR_it represents the knowledge recombination reuse of enterprise i in year t. β and $\mathrm{\theta }$ are the coefficient of the variables.

To examine the moderating role of market competition in the relationship between knowledge recombination and enterprise breakthrough innovation, the following moderating effect models are constructed:

$${\mathrm{HIT}}_{\mathrm{it}}={\mathrm{\gamma }}_{01}+{\mathrm{\gamma }}_{11}{\mathrm{KRC}}_{\mathrm{it}}+{\mathrm{\gamma }}_{21}{\mathrm{Controls}}_{\mathrm{it}}+{\mathrm{\mu }}_{\mathrm{i}}+{\mathrm{\delta }}_{\mathrm{t}}+{\mathrm{\epsilon }}_{\mathrm{it}}$$

(6)

$${\mathrm{HIT}}_{\mathrm{it}}={\mathrm{\xi }}_{01}+{\mathrm{\xi }}_{11}{\mathrm{KRC}}_{\mathrm{it}}+{\mathrm{\xi }}_{21}\mathrm{PCM}+{\mathrm{\xi }}_{31}{\mathrm{KRC}}_{\mathrm{it}}\times {\mathrm{PCM}}_{\mathrm{it}}+{\mathrm{\xi }}_{41}{\mathrm{Controls}}_{\mathrm{it}}+{\mathrm{\mu }}_{\mathrm{i}}+{\mathrm{\delta }}_{\mathrm{t}}+{\mathrm{\epsilon }}_{\mathrm{it}}$$

(7)

$${\mathrm{HIT}}_{\mathrm{it}}={\mathrm{\gamma }}_{02}+{\mathrm{\gamma }}_{12}{\mathrm{KRR}}_{\mathrm{it}}+{\mathrm{\gamma }}_{22}{\mathrm{Controls}}_{\mathrm{it}}+{\mathrm{\mu }}_{\mathrm{i}}+{\mathrm{\delta }}_{\mathrm{t}}+{\mathrm{\epsilon }}_{\mathrm{it}}$$

(8)

$${\mathrm{HIT}}_{\mathrm{it}}={\mathrm{\xi }}_{02}+{\mathrm{\xi }}_{12}{\mathrm{KRR}}_{\mathrm{it}}+{\mathrm{\xi }}_{22}\mathrm{PCM}+{\mathrm{\xi }}_{32}{\mathrm{KRR}}_{\mathrm{it}}\times {\mathrm{PCM}}_{\mathrm{it}}+{\mathrm{\xi }}_{42}{\mathrm{Controls}}_{\mathrm{it}}+{\mathrm{\mu }}_{\mathrm{i}}+{\mathrm{\delta }}_{\mathrm{t}}+{\mathrm{\epsilon }}_{\mathrm{it}}$$

(9)

Equation (6) is used to test the impact of knowledge recombination creation on enterprise breakthrough innovation, and Equation (7) is used to test the moderating effect of market competition in this process, where PCM_it represents the market competition of enterprise i in year t. Equation (8) is used to test the impact of knowledge recombination reuse on enterprise breakthrough innovation, and Equation (9) is used to test the moderating effect of market competition in this process. γ and $\mathrm{\zeta }$ are the coefficient of the variables.

4. Empirical Results Analysis

4.1. Descriptive Statistics and Correlation Analysis

The descriptive statistics and correlation statistics of variables are shown in

Table 3. Correlation and descriptive analysis.

