Generative AI and Sustainable Performance in Manufacturing Firms: Roles of Innovations and AI Regulation

Abstract

This study scrutinizes the effects of generative artificial intelligence (GenAI) on sustainable performance (SP) in Chinese manufacturing firms through the mediating role of novelty-centered and efficiency-centered business model innovations (BMIs). It also explores the moderating effect of AI regulation on the GenAI–BMIs and GenAI–SP relationships. Data were collected from 1192 middle-level managers across 500 Chinese manufacturing firms using a two-wave survey design. Partial least squares structural equation modeling (PLS-SEM) was employed to test direct, mediating, and moderating relationships. The findings show that GenAI adoption has a significant positive effect on novelty-centered BMI, efficiency-centered BMI and sustainability performance. The GenAI–SP relationship is mediated by both BMIs, indicating that GenAI contributes to sustainability both directly and through innovative business practices. Moreover, AI regulation significantly strengthens the effects of GenAI on both BMI and SP, emphasizing the importance of regulatory alignment in maximizing technological benefits. This research shows that firms should emphasis AI tools and strategies to innovate their business model for better sustainable outcomes. Firms need to follow regulations and rules embedded into digitalization to ensure a sustainable competitive position in the market.

1. Introduction

Over recent decades, climate change has shifted from a peripheral concern to a critical global issue, disproportionately affecting vulnerable communities with limited adaptive capacity [1]. Grasping the depth of this reality, the United Nations created the Sustainable Development Goals (SDGs), which require radical and root-level efforts from all sectors to develop environmental integrity, social justice, and economic resilience. The manufacturing industry is pioneering in this global movement towards sustainability [2]. Although manufacturing is a source of employment and innovation, manufacturing also creates substantial environmental impacts through its energy requirements, generation of wastes, and use of natural resources [3,4]. Due to these factors, improving manufacturing sustainability is a necessity and a long-term value creation strategy [5].

Sustainable manufacturing performance is the triple alignment of environmental responsibility, social responsibility, and economic effectiveness [6]. These pillars represent a firm’s collective effort towards sustainable development [7]. Green manufacturing practices reduce environmental impacts, win customers’ trust, and adhere to more stringent environmental regulations [8,9]. Green manufacturing-oriented organizations are best positioned to adapt to evolving marketplace and customer demands. In so doing, they ensure their long-term prosperity and build a cleaner and healthier world for future generations [10]. However, despite its significance, research on effectively incorporating sustainability into manufacturing processes is still lacking [2]. Most manufacturing firms, especially in emerging and transition economies, are still far from mainstreaming sustainable practices into their operations [11]. The issues are, however, generally related to old production systems and a failure to possess the ability to evolve and incorporate new technologies properly.

With the constantly rising business challenges of the current era, technological innovation has emerged as a potent force for sustainable change, and Generative AI (GenAI) has been one of the most significant innovations for the manufacturing sector [12]. In contrast to traditional AI systems that merely work on available data, the special ability of GenAI to create original content, mimic intricate situations, and create novel solutions provides manufacturers with unrivaled opportunities to rethink processes, boost efficiency, and lead sustainable practices [13]. This transformative potential stems from the diversified applications of GenAI, ranging from maximizing resource allocation and making predictive maintenance possible to sustainable product design facilitation and supporting better decision-making through sophisticated data modeling [14].

With strategic application, GenAI can propel an organization towards sustainability [15]. GenAI can help optimize resource planning, enable predictive maintenance, support climate-friendly product design, and help in decision-making in a more informed way through advanced data modeling [16]. Such capabilities are highly adaptable to the higher environmental and operating efficiency goals, and thus GenAI is a catalytic driver of sustainable manufacturing development. However, unlocking the full potential of GenAI for sustainability entails navigating substantial implementation challenges, such as creating new technical capabilities, keeping pace with changing regulations, creating strong data governance, and coping with organizational change [17]. While academic and industry interest in GenAI has grown rapidly, most research has been focused on its technological potential rather than real applications to sustainability, particularly in manufacturing. More empirical research is necessary to understand how the uptake of GenAI translates into tangible sustainability benefits in real-world contexts.

In real-world business settings, two critical but overlooked dimensions shape this translation from potential to impact: business model innovation (BMI) and AI regulation. BMI helps firms reconceive how they generate and deliver value in a meaningful way that allows for the introduction of GenAI. BMI includes novelty-centered innovation, new concepts and products, and efficiency-centered innovation, improving current processes [18]. At the same time, how AI is regulated is highly significant. Positive regulations and clear guidelines can enable the broader use of GenAI by businesses, whereas binding or vague regulations can stifle innovation [12]. However, little is known about how these interact in practice, especially in sectors such as Chinese manufacturing, with environmental pressure coupled with the strong incentive for digitalization. Our study addresses these complexities by investigating direct and indirect pathways through which GenAI influences sustainable performance, while accounting for how regulations shape these relationships. We specifically investigate GenAI’s direct effects on sustainable performance and its indirect effects through innovation in a novelty-centered and efficiency-centered business model. We also examine how AI regulation moderates these relationships.

This study makes significant theoretical contributions in advancing the literature on digitalization and SDGs in three important ways. We extend RBV theory first by showing how Generative AI acts as a meta-resource that produces enduring competitive advantage through changing interactions with business model innovation. Unlike traditional resources, the worth of GenAI stems from its ability to support resource leverage (via efficiency-oriented BMI) and resource transformation (via novelty-oriented BMI) simultaneously, resolving theoretical ambiguities about digital resources in sustainability contexts. Second, we theorize BMI through the empirical testing and differentiation of the two paths. This illustrates how manufacturers should prioritize radical innovation (from novelty-driven BMI) or incremental improvement (through efficiency-driven BMI) depending on their sustainability objectives. Third, we bridge institutional theory with RBV by revealing how AI regulation operates as a multi-level boundary condition that shapes strategic technological resources’ deployment and value capture. This introduces a new “regulated resource advantage” perspective crucial for understanding technology adoption in constrained environments. Besides theoretical insights, our findings yield practice-relevant outcomes. We provide managers with a framework for moving beyond shallow GenAI adoption by emphasizing (1) strategic alignment of AI capabilities with appropriately compatible BMI types, and (2) regulatory intelligence that considers compliance as scaffolding innovation rather than constraint. Along with these practical and theoretical advances, our study positions itself as relevant to scholars examining digital transformation, transitions to sustainability, and technology policy, and also provides concrete guidance for practitioners confronting complex implementation contexts.

2. Literature Review

GenAI is leading the worldwide business and human practice revolution. Besides maximizing productivity, GenAI is also a powerful economic and social disruptor [19], with particular implications for sustainability and corporate responsibility in the emerging global marketplace [20]. This technology raises such fundamental questions about responsible deployment to serve social and environmental purposes, a crucial consideration of significant concern to manufacturing firms that are industrial pillars [21]. Manufacturers are increasingly applying GenAI to green their production processes, create green products, and apply data analytics to optimize the use of resources and reduce waste [13,22]. This strategic use showcases GenAI’s ability to drive innovation while also responding to urgent ecological issues.