Variables	HIT	AI	KRC	KRR	PCM	Age	ROA	RGR	OC	BS
HIT	1.000
AI	0.090 ***	1.000
KRC	0.741 ***	0.049 ***	1.000
KRR	0.758 ***	0.028 ***	0.879 ***	1.000
PCM	0.031 ***	0.052 ***	0.014	0.007	1.000
Age	0.077 ***	−0.034 ***	0.105 ***	0.081 ***	0.006	1.000
ROA	−0.006	0.019 **	−0.011	−0.008	0.003	−0.019 **	1.000
RGR	0.008	−0.037 ***	0.021**	0.013	0.014	−0.034 ***	0.012	1.000
OC	−0.009	−0.152 ***	0.017 *	0.007	−0.083 ***	−0.119 ***	0.009	0.161 ***	1.000
BS	0.090 ***	−0.066 ***	0.078 ***	0.075 ***	−0.043 ***	0.122 ***	0.007	0.024 **	0.028 ***	1.000
Mean	31.499	2.009	106.861	19.464	0.876	18.617	0.46	0.031	58.557	8.353
Std. Dev.	166.314	1.163	358.770	108.818	0.120	5.862	5.707	0.136	15.235	1.763
Min	0.000	0.693	0.000	0.000	0.000	3.000	−7.655	−7.700	8.779	0.000
Max	3950.000	6.250	14,769.000	6156.000	0.968	64.000	434.593	1.408	100.970	18.000
VIF	—	1.05	4.68	4.62	1.01	1.11	1.00	1.03	1.07	1.11

Note: *** denotes p < 0.01, ** denotes p < 0.05, * denotes p < 0.1.

. Correlation analysis results reveal that AI application exhibits a significant positive correlation with both knowledge recombination and enterprise breakthrough innovation. This alignment with the theoretical logic underlying the research hypotheses provides preliminary empirical support for the proposed relationships, laying a solid foundation for subsequent in-depth analysis. In terms of multicollinearity testing, among all the variables included in the model, the maximum variance inflation factor (VIF) is 4.68, which is well below the commonly accepted threshold of 10. Meanwhile, most of the correlation coefficients between variables do not exceed 0.8, further confirming that there is no serious multicollinearity problem in the dataset. These results are crucial as they ensure that the regression coefficients obtained in the subsequent regression analysis will be stable and reliable, allowing for valid inferences to be drawn about the relationships between the variables.

4.2. Regression Analysis Results

This paper uses the Hausman test to determine the form of the panel data model. This test determines whether individual effects are correlated with explanatory variables to distinguish between fixed and random effects models. The results showed that the p-value was less than 0.01, rejecting the null hypothesis at the 1% significance level. This indicates the existence of significant individual heterogeneity, which is related to the explanatory variables in the model. At this point, the fixed effects model can effectively control unobserved individual characteristics that do not change over time, avoid estimation bias caused by omitted variables, and ensure consistency in parameter estimation. Therefore, this article ultimately chooses the fixed effects model for subsequent analysis. To ensure that the calculated data are both comparable and unified, we use the dimensionless normalization method to process all data. The double fixed effects model is shown in

Table 4. Results of regression analysis.

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
	HIT	KRC	KRR	HIT	HIT	HIT
AI	0.051 ** (2.239)	0.005 *** (2.656)	0.040 * (1.906)	0.019 (1.529)	0.026 ** (2.100)	0.019 (1.606)
KRC				0.595 *** (7.285)		0.416 *** (5.0243)
KRR					0.628 *** (6.65)	0.244 ** (2.0802)
Controls	Yes	Yes	Yes	Yes	Yes	Yes
Firm fixed	Yes	Yes	Yes	Yes	Yes	Yes
Year fixed	Yes	Yes	Yes	Yes	Yes	Yes
N	9322	9322	9322	9322	9322	9322
R2	0.054	0.068	0.039	0.446	0.401	0.462

Note: t-values of clustered standard errors are in parentheses, ***, ** and * denote significance levels of 1%, 5%, and 10%, respectively.