GenAI has numerous potential uses in enhancing manufacturing sustainability performance. GenAI algorithms can be used to comb through vast datasets for their potential in ensuring maximum utilization of resources, minimizing waste, and achieving optimum energy efficiency. For example, AI systems can monitor real-time energy consumption and adjust production schedules to minimize total energy consumption [23]. AI-based interventions have exhibited considerable waste minimization. AI-powered waste reduction initiatives reported 12% less waste and 24% fewer plastic waste creation [24]. GenAI can perform predictive maintenance, reducing downtime and extending the lifespan of machinery. Predictive maintenance is the process through which every possible breakdown can be predicted to prevent costly breakdown, avoiding unnecessary wastage of resources [25]. GenAI can optimize supply chain operations such that resources are well utilized and wastage is curbed. It encompasses logistics optimization, avoidance of transport costs, and better inventory control [26]. Green AI-based business strategies are able to drive green process innovation and achieve improved environmental performance [27]. Their work asserted that AI can be capable of processing information in the purpose of seeking opportunities where the processes could be carried out in a sustainable manner, and AI can be applied in designing as well as implementing new sustainable processes.

Researchers have increasingly explored how AI can support sustainable development, particularly in areas like resource planning, energy efficiency, and circular economy practices in industrial settings [28,29]. Although prior studies have explored the broader effects of artificial intelligence on sustainability [30] and manufacturing [31], there remains limited insight into how the distinct capabilities of generative AI—such as advanced problem-solving and autonomous content creation—contribute to social and environmental outcomes within manufacturing firms. Addressing this gap is particularly relevant in light of Technology in Society’s recent emphasis on investigating the societal implications of emerging technologies like GenAI [20]. Current literature doesn’t dig deep enough into how GenAI transforms manufacturing processes or improves environmental performance. Some studies touch on AI’s role in sustainable operations management [32], but we need a sharper focus on how GenAI reshapes decision-making and organizational behavior in the push for greener practices.

Our research tackles these gaps by investigating GenAI’s specific role in boosting sustainability within manufacturing firms. We look at two key pathways—novelty-driven innovation (breakthrough solutions) and efficiency-driven innovation (optimizing existing processes)—to understand how GenAI drives sustainable performance. We also consider how AI regulations shape these outcomes, adding much-needed insights into responsible AI governance in industry.

Manufacturing companies are the backbone of world economies but are increasingly being challenged to provide productivity as well as sustainability. Existing research estimates GenAI to have the potential to “democratize” innovation so that manufacturers can access high-end capabilities [12]. But it is not yet apparent how production firms may strategically use this technology to advance both environmental and social agendas. We focus on Chinese manufacturers due to three reasons: Firstly, China is a leading country in manufacturing and AI innovation, and thus it is an ideal case study on how GenAI can influence industrial sustainability [33]. Secondly, the Chinese economy is manufacturing-based, and manufacturing businesses are spearheading digital transformation [34]. Their adoption of GenAI offers lessoning imperative to global industry. Third, China’s distinct regulatory landscape, characterized by robust state controls imposing green manufacturing and AI ethics [35], provides an opportunity to reflect on the intricate way in which regulatory policies intersect with technological breakthroughs in shaping innovation outcomes.

Borrowing from the RBV theory as our main theoretical framework, we frame GenAI capability as a strategic organizational asset that has the potential to create sustained competitive advantage. RBV theory argues that companies may attain superior performance through the utilization of unique, valuable, rare, and inimitable resources [36]. In manufacturing context, GenAI capability represents such a strategic resource that enables firms to enhance their sustainability performance through innovative management practices. The theory supports our model by explaining how GenAI capability, as a valuable organizational resource, can be deployed through novelty-centered and efficiency-centered innovation-based management to achieve superior sustainability outcomes. Furthermore, RBV acknowledges the role of external factors such as regulation in shaping how firms utilize their resources, providing theoretical justification for our inclusion of AI regulation as a moderating factor in the relationship between GenAI capability and innovation management practices.

Building on the Resource-Based View (RBV), we conceptualize GenAI as a strategic organizational resource that can generate sustained competitive advantage if it is valuable, rare, inimitable, and non-substitutable [36]. In the context of sustainable manufacturing, GenAI is highly valuable as it enables firms to optimize energy use, reduce waste, and innovate green products, thereby directly contributing to environmental, social, and economic sustainability [37]. The rarity of GenAI in this domain stems from its current limited adoption and the specialized technical expertise required for effective deployment, particularly among smaller manufacturers in emerging economies [38]. Its inimitability arises from firm-specific configurations of data infrastructure, human–AI collaboration practices, and integration with existing systems, making replication difficult even if competitors have access to similar technologies [39]. Furthermore, while narrow automation tools may be substitutable, GenAI’s capacity for autonomous problem-solving, content generation, and cross-functional innovation renders it non-substitutable for driving holistic business model transformation toward sustainability [40]. Thus, GenAI satisfies the VRIN criteria and functions as a meta-resource capable of enabling higher-order capabilities such as novelty- and efficiency-centered BMIs. According to RBV, possessing such a resource provides the foundation for superior performance—but only when it is effectively orchestrated within the organization. This aligns with our model, which posits that GenAI enhances sustainable performance not only directly but also through its role in enabling strategic BMI. Furthermore, RBV recognizes that external institutional conditions—such as regulatory frameworks—can influence how resources are leveraged, supporting our inclusion of AI regulation as a moderating force that shapes the value creation process.

Research Gaps and Theoretical Contributions

While growing research acknowledges the potential of artificial intelligence in advancing sustainability, significant conceptual and empirical gaps remain. First, most studies treat AI as a monolithic technology, failing to distinguish between traditional machine learning systems and the emerging class of GenAI tools—such as large language models and generative neural networks—that possess unique capabilities in content creation, simulation, and autonomous problem-solving [13,41]. Unlike rule-based automation, GenAI enables firms to explore novel business models and simulate sustainable futures, yet its strategic implications for organizational innovation remain undertheorized.

Second, there is an ongoing debate about whether digital technologies drive sustainability through direct operational improvements (e.g., predictive maintenance reducing energy use) or indirect strategic reconfiguration via BMI. Some scholars argue that efficiency gains alone are sufficient [23], while others emphasize that lasting impact requires systemic change in value creation logic [18,42]. Our study contributes to this debate by testing both pathways and showing that BMI serves as a critical mediating mechanism.

Third, existing work often overlooks the role of institutional context—particularly regulatory frameworks—in shaping how firms deploy AI for sustainability. While some view regulation as a barrier to innovation [43], others frame it as an enabler of trust and responsible adoption [12]. This tension reflects a broader theoretical divide between instrumental views of regulation (as constraint) and institutional views (as legitimacy provider).

Finally, despite increasing interest in digital sustainability, few studies offer empirical evidence from manufacturing firms in emerging economies—where digital transformation intersects with urgent environmental challenges and evolving policy landscapes [2]. Most research focuses on Western or tech-centric sectors, limiting generalizability.

Our study directly addresses these gaps by (1) focusing specifically on GenAI, not generic AI; (2) examining dual innovation pathways (novelty vs. efficiency-centered BMI) as mediators; (3) integrating AI regulation as a moderating force within an RBV-institutional framework; and (4) providing empirical insights from Chinese manufacturing firms, a high-relevance context for global industry. By doing so, we advance understanding of how digital resources translate into sustainable performance in complex, real-world settings.

3. Hypothesis Development

3.1. GenAI and Sustainable Performance

GenAI presents a multifaceted potential to enhance sustainability performance across its social, environmental, and economic dimensions [44]. GenAI can optimize processes, foster innovative solutions, and discern patterns within extensive datasets pertaining to sustainability, paving the way for more efficient resource allocation, waste reduction, and the creation of circular economy models [45]. From a social perspective, GenAI can simulate societal systems, thereby improving access to information, promoting inclusivity, and ensuring equitable resource distribution [25]. Environmentally, AI adoption can lead to reductions in climate change, improvements in agriculture, enhancements in ocean health and water resource management, as well as more accurate weather forecasting and disaster resilience [46,47]. Economically, AI-driven automation and innovation can drive cost savings, boost productivity, and stimulate economic growth, while green knowledge management and human capital development, enhanced by AI, can further contribute to sustainable performance [48]. Thus, integration of GenAI not only optimizes existing systems but also fosters the development of novel, sustainable solutions, making it a crucial tool for achieving comprehensive sustainability goals. Therefore, we posit that:

H1. 

GenAI exerts a significant positive effect on firm sustainable performance.

3.2. GenAI and Novelty-Centered BMI

GenAI holds promise for enhancing novelty-centered BMI by enabling the exploration of a wider solution space and the generation of novel ideas [49]. Their study contains that GenAI’s content generation capabilities can assist human teams in overcoming cognitive fixedness, which is a psychological constraint that hinders the ability to generate new ideas. By automating the initial stages of idea generation, GenAI can free up human resources to focus on higher-level strategic thinking, experimentation, and refinement of business models. The adoption of AI can significantly enhance organizational innovation novelty through human–AI collaboration [50]. Furthermore, cognitive analytics enabled by AI can generate deeper insights from complex data sources, which can inform the development of novel business models in various sectors, including healthcare [51]. Companies pursuing GenAI innovation may show lower top management team turnover [52]. Therefore, the capacity of GenAI to facilitate idea generation, innovation in business model, augment human creativity, and derive insights from data supports the generation of the hypothesis that

H2. 

GenAI exerts a significant positive effect on novelty-centered BMI.

3.3. GenAI and Efficiency-Centered BMI

Business use of GenAI is reconfiguring traditional business models in terms of enhanced operations and savings [15]. Efficiency-oriented BMI tries to enhance existing operations, save resources, remove wastage, and increase value delivery through engineered processes [18]. GenAI technologies, with their ability to automate content creation, predictive analysis, process optimization, and intelligent decision-making, feed directly into these objectives of efficiency. For instance, through optimizing customer care interaction or developing data-driven forecasts, GenAI minimizes human effort, decreases response time, and improves precision, all of which are all typical of efficiency-gain innovation [53]. In addition, GenAI enables companies to redefine internal processes dynamically in response to shifting markets at a minimal cost, thereby enabling lean and agile business models [54]. Merging GenAI with Business Intelligence (BI) has the potential to create Generative BI, which enables real-time data analysis and informed decision-making, optimizing business functions and increasing efficiency [55]. The logic follows the RBV, which posits that competitive advantage stems from rare, valuable, and inimitable resources, characteristics that GenAI systems increasingly embody due to their learning capabilities and adaptability [12]. It is therefore theoretically and empirically possible to postulate that:

H3. 

GenAI exerts a significant positive effect on efficiency-centered BMI.

3.4. Novelty-Centered BMI and Firm Sustainability Performance

Novelty-centered BMI involves the development of entirely new ways of creating, delivering, and seizing value through reconfiguring customer engagement, value propositions, or revenue mechanisms [18]. Novelty-centered BMI can potentially enhance the sustainability of the firm in social, environmental, as well as economic dimensions by generating new value propositions and competitiveness [56]. Their study highlights that original business models, characterized by their novelty, can lead to significant impacts on both enterprises and societal development. By creating new value creation, delivery, and capture solutions, organizations are able to unlock new market potential and attract consumers interested in sustainability [57]. This will enhance the economic performance of the company while simultaneously addressing social and environmental problems. Additionally, business model innovation can result in the embracement of environmentally friendly business practices like efficiency in the use of resources, minimized wastage, and sustainable sourcing that can help a company lower its carbon footprint and gain improvement in its image [42]. By pursuing innovation, integrating sustainability aspects into a firm’s business model can also create win–win outcomes for stakeholders, such as enhanced sustainability performance and enhanced technological competence [58]. Moreover, from an RBV, novel business models become intangible strategic resources—difficult to imitate and uniquely valuable in fast-changing environments [18]. Therefore, the following hypothesis is reasonable:

H4. 

Novelty-centered BMI exerts a positive influence on firm sustainable performance.

3.5. Efficiency-Centered BMI and Firm Sustainability Performance

Efficiency-based BMI helps achieve company sustainability in its economic, environmental, and social dimensions through optimum utilization of resources and avoidance of wastage with process improvement [57]. Efficiency-based emphasis helps firms rein in wastage, reduce costs, and enhance productivity, which supports economic sustainability [42]. Environmentally, efficiency-driven innovations generate less energy use, lower emissions, and a reduced environmental footprint, paving the way for environmental sustainability [59]. Socially, efficient business models are able to generate stakeholder value through the provision of low-cost and widely available products or services, improved working conditions, and even promoting ethical behavior, thereby contributing to social sustainability [58]. Further, their study highlights that digital technology leveraging and innovative practices, driven by pursuit of efficiency, are able to enhance technological performance and develop win–win stakeholder relationships, further boosting sustainability performance. Rooted in the RBV, efficiency improvements enhance a firm’s internal capabilities by leveraging valuable, rare, and inimitable operational routines that strengthen the sustainability of its competitive advantage. Accordingly, an emphasis on efficiency in BMI supports the hypothesis that

H5. 

Efficiency-centered BMI has a positive impact on firm sustainable performance.