. It can be seen from Model 1 that the AI application has a significant positive impact on enterprise breakthrough innovation. Therefore, H1 is validated. Model 2 and Model 3 are used to test the impact of the independent variable on the mediating variables (knowledge recombination creation and knowledge recombination reuse). The results show that the AI application has a significant positive impact on its knowledge recombination creation and knowledge recombination reuse, that is, hypotheses H2a and H2b are verified. Model 4 and Model 5, respectively, include knowledge recombination creation, knowledge recombination ability, AI application, breakthrough innovation and control variables. The results show that the knowledge recombination creation has a significant positive impact on enterprise breakthrough innovation, and the enterprise knowledge recombination ability has a significant positive impact on its core technology breakthrough ability. Model 6 simultaneously includes breakthrough innovation, AI application, knowledge recombination creation, knowledge recombination reuse and control variables. The results show that both creative knowledge recombination and knowledge recombination for reuse play mediating roles between AI application and enterprise breakthrough innovation, thus verifying Hypotheses H3a and H3b.

Since there are defects in the determination of the existence of mediating effects by the stepwise regression method, this study uses the Bootstrap method to further test the mediating effects of knowledge recombination to enhance the completeness and credibility of the mechanism test. The number of iterations is set to 1000 times, and the confidence interval is 95%. The results are shown in

Table 5. Bootstrap mediation effects test—knowledge recombination creation.

Mediating Variable	Type of Effect	Coefficient	Standard Error	Confidence Interval
Knowledge Recombination Creation	Indirect Effect	0.012	0.002	[0.007, 0.017]
	Direct effect	0.013	0.002	[0.010, 0.017]
	Total effect	0.026	0.003	[0.019, 0.032]

and

Table 6. Bootstrap mediation effect test—knowledge recombination reuse.

Mediating Variable	Type of Effect	Coefficient	Standard Error	Confidence Interval
Knowledge Recombination Reuse	Indirect Effect	0.008	0.003	[0.003, 0.013]
	Direct effect	0.018	0.002	[0.013, 0.022]
	Total effect	0.026	0.003	[0.019, 0.032]

. Table 5 shows the Bootstrap test results of the mediating effect of knowledge recombination creation. It can be seen that the confidence intervals of the direct effect and the indirect effect do not contain 0, that is, both the direct effect and the indirect effect are significant. Knowledge recombination creation plays a partial mediating role in the impact of AI application on enterprise breakthrough innovation. Similarly, knowledge recombination reuse also plays a mediating role in the impact of AI application on enterprise breakthrough innovation.

To verify the moderating effect of market competition intensity in the relationship between knowledge recombination and enterprise breakthrough innovation, the interaction terms of knowledge recombination and market competition intensity after decentralization are added to the model for regression analysis. The results are shown in

Table 7. Moderating effects test.

	Model 1	Model 2	Model 3	Model 4
	HIT	HIT	HIT	HIT
KRC	0.596 *** (7.269)	−0.325 * (−1.688)
KRC×PCM		0.881 *** (3.388)
KRR			0.628 *** (6.639)	−0.336 * (−1.869)
KRR×PCM				0.869 *** (3.510)
Controls	Yes	Yes	Yes	Yes
Firm fixed	Yes	Yes	Yes	Yes
Year fixed	Yes	Yes	Yes	Yes
N	9303	9303	9303	9303
R2	0.446	0.460	0.401	0.418

Note: ***, and * denote significance levels of 1%, and 10%, respectively.

. From Model 2 in Table 7, the interaction term coefficient of knowledge recombination creation and market competition intensity is significantly positive, that is, market competition intensity can positively moderate the relationship between knowledge recombination creation and enterprise breakthrough innovation, and hypothesis H4a is verified. From Model 4 in Table 7, the interaction term coefficient of knowledge recombination reuse and market competition intensity is significantly positive, that is, market competition intensity can positively moderate the relationship between knowledge recombination reuse and enterprise breakthrough innovation, and hypothesis H4b is verified.

4.3. Robustness Test

4.3.1. Using Different Models

Due to the fact that enterprise breakthrough innovation is measured by the number of enterprise invention patents, and there are a large number of zero values in the patent number, showing the characteristics of censored data, referring to the research of Faleye et al. [86], the Tobit model is used to further test the relationship between the AI application and enterprise breakthrough innovation. The results are shown in

Table 8. Robustness test.