3.6. Mediation Effects

Business model innovation, comprising both novelty-driven and efficiency-driven elements, serves as a critical mediating process through which the capability of GenAI impacts a firm’s sustainable performance. This mediating role is outlined within the RBV, which enables firms to achieve a sustainable competitive advantage not only by owning valuable resources but also by recombining them into strategic capabilities [36]. GenAI is a scarce, valuable, and increasingly irreplaceable technological resource [13]. It enables firms to seek new value propositions and leverage accessible operating efficiencies simultaneously—two core activities of strategic renewal.

Novelty-based BMI embraces a firm’s ability to innovate the way it produces, distributes, and captures value—such as using AI-enabled product-as-a-service propositions, co-designing circular economy products with others, or adopting generative design processes for sustainable products [18]. GenAI facilitates this function of discovery by generating new ideas, simulating customers’ demand under sustainability challenges, and reducing prototyping cycles [49]. In contrast, efficiency-centered BMI focuses on refining internal processes to minimize waste, reduce costs, and improve resource utilization—outcomes directly enabled by GenAI’s capacity for predictive maintenance, real-time energy optimization, and intelligent decision-making [23].

This two-step strategy aligns with dynamic capabilities theory [60], where GenAI serves as the “sensing” and “seizing” stage, and BMI represents the “transforming” stage—the redesign of organizational routines into sustainable business models. As Teece [61] argues, business model adjustment and transformation constitute high-order dynamic capabilities that facilitate the recombination and realignment of firms’ lower-order processes and resources in response to environmental change. Empirical data support this relationship: Torrent-Sellens, Enache-Zegheru and Ficapal-Cusí [48] demonstrate that digital transformation enhances sustainability only when complemented by innovation in value creation logic, whereas Khan et al. [62] show that the adoption of GenAI achieves improved performance yields primarily when accompanied by structural realignment. Similarly, Najafi-Tavani, Zantidou, Leonidou and Zeriti [18] illustrate how business model innovation enables firms to reconfigure resources and capture export value in constantly evolving contexts, thereby cementing their status as a key driver of strategic renewal. Furthermore, Ghobakhloo, Fathi, Iranmanesh, Vilkas, Grybauskas and Amran [32] argue that GenAI facilitates Industry 5.0 ambitions through not only task automation but also systemic change towards human-cantered and sustainable production—highlighting the need for organizational reconfiguration beyond technological adoption. This viewpoint concurs with Li et al. [63] resource-based definition of BMI, which places new business models in the context of outcomes of bundles of valuable resources being strategically recombined to change value creation and delivery. Therefore, BMI is not just one possible outcome of GenAI application but a theoretically necessary mediator because it converts distant technological achievements into systemic, scalable, and strategically oriented sustainability outcomes. As such, it is appropriate to posit that:

H6. 

Novelty-centered and efficiency-centered BMI mediate the relationship between GenAI capability and firm sustainable performance.

3.7. Moderation Effects of AI Regulation

AI regulation can moderate the relationship between Generative AI (GenAI) capability and novelty-centered business model innovation (BMI), efficiency-centered BMI, and firm sustainability performance by affecting the evolution and application of GenAI technologies and business models. Policy guidelines on responsible AI practices, data privacy, and algorithmic explainability can result in higher trust and adoption of GenAI-enabled technology, forcing entrepreneurs to explore new uses and business models that align with social agendas and sustainability [51]. Conversely, excessive regulation or unclear guidelines can stifle innovation by imposing additional compliance costs, limiting experimentation, and generating uncertainty about the legal and ethical boundaries of GenAI use [43]. For efficiency-centered BMI, AI policy can govern GenAI used in minimizing wastage, optimal utilization of resources, and automation of processes in eco-friendly and economically sustainable manners [12]. Also, AI regulation can determine how environmental performance is affected by AI [64]. Their findings indicate that by setting standards for data quality, algorithmic fairness, and environmental impact assessment, regulators can enable companies to develop and implement GenAI solutions that enhance sustainability without increasing existing disparities or creating new environmental risks. Invoking contingency theory, where the effectiveness of a capability depends on congruence with external contextual variables, AI regulation serves as a boundary condition. Therefore, the effectiveness of AI regulation in the promotion of these relations depends on its ability to manage innovation promotion against preventing related risk such that GenAI functionality is applied in promoting novelty and efficiency in business models and improving firm sustainability. We propose here that:

H7. 

AI regulation positively moderates the relationship between GenAI capability and (a) novelty-centered BMI, (b) efficiency-centered BMI, and (c) firm sustainable performance.

Figure 1 indicates the proposed model of this study.

4. Methodology

4.1. Data Collection

To empirically test the proposed model, data were collected from manufacturing firms located in key industrial regions of China, including Shenzhen, Guangzhou, Foshan, Dongguan, and Zhuhai. These cities were selected due to their strong presence in advanced manufacturing and digital innovation [65], which made them particularly relevant for examining GenAI capability, business model innovation, and sustainability performance. The sample encompasses a range of manufacturing subsectors, including electronics and electrical equipment, machinery, automotive parts, chemicals, consumer goods, textiles, food and beverage—reflecting the industrial diversity of China’s Pearl River Delta region.

We made sure to clearly explain the study’s purpose to each company’s HR department. Participation was completely optional, and we emphasized that all data would only be used for academic research. We guaranteed respondents full anonymity—no individual or company would be identified in any way. Informed consent was obtained from over 500 manufacturing firms. We distributed approximately 2000 questionnaires. Middle managers were selected as the primary respondents. They are uniquely positioned within the organizational hierarchy—bridging strategic decisions from top management and operational realities faced by front-line staff [66]. This position allows them to offer comprehensive insights into organizational capabilities, particularly in areas such as sustainability practices, GenAI implementation, and innovation strategies [2]. Moreover, using multiple respondents from the same firm enabled a deeper and more accurate understanding of firm-level dynamics and helped mitigate the risk of individual-level biases [67].

The survey was conducted in two waves to reduce common method variance. In the first wave, respondents provided information on GenAI capabilities, AI regulation, and demographics. Those who completed the first survey were invited through their email which was obtained in the first wave, to participate in the second wave, which captured data on business model innovation and sustainable performance. Out of the distributed questionnaires, 1540 responses were received in the first wave, and 1192 matched responses were collected in the second wave. The final usable sample consisted of 1192 responses, resulting in an effective response rate of approximately 59.6%.

4.2. Measures

For this study, we used established measurement scales that we adapted to fit our specific research context. Participants provided their responses using a standard 7-point rating system, where 1 meant “Strongly Disagree” and 7 meant “Strongly Agree.” The survey began with basic background questions about company characteristics like size and export performance. Sustainable performance (SP) was measured using a multi-dimensional scale adapted from [2,68], covering environmental, social, and economic dimensions. Environmental SP items focused on tangible outcomes such as waste reduction and emission control (e.g., “Our organization actively reduces waste and emissions through innovation”). One item mentions GenAI (“GenAI applications support environmentally sustainable decisions”), as it aims to capture managers’ perceptions of how technology contributes to sustainability—an approach consistent with prior studies on digital transformation [48]. Linking the outcome (sustainability) directly to the driver (GenAI) in this item could introduce some overlap between constructs—a concern known as criterion contamination. However, the majority of SP items are outcome-focused and do not reference GenAI”. The Cronbach’s alpha for EP was 0.831, indicating strong internal consistency. Social sustainable performance was measured with three items, such as “We actively support community development” and “Our employees are treated fairly and ethically,” with a Cronbach’s alpha of 0.808. Economic sustainable performance was assessed with four items; sample statements include “Our organization develops new products, services, or programs that enhance competitiveness” and “GenAI-driven solutions have contributed to cost savings or efficiency improvements,” showing a Cronbach’s alpha of 0.857. GenAI capability was measured using nine perception-based items and AI regulations items were adapted from [12,62]. These items reflect a firm’s strategic orientation toward GenAI, including the existence of a defined implementation roadmap, active exploration of innovation opportunities, cross-functional integration plans, and leadership commitment to AI adoption. Example items include: “Our Company has a well-defined strategy for implementing GenAI across operations” and “We consistently explore ways to use GenAI for innovation.” While these items capture perceived strategic preparedness and organizational intent rather than granular technical metrics (e.g., model sophistication or infrastructure), they align with recent organizational studies that conceptualize digital capability not solely as technological assets but as an integrated socio-technical capacity encompassing vision, alignment, and exploratory action [12,62]. Thus, our measure reflects strategic GenAI capability—a higher-order construct indicating organizational readiness and intentional investment in leveraging GenAI for competitive advantage, particularly relevant in the early-to-mid stages of adoption observed in Chinese manufacturing firms.” This scale showed excellent reliability (α = 0.971). For AI regulation, we assessed four statements including the following: “We must follow legal requirements for AI use in our operations,” and “There are clear regulatory boundaries for how we can apply AI technologies.”

These items demonstrated strong consistency (α = 0.92). Novelty-centered and efficiency-centered BMI were adapted from [18]. These constructs were measured with ten indicators such as “Our GenAI-based business model introduces new combinations of products, services, and information aimed at improving sustainability outcomes. And our business model leverages GenAI to establish new partnerships (e.g., with customers, suppliers, or financiers) for sustainable operations” and eleven items such as “Our GenAI-supported business model simplifies sustainable transactions from the customer’s perspective, and our GenAI-based model reduces additional costs (e.g., marketing, communication, transaction processing) for our stakeholders”, respectively, with a Cronbach’s alpha of 0.932 and 0.947. All Cronbach’s alpha values exceed the 0.70 threshold, confirming the reliability of the constructs, and average variance extracted (AVE) values exceeded 0.50 for all, ensuring convergent validity.

We validated our survey through a rigorous multi-step process. First, five experts in corporate sustainability and digital transformation reviewed the questionnaire to assess its clarity and appropriateness for our research context. Based on their recommendations, we refined the wording of several items to improve comprehension. Before launching the full study, we pilot-tested the instrument on 40 respondents to try out the reliability of the instrument. The results were excellent—all the constructs measured were found to have high internal consistency with Cronbach’s alpha values well over 0.70, and all items had high factor loadings of over 0.60. These results confirm that our survey instrument is not only psychometrically sound but also suitable for large-scale data collection.

4.3. Control Variables

Firm age, size, export orientation, and type of ownership are control variables in the research. Organizational maturity, which may affect decision-making and performance, is reflected through firm age. Firm size is traditionally associated with access to resources, complexity of operations, and market power, all of which may have implications for strategic outcomes. Export firms would most probably be subject to a variety of external pressures and institutional forces, e.g., international standards. Ownership structure, private or state, would most likely be subject to differential systems of governance, incentives, and regulatory systems that would shape the behavior of firms. Holding these constant enables us to isolate the direct effects of the research model.

5. Results

Data were statistically tested through Partial Least Squares Structural Equation Modeling (PLS-SEM) with the support of SmartPLS 4.0 because it is easy in predictive research models and very complicated interactions of several latent constructs [69]. Measurement model testing and structural model testing were part of the analysis in two phases. We also performed screening tests, post hoc examination, and common method bias (CMB) issues in the data to ensure the validity of the findings.

5.1. Common Method Bias (CMB)

To mitigate the probable issue of CMB, statistical controls and procedural controls were used [12]. Procedurally, a two-wave time-lagged survey design was applied in an attempt to limit the ability of the respondents to use the same cognitive schemata to all answers. Statistically, Harman’s single-factor test was conducted with exploratory factor analysis to ascertain the existence of CMB. In the unrotated solution, the variance explained by the first factor was 35.7%, which is less than the conservatively chosen cut-off of 50%, indicating common method variance is not a concern for this research [70]. These measures enhance the validity of our findings and establish the stability of the measurement model.

5.2. Measurement Model

To determine the robustness of the measurement model, we examined the reliability and convergent validity of all constructs. From

Table 1. Measurement model.

Construct	Items	FL	VIF	α	Cr	AVE
AI Regulation	AIR1	0.953	3.597	0.92	0.933	0.776
	AIR2	0.846	3.107
	AIR3	0.817	2.953
	AIR4	0.903	2.614
Efficiency-Centered IBM	EBMI1	0.757	2.023	0.947	0.954	0.653
	EBMI10	0.833	2.682
	EBMI11	0.765	2.064
	EBMI2	0.824	2.61
	EBMI3	0.8	2.34
	EBMI4	0.826	2.624
	EBMI5	0.857	3.057
	EBMI6	0.79	2.261
	EBMI7	0.813	2.442
	EBMI8	0.841	2.801
	EBMI9	0.774	2.13
GenAI	GnAI1	0.894	3.994	0.971	0.975	0.812
	GnAI2	0.906	4.489
	GnAI3	0.917	4.989
	GnAI4	0.866	3.237
	GnAI5	0.92	4.15
	GnAI6	0.922	3.256
	GnAI7	0.896	4.09
	GnAI8	0.864	3.235
	GnAI9	0.924	3.421
Novelty-centered IBM	NBMI1	0.776	2.073	0.932	0.943	0.623
	NBMI10	0.726	1.802
	NBMI2	0.777	2.113
	NBMI3	0.797	2.233
	NBMI4	0.785	2.172
	NBMI5	0.824	2.504
	NBMI6	0.781	2.13
	NBMI7	0.838	2.683
	NBMI8	0.724	1.799
	NBMI9	0.851	2.864
Sustainable Performance	SPr1	0.811	2.391	0.938	0.947	0.642
	SPr10	0.819	2.463
	SPr2	0.756	1.964
	SPr3	0.806	2.338
	SPr4	0.841	2.725
	SPr5	0.859	2.992
	SPr6	0.738	1.878
	SPr7	0.758	1.977
	SPr8	0.797	2.266
	SPr9	0.816	2.44

, it can be observed that internal consistency for all constructs is satisfactory, with Cronbach’s alpha (α) and CR all above 0.70 [71]. Further, convergent validity was also tested through AVE where all the value surpass the minimum criterion of 0.50, confirming that a substantial explained by indicators for each construct and not error [71]. Also, all the factor loadings were highly significant and above 0.70, confirming measures’ reliability. Figure 2 illustrates the standardized factor loadings and p-values graphically confirming high correlation between observed indicators and latent constructs. Overall, these results lend strong support for the convergent validity and reliability of the measurement model.