	Model 1	Model 2	Model 3	Model 4	Model 5
	HIT	HIT	HIT	HIT	HIT
AI	0.100 *** (8.865)	0.041 * (1.877)
L.AI			0.047 ** (2.438)
L2.AI				0.049 ** (2.44)
L3.AI					0.054 ** (2.417)
Controls	Yes	Yes	Yes	Yes	Yes
Firm fixed	Yes	Yes	Yes	Yes	Yes
Year fixed	Yes	Yes	Yes	Yes	Yes
N	9322	8454	6275	4857	3813
R2	0.055	0.058	0.055	0.064	0.060

Note: ***, ** and * denote significance levels of 1%, 5%, and 10%, respectively.

Model 1. The AI application obviously promotes enterprise breakthrough innovation, which is consistent with the results of the benchmark regression, indicating that the experimental results have a certain robustness.

4.3.2. Changing the Measurement Method of the Dependent Variable

Referring to the study of Liu et al. [87], the patents with new knowledge elements appearing in the t year compared with the previous five years of an enterprise are calculated, and the natural logarithm is taken after adding 1 to represent the enterprise breakthrough innovation. The knowledge elements are characterized by the data before the “/” in the IPC classification number. The results are shown in Table 8, Model 2. The impact of the AI application on enterprise breakthrough innovation is still significant.

4.3.3. Lagging the Independent Variable

Considering the possible time lag in AI application by enterprises, the variable AI application is lagged by 1–3 periods for testing. The results are shown in models 3, 4, and 5 of Table 8. The impact of the AI application of the enterprise on its breakthrough innovation is still significant.

4.4. Heterogeneity Analysis

4.4.1. Property Rights Heterogeneity

Compared with non-state-owned enterprises, state-owned enterprises generally have more sufficient resource reserves and higher-end talent support. Due to their special ownership nature, state-owned enterprises can often mobilize accumulated resources more quickly when implementing new technology applications and can stimulate technological vitality in a short time. To verify the impact of AI application on enterprise breakthrough innovation with different ownership natures, this study divides all samples into two groups: state-owned enterprises and non-state-owned enterprises. The regression results are shown in

Table 9. Heterogeneity analysis results.

	Model 1	Model 2	Model 3	Model 4	Model 5
	State-Owned	Non-State-Owned	The East	The Central	The West
AI	0.119 * (1.78)	0.050 ** (2.153)	0.055 ** (2.297)	−0.065 (−1.127)	0.004 (0.063)
Controls	Yes	Yes	Yes	Yes	Yes
Firm fixed	Yes	Yes	Yes	Yes	Yes
Year fixed	Yes	Yes	Yes	Yes	Yes
N	2034	7124	7751	959	597
R2	0.125	0.027	0.057	0.080	0.090

Note: ** and * denote significance levels of 5%, and 10%, respectively.

Models 1 and 2. The results show that in both types of enterprises with different ownerships, the research hypotheses in this study are established. However, the promoting effect of AI application on breakthrough innovation in state-owned enterprises is better than that in non-state-owned enterprises, which verifies the heterogeneity analysis of enterprise ownership.

4.4.2. Regional Heterogeneity

In fact, due to different resource endowments, there are differences in the development levels among different regions, and there may be large differences in innovation habits, AI application scales, etc., among various regions. In view of this, this study divides the regions where enterprises are located into eastern, central, and western regions to explore the changes in the impact of AI application on enterprise breakthrough innovation under the above differences. The regression results are shown in Table 9 Models 3, 4 and 5. The results indicate that for enterprises in the eastern region, the research hypotheses in this study are established. However, for enterprises in the central and western regions, the hypotheses are not significant. The possible reason is that the digital infrastructure construction in the eastern region is more perfect, the innovation level is higher, and emerging technologies are applied more quickly. Compared with enterprises in the eastern region, enterprises in the central and western regions apply AI technology later and have not yet been able to combine AI application with enterprise production and operation better, and thus cannot fully exert its vitality.

5. Discussion

This study analyzes the impact of AI application and enterprise breakthrough innovation by considering the mediating role of knowledge recombination and the moderating role of market competition intensity. Using a sample of unbalanced panel data stocks of A-share listed companies in China from 2011 to 2023, we proved seven hypotheses. The main conclusions of this study are reflected in the following aspects.

Firstly, AI application positively promotes enterprise sustainable breakthrough innovation in enterprises is supported. This result is consistent with recent academic and practitioner encouragement for enterprises to apply AI to increase enterprise innovation performance [88,89]. By enhancing the efficiency of knowledge acquisition, accelerating the iteration of R&D processes, and enabling more accurate identification of unmet market needs, AI not only optimizes incremental improvements but also fosters the radical shifts in technology and business models that define breakthrough innovation. Practitioners increasingly emphasize AI adoption as a strategic priority for staying competitive in rapidly evolving markets, with numerous case studies documenting how enterprises leveraging AI have successfully developed disruptive products, services, or processes. Our results thus reinforce these emerging insights, providing empirical evidence that AI application is not merely a tool for operational efficiency but a key driver of transformative innovation in enterprises, one that accelerates progress toward sustainability by enabling breakthroughs in sustainable business models.

Secondly, the role of knowledge recombination capabilities acting as mediators in the relationship between AI application and enterprise breakthrough innovation is also supported. Knowledge-related capabilities have been the focus of existing scholars in research on innovation. A growing number of scholars have argued that enterprises’ innovations come from the ability to apply the knowledge they have absorbed [90,91]. Consistent with the mainstream view, this study also confirms the mediating role of knowledge recombination in the process of AI application affecting enterprise sustainable breakthrough innovation, AI as a revolutionary technology, to bring changes to the organization is by no means a single, the embeddedness of this technology will affect the operation of the organization in many ways, in the process of the enterprise on the acquisition of knowledge, assimilation and utilization of the habits of the enterprise will also receive a certain impact. In the process of enterprises’ acquisition, assimilation, and utilization of knowledge, the habits of enterprises will also be affected, especially in promoting the formation of knowledge management habits oriented to sustainable development. It encourages enterprises to pay more attention to the collection and integration of knowledge about sustainable operations, thereby laying a solid knowledge foundation for achieving long-term sustainable development goals while promoting sustainable breakthrough innovation.

Thirdly, the moderating effect of market competition intensity on the relationship between knowledge recombination and enterprise sustainable breakthrough innovation is also confirmed. The external competitive environment is critical for enterprises’ innovation decisions: intense competition heightens pressure to maintain market position, prompting more aggressive innovation strategies that increasingly align with sustainable development goals. In such contexts, knowledge recombination capabilities become particularly vital—they enable enterprises to break existing business and technological boundaries by rapidly integrating scattered resources and knowledge. Moderately intense competition not only accelerates the speed of translating knowledge into tangible innovations but also enhances the effectiveness of such innovations by ensuring they are better aligned with market needs and competitive gaps. This not only accelerates the speed of translating knowledge into tangible innovations with sustainable value but also enhances the effectiveness of such sustainable innovations by ensuring they are better aligned with market needs for social responsibility and competitive gaps. Thus, market competition amplifies the role of knowledge recombination in driving transformative innovation with sustainable impact, turning competitive pressure into a catalyst for both market success and long-term innovation and sustainable development.

6. Main Research Conclusions and Implications

6.1. Main Research Conclusions

Based on data from Chinese A-share listed companies spanning 2011 to 2023, this study explores the relationship between AI application and enterprise breakthrough innovation, along with its underlying mechanisms and boundary conditions, providing practical basis for enterprises to achieve sustainable development. The key findings are as follows:

Firstly, AI application significantly promotes enterprise breakthrough innovation by efficiently processing massive heterogeneous data, driving business model innovation, and optimizing organizational management models, thereby laying the foundation for the novel knowledge, advanced technologies, and scientific decision-making required for sustainable breakthrough innovation.