Discriminant validity was assessed using both the Fornell and Larcker [72] criterion and the HTMT ratio of correlations. As presented in

Table 2. Discriminant validity.

Construct	1	2	3	4	5
1. AI Regulation	0.881
2. Efficiency-Centered BMI	0.023	0.808
3. GenAI	0.025	0.524	0.901
4. Novelty-Centered IBM	0.008	0.42	0.615	0.789
5. Sustainable Performance	0.026	0.61	0.691	0.606	0.801
HTMT
1. AI Regulation
2. Efficiency-Centered BMI	0.022
3. GenAI	0.024	0.546
4. Novelty-Centered IBM	0.024	0.447	0.645
5. Sustainable Performance	0.026	0.645	0.723	0.646

, the square root of the AVE for each construct was greater than its correlations with other constructs, satisfying the criterion of Fornell and Larcker [72]. Additionally, HTMT values for all construct pairs were below the conservative threshold of 0.9 [73], further confirming adequate discriminant validity. These results affirm that each construct in the model is conceptually and statistically distinct, indicating the robustness of the measurement model.

5.3. Model Fit

The model fit indices (

Table 3. Model fit.

	Estimated Model
SRMR	0.026
d_ULS	0.778
d_G	0.241
Chi-square	1645.101
NFI	0.962

) indicate an excellent overall fit for the estimated theoretical model. The SRMR (Standardized Root Mean Square Residual) value of 0.026 is well below the recommended threshold of 0.08, indicating a close fit between the observed and predicted correlation matrices [74]. The NFI (Normed Fit Index) of 0.962 exceeds the commonly accepted cutoff of 0.95, suggesting that the model fits significantly better than the baseline independence model [72].

5.4. Structural Model

After verifying the measurement model, we proceeded with hypothesis testing to examine the structural relationships of the research model. The findings in

Table 4. Hypothesis results.

Path	Direct Effects	Moderating Effects	Indirect Effects	f2	Supported
Control Effects
Firm Age → sustainable performance	0.008 (0.454)	—	—	0.001	-
Firm size → sustainable performance	0.034 (0.089)	—	—	0.003	-
Export oriented → sustainable performance	0.004 (0.535)	—	—	0.001	-
Ownership type → sustainable performance	0.002 (0.93)	—	—	0.001	-
Main Effects
GenAI → Sustainable performance	0.424 (0.000)	0.131 (0.000)	—	0.232	Yes
GenAI → Novelty-centered BMI	0.625 (0.000)	0.213 (0.000)	—	0.673	Yes
Novelty-centered BMI → Sustainable performance	0.206 (0.000)	—	—	0.063	Yes
Efficiency-centered BMI → Sustainable performance	0.280 (0.000)	—	—	0.138	Yes
GenAI → Efficiency-centered BMI	0.532 (0.000)	0.196 (0.000)	—	0.410	Yes
GenAI → Novelty-centered BMI → sustainable performance	—	—	0.155 (0.000)		Yes
GenAI → Efficiency-centered BMI → sustainable performance	—	—	0.231 (0.000)		Yes

indicate that GenAI significantly contribute to the sustainable performance (β = 0.424, t = 16.106, p < 0.001), novelty-centered BMI (β = 0.625, t = 36.649, p < 0.001) and efficiency-centered BMI (β = 0.532, t = 28.204, p < 0.001). Hence, H1, H2 and H3 are supported. Furthermore, the results show that novelty-centered (β = 0.206, t = 7.797, p < 0.001), and efficiency-centered BMI (β = 0.280, t = 11.318 p < 0.001) significantly contribute to sustainable performance. Thus, H4 and H5 are supported. Novelty-centered (β = 0.155, t = 8.369, p < 0.001) and efficiency-centered BMI (β = 0.231, t = 13.254, p < 0.001), significantly mediate in the relationship between GenAI and Sustainable performance. Hence, H6 is accepted. The moderating effects of AI regulation on the relationship between GenAI and novelty-centered BMI (β = 0.213, t = 5.577, p < 0.001), GenAI and efficiency-centered BMI (β = 0.196, t = 5.146, p < 0.001) and GenAI and sustainable performance (β = 0.131, t = 4.938, p < 0.001) are statistically significant. So, hypothesis is H7 accepted. Insignificant effects of all controls were found. As signaled by the R² value, the present study accounts for 31.1% of the variance in efficiency-centered BMI, 42.2% in novelty-centered BMI and 0.61.4% in sustainable performance. The Q² values of 0.201, 0.260 and 0.390 efficiency-centered BMI, in novelty-centered BMI and in sustainable performance, respectively, represents good predictive power of the model.

5.5. Post Hoc Analysis

In addition to the first-order construct, we assessed the influence of the research model on second-order constructs; environmental sustainable performance and social sustainable performance and economic sustainable performance. The analysis showed in

Table 5. Post hoc analysis.

Path	β	T Value	p Value
Efficiency-Centered BMI → Economic SP	0.272	10.779	0.000
Efficiency-Centered BMI → Environmental SP	0.27	10.53	0.000
Efficiency-Centered BMI → Social SP	0.251	9.593	0.000
GenAI → Economic SP	0.401	14.872	0.000
GenAI → Efficiency-Centered BMI	0.532	28.036	0.000
GenAI → Environmental SP	0.389	13.929	0.000
GenAI → Novelty-Centered IBM	0.625	36.818	0.000
GenAI → Social SP	0.41	14.192	0.000
Novelty-Centered IBM → Economic SP	0.191	6.867	0.000
Novelty-Centered IBM → Environmental SP	0.215	7.757	0.000
Novelty-Centered IBM → Social SP	0.174	6.014	0.000
AIR × GenAI → Economic SP	0.128	4.67	0.000
AIR × GenAI → Efficiency-Centered BMI	0.193	5.054	0.000
AIR × GenAI → Environmental SP	0.123	4.563	0.000
AIR × GenAI → Novelty-Centered IBM	0.212	5.452	0.000
AIR × GenAI → Social SP	0.122	4.286	0.000
Efficiency-Centered BMI → Economic SP	0.272	10.779	0.000
GenAI → Novelty-Centered IBM → Environmental SP	0.134	7.701	0.000
GenAI → Efficiency-Centered BMI → Economic SP	0.145	9.934	0.000
GenAI → Efficiency-Centered BMI → Environmental SP	0.144	9.746	0.000
GenAI → Novelty-Centered IBM → Social SP	0.109	6.018	0.000
GenAI → Efficiency-Centered BMI → Social SP	0.134	8.898	0.000
GenAI → Novelty-Centered IBM → Economic SP	0.12	6.868	0.000

revealed significant positive relationships between GenAI and environmental sustainable performance (β = 0.389, t = 13.929, p < 0.001), social sustainable performance (β = 0.41, t = 14.192, p < 0.001), economic sustainable performance (β = 0.128, t = 4.67, p < 0.001), novelty-centered BMI (β = 0.626, t = 36.818, p < 0.001), and efficiency-centered BMI (β = 0.532, t = 28.036, p < 0.001). Efficiency-centered significantly affects economic sustainable performance (β = 0.272, t = 10.779, p < 0.001), environmental sustainable performance (β = 0.27, t = 10.53, p < 0.001) and social sustainable performance (β = 0.251, t = 9.593, p < 0.001). Novelty-centered significantly affects economic sustainable performance (β = 0.191, t = 6.867, p < 0.001), environmental sustainable performance (β = 0.215, t = 7.757, p < 0.001) and social sustainable performance (β = 0.174, t = 6.014, p < 0.001). The mediating impacts of novelty-centered and efficiency-centered BMIs in the link between GenAI and environmental, economic and social sustainable performance are significant. Additionally, AI regulation significantly moderates the relationships between novelty-centered and efficiency-centered BMIs, environmental, economic and social sustainable performance and GenAI.

5.6. Robustness Check

We examined whether nonlinear relationships might exist in our structural model by implementing two established tests [75]. First, we applied the Ramsey [76] RESET test to the latent variable scores from our PLS-SEM analysis. The results showed no evidence of nonlinearity for either the relationship between novelty-centered innovation and sustainable performance (F = 1.36, p = 0.259) or efficiency-centered innovation and sustainable performance, nor for the combined effects of these variables with GenAI adoption (F = 1.265, p = 0.284).

Second, we introduced quadratic (squared) forms of GenAI capability into the model and tested their significance using 5000 bootstrap samples. As shown in

Table 6. Assessment of nonlinear effects.

Nonlinear Relationship	Coefficient	p Value	f2	Ramsey’s RESET
QE (GenAI) → Efficiency-Centered BMI	−0.028	0.192	0.002	F = 1.265, p = 0.284
QE (GenAI) → Novelty-Centered IBM	−0.01	0.662	0.001
QE (GenAI) → Sustainable Performance	−0.024	0.174	0.002
QE (Novelty-Centered IBM) → Sustainable Performance	−0.014	0.318	0.001	F = 1.36, p = 0.259
QE (Efficiency-Centered BMI) → Sustainable Performance	0.023	0.077	0.003	F = 0.543, p = 0.582

, none of these squared terms were statistically significant. These findings indicate that the relationships remain linear across observed levels of GenAI capability, with no evidence of diminishing returns or inverted-U patterns.

Additionally, mediation and moderation effects were re-estimated using Hayes’ PROCESS macro (Models 4 and 1) with 5000 bootstraps, confirming consistent indirect and interactive effects across analytical methods (see

Table 7. Hayes’ PROCESS macro results.

Mediation
Path		Effect	BootSE		Confidence Intervals
GenAI → Novelty-Centered IBM → Sustainable Performance		0.104	0.012		(0.0817, 0.1280)
GenAI → Efficiency-Centered BMI → Sustainable Performance		0.113	0.011		(0.0909, 0.1360)
Moderation
Path	β	se	t value	p value	Confidence Intervals
GenAI*AI regulation → Sustainable Performance	0.134	0.012	11.498	0.000	(0.1109, 0.1566)
GenAI*AI regulation → Novelty-Centered IBM	0.111	0.011	9.437	0.000	(0.0878, 0.1339)
GenAI*AI regulation → Efficiency-Centered BMI	0.112	0.014	8.279	0.000	(0.0853, 0.1383)

). Together, these checks support the robustness and appropriateness of our linear specification.

5.7. Discussion

The increasing integration of GenAI into business operations has raised critical questions about its strategic implications for firm sustainability. While prior research has explored the impact of traditional AI on performance outcomes, limited attention has been paid to the specific role of GenAI in driving business model innovation (BMI) and sustainable performance, particularly in the context of emerging regulatory frameworks. Moreover, past studies have rarely examined how AI-driven innovation translates into firm-level environmental, social, and economic benefits through distinct innovation mechanisms such as novelty- and efficiency-centered BMIs. Addressing this gap, the present study aimed to investigate the influence of GenAI on firm sustainable performance, mediated by two types of BMI, and moderated by AI regulation. The analysis revealed a strong model fit with substantial predictive power, as indicated by the coefficient of determination (R²) values for the key endogenous constructs (Novelty-BMI = 0.422, Efficiency-BMI = 0.311, and Sustainable Performance = 0.614), and Stone–Geisser’s Q² value 0.390 exceeding the threshold of 0.35, confirming the model’s high predictive relevance.

The study results show that GenAI has a significant positive effect on both novelty-centered and efficiency-centered business model innovation (BMI), as well as on firm sustainable performance. This finding suggests that the adoption of GenAI technologies not only enables firms to introduce novel products, services, or delivery methods but also allows them to optimize internal processes and enhance operational efficiency. These two streams of innovation are central to improving a firm’s economic, social, and environmental performance. This result follows research that indicates GenAI will enhance creativity, automation, and data-driven decision-making, all of which are conditions for sustainable value creation [12]. In line with RBV, GenAI acts as a strategic enabler to empower firms to reconfigure capabilities and resources toward sustainable development. Therefore, the research highlights that GenAI is not only a technological tool but a dynamic capacity that enables innovation-driven organizational sustainability performance (Simo et al., 2024) [77].

Further, the results indicate that novelty-oriented and efficiency-oriented BMI significantly mediate the effect of GenAI on firm sustainable performance. To illustrate how these mediating pathways operate in practice, consider two plausible scenarios. First, a high-tech electronics manufacturer uses GenAI to co-design modular, recyclable products with customers—launching a product-as-a-service platform that redefines value creation (novelty-centered BMI). Over time, the same firm applies GenAI to optimize logistics, reduce material waste, and automate customer support, enhancing operational sustainability (efficiency-centered BMI). Second, a textile company adopts GenAI to improve cutting precision and predict machine failures, minimizing energy use and downtime (efficiency-centered BMI). With improved margins and stability, it then invests in AI-generated sustainable fashion lines and blockchain-tracked sourcing (novelty-centered BMI). These examples reflect how firms may follow different innovation trajectories—novelty-first or efficiency-first—yet both lead to enhanced sustainability outcomes through strategic reconfiguration.