Secondly, knowledge recombination plays a mediating role in this relationship: through knowledge recombination creation and reuse, external knowledge is transformed into resources needed for corporate innovation, helping enterprises deepen their understanding of knowledge, integrate new technologies and innovation models, and facilitate the implementation of sustainable breakthrough innovation. This process continuously injects innovative vitality into enterprises and is a key mechanism to ensure their sustainable development.

Finally, market competition intensity positively moderates the relationship between knowledge recombination and breakthrough innovation. Fierce competition prompts enterprises to accelerate the conversion of knowledge recombination into sustainable breakthrough innovation, expand the scope of knowledge recombination, and develop new technologies to form innovation barriers and enhance competitive advantages, and thereby help enterprises consolidate their development foundation in dynamic markets, achieving the improvement of sustainable development capabilities.

In addition, heterogeneity analysis reveals that AI application has a stronger promoting effect on sustainable breakthrough innovation in state-owned enterprises than in non-state-owned enterprises, as the former can more easily mobilize internal and external resources to integrate AI with operations. Additionally, the positive impact of AI application on breakthrough innovation is more pronounced in enterprises in eastern China compared to those in central and western regions, benefiting from earlier exposure to AI technology and higher maturity in application. This provides stronger technological support for breakthrough innovation and sustainable development of enterprises in the region.

In summary, AI empowers enterprise sustainable breakthrough innovation through knowledge recombination, a process driven by market competition and characterized by heterogeneity due to differences in enterprise attributes and regions.

6.2. Implications

6.2.1. Implications for Government

Based on the above research conclusions, in order to better promote enterprises to achieve sustainable innovation through AI application and knowledge recombination, this paper proposes the following policy recommendations for the government.

Firstly, establish a collaborative support system for AI application and knowledge recombination to consolidate the foundation for enterprises’ sustainable development. Set up special funds to support enterprises in introducing AI technologies to optimize knowledge management systems, with a focus on supporting projects of knowledge recombination creation (such as cross-domain knowledge integration platforms) and reuse (such as the construction of industry knowledge bases). Provide tax reductions or R&D subsidies to enterprises that achieve efficient knowledge recombination through AI, generate sustainable breakthrough innovation results, and align with the orientation of sustainable development.

Secondly, optimize the market competition environment to drive the momentum for enterprises’ sustainable development through moderate competition. Improve anti-monopoly and anti-unfair competition laws and regulations to create a market ecosystem of “moderate competition”. On the one hand, stimulate enterprises’ innovation motivation by breaking industry monopolies, prompting them to seize technological heights through knowledge recombination and cultivate sustainable competitive advantages; on the other hand, avoid excessive competition that leads enterprises into short-termism, and guide enterprises to allocate knowledge recombination resources to long-term sustainable breakthrough innovations, especially technological R&D in fields related to sustainable development such as ecological protection and resource conservation.

Finally, strengthen AI infrastructure and talent cultivation to solidify the support system for enterprises’ sustainable development. Accelerate the construction of new infrastructure such as 5G and computing power centers to reduce the technical threshold for enterprises to apply AI. Offer interdisciplinary programs of “Artificial Intelligence + Knowledge Management” in universities, integrating courses on sustainable development concepts and practices, to cultivate compound talents who understand AI technology, are proficient in industry knowledge, and have a sense of social responsibility. This will provide talent support for enterprises to carry out sustainable breakthrough innovation through knowledge recombination and achieve the coordinated improvement of economic and environmental benefits.

6.2.2. Implications for Enterprises

In view of the above research conclusions, in order to assist enterprises in enhancing sustainable breakthrough innovation capabilities and consolidating the foundation of sustainable development through artificial intelligence and knowledge recombination, this paper proposes the following management suggestions.

Firstly, proactively lay out innovation paths where artificial intelligence empowers knowledge recombination to consolidate the foundation for sustainable development. Invest resources in developing or introducing intelligent knowledge management tools, and utilize technologies such as natural language processing and machine learning to mine implicit knowledge from internal and external data, with particular attention to knowledge elements related to sustainable development such as green technologies and low-carbon processes, so as to improve the efficiency of knowledge recombination creation and its environmental protection orientation. Meanwhile, build an internal knowledge-sharing platform within the enterprise, and realize precise matching and reuse of knowledge through AI algorithms to avoid resource waste caused by repeated research and development, thus promoting the transformation of enterprises towards a resource-saving and environment-friendly development model.

Secondly, dynamically adjust knowledge recombination strategies according to market competition intensity, and balance competitiveness and sustainable development. In a fiercely competitive environment, focus on the “sustainable breakthrough direction” of knowledge recombination, explore new technical paths through knowledge integration across industries and technical fields, and give priority to investing recombination resources in fields such as clean energy and circular economy that have both market potential and sustainable value, so as to form differentiated competitive advantages. In periods of moderate competition, pay attention to the “in-depth accumulation” of knowledge recombination, consolidate core technical barriers, and reserve momentum for sustainable development for subsequent breakthroughs. At the same time, establish a market competition monitoring mechanism to adjust the investment ratio of knowledge recombination resources between short-term benefits and long-term sustainable goals in real time.

Finally, build an organizational culture that encourages knowledge recombination and sustainable innovation to activate long-term development momentum. Establish an innovation fault-tolerance mechanism to allow trial-and-error costs incurred in the process of knowledge recombination for exploring green technologies and sustainable business models, so as to stimulate employees’ enthusiasm for participating in cross-departmental knowledge collaboration. Through equity incentives, project dividends and other methods, link the sustainable development achievements generated by knowledge recombination with employee incentives, promote the in-depth integration of artificial intelligence technology with organizational innovation vitality and sustainable development concepts, and realize the coordinated growth of enterprises’ economic and environmental benefits.

6.3. Research Limitations

Although this study has achieved certain results in the study of the relationship between AI application, knowledge recombination, market competition intensity, and enterprise breakthrough innovation, there are still the following limitations.

Firstly, in terms of research methodology, this study mainly uses fixed effects regression analysis to verify the relationships between variables, which can effectively control confounding factors and identify overall effects. However, it is difficult to deeply reveal the nonlinear relationships, dynamic evolution processes, and multi-stage characteristics of causal chains between variables. Future research can combine methods such as structural equation modeling, panel threshold modeling, or dynamic network analysis to explore potential mechanisms from richer statistical and econometric tools, in order to improve the interpretability and robustness of conclusions.

Secondly, in terms of variable measurement, this study uses the number of invention patents granted by enterprises in the current year as an indicator for measuring enterprise breakthrough innovation. Although this method can reflect the sustainable breakthrough innovation capability of enterprises to a certain extent, it has not responded well to the technological value and market impact of patents, and there is still a certain deviation from the actual level of sustainable breakthrough innovation. Subsequent research can introduce indicators such as the number of patent citations and the duration of patent rights maintenance to reflect the influence and sustainability of patents, or combine semantic analysis-based patent quality evaluation methods to provide multidimensional characterization of sustainable breakthrough innovation, in order to obtain more comprehensive and accurate measurement results.

Finally, this study mainly uses data from Chinese A-share manufacturing listed companies. Although it can reflect the relationship between AI application and enterprise breakthrough innovation in the Chinese context, it may limit the cross-regional generalizability of the conclusions. Due to differences in institutional environments, market structures, and technological foundations across countries, the enabling effect of AI may exhibit heterogeneity. Future research can be extended to cross-country comparative studies to explore the similarities and differences in variable relationships under different institutional backgrounds. Meanwhile, non-listed enterprise samples can be included to test the applicability of the conclusions in a wider range of industries, thereby further enriching the external validity of the research.