The strength of the relationships provides meaningful insights for both theory and practice. First, the direct effect of GenAI on sustainable performance (β = 0.424) indicates a robust positive impact—suggesting that firms adopting GenAI are likely to see measurable improvements across environmental, social, and economic outcomes. However, nearly 37% of this total effect flows through business model innovation (15.5% via novelty-centered BMI and 23.1% via efficiency-centered BMI), highlighting that while GenAI enables progress, its full potential is unlocked only when paired with strategic reconfiguration.

Notably, efficiency-centered BMI (β = 0.280) exerts a stronger influence on sustainable performance than novelty-centered BMI (β = 0.206), suggesting that in the current stage of digital maturity, Chinese manufacturers gain greater returns from optimizing existing processes—such as predictive maintenance or waste reduction—than from launching radical new models. This may reflect an incremental approach to digital transformation, where firms prioritize cost savings and operational stability before venturing into high-risk innovation.

Furthermore, the significant moderating role of AI regulation (β = 0.131–0.213) shows that clear regulatory frameworks act as catalysts rather than constraints. Firms operating under transparent rules are better able to justify investments in GenAI and align them with green innovation strategies. This implies that instead of fearing regulation, managers should view compliance as a foundation for responsible innovation, while policymakers should design agile, enabling regulations that balance oversight with flexibility.

This points to the fact that the effect of GenAI on sustainability performance is not only direct but also directs its effects through mechanisms of strategic innovation. According to the RBV, this means GenAI provides the technological know-how (capabilities), but reconfiguring and putting the capabilities into action through mechanisms of innovative business models is what truly leads to improved sustainable performance. The strong mediation of novelty-focused BMI shows that GenAI facilitates companies to venture into new value propositions, customer bases, and revenue streams, supporting previous research that identified GenAI crucial contribution towards innovation and sustainability [12]. Likewise, the mediation of efficiency-focused BMI captures how GenAI supports companies’ intrinsic capabilities to streamline operations, minimize wastage of resources, and increase cost-effectiveness, key factors in achieving economic and environmental goals. These two paths reaffirm that GenAI becomes important to strategy when it is embedded in core business innovation activities, substantiating the RBV assertion that resources alone do not guarantee competitive advantage, it is their effective orchestration that matters [60].

The study further established that AI regulation moderates significantly the relationships between GenAI and both types of BMIs, as well as between GenAI and firm sustainable performance. The results confirm regulatory regimes shape how firms gain strategically from GenAI technologies. Specifically, where regulatory certainty, compliance processes, and legal frameworks are robust, firms deploy GenAI in ways that enhance novelty-driven and efficiency-driven BMI. This finding is in line with RBV, which suggests external conditions can enhance the performance of a firm [78]. Literature also suggests that enabling regulatory frameworks are institutional enablers, that enable firms to use AI for sustainable transformation [79]. Thus, regulation of AI is not solely restriction, but rather it can be a force of change deployed strategically, influencing GenAI adoption towards more and more innovative, sustainability-oriented, and sustainable business outcomes. These robust moderation effects underscore the importance of institutional and environmental contingencies being considered in utilizing high doses of cutting-edge technologies for long-term performance as well as competitive advantage.

6. Conclusions

This study explored the impact of GenAI on firm sustainable performance through the mediating effects of novelty-centered and efficiency-centered BMI and the moderating effect of AI regulation. Following the RBV, this study highlights that GenAI is a strategic resource that can enhance both innovation and sustainability performance. Data were collected from over 500 Chinese manufacturing firms, with 1192 middle-level managers’ replies in two time waves to attain temporal separation and reduce common method bias. The findings indicate that GenAI profoundly enhances both forms of BMI, which in turn mediate its positive effect on firm sustainable performance. Additionally, AI regulation was found to significantly strengthen the relationships between GenAI and both BMI types, as well as the direct path to sustainable performance. These results offer important implications.

6.1. Theoretical Implications

This research makes three key theoretical contributions. First, it extends the Resource-Based View (RBV) by illustrating the specific pathways through which Generative Artificial Intelligence (GenAI) enhances firm sustainable performance. By identifying GenAI as a strategic, valuable, and dynamic resource, this study provides empirical support that such digital technologies can strengthen firms’ capabilities to innovate and perform sustainably. This aligns with foundational view of Barney [36] that internal resources—when rare, inimitable, and organized—can provide sustained competitive advantage, particularly when embedded within advanced business model innovations. Second, the study deepens the understanding of BMI by distinguishing its dual dimensions—novelty-centered and efficiency-centered—as mechanisms through which GenAI drives sustainability outcomes. The significant mediation effects show that GenAI alone does not guarantee performance benefits unless supported by the strategic reconfiguration of how firms create and deliver value. Third, this research incorporates AI regulation as a boundary condition that moderates the effects of GenAI on BMI and sustainable performance. By showing that regulatory environments shape how effectively GenAI resources are deployed, the study integrates institutional perspectives into RBV-based models. It highlights that the interplay between organizational capabilities and external governance mechanisms is critical for maximizing the value of emerging technologies. This regulatory lens not only enriches the RBV but also emphasizes the importance of compliance and alignment in responsible AI adoption.

6.2. Implications for Practice

The findings of this study offer several valuable implications for managers, policymakers, and industry practitioners in the manufacturing sector. First, the significant positive effect of GenAI on both types of business model innovation—novelty-centered and efficiency-centered—suggests that firms should actively invest in generative AI technologies not merely as a tool for automation, but as a strategic enabler of innovation. Managers are recommended to build organizational readiness by building digital capabilities and training teams to integrate GenAI into operational and creative processes. Second, the findings emphasize that GenAI not only drives innovation but also directly and indirectly influences firm sustainable performance on environmental, social, and economic dimensions. This suggests that GenAI utilization can support broader sustainability goals, particularly when innovation is strategically coupled with green agenda and stakeholders’ expectations. For practitioners, this emphasizes the importance of leveraging AI not just for profitability but also for sustainable and responsible value creation. Third, the significant moderating effect of AI regulation suggests that firms with clearer and more favorable regulatory conditions are better placed to leverage GenAI. Therefore, managers should closely monitor regulatory trends and proactively align their AI initiatives with compliance standards and ethical norms. Similarly, policymakers are encouraged to design regulations that do not hinder innovation but instead guide firms toward the responsible use of AI technologies. Finally, by collecting data from over 500 Chinese manufacturing firms, this study provides practical insights particularly relevant to developing economies undergoing digital transformation. It demonstrates that with the right mix of technology, innovation, and regulatory awareness, firms in emerging markets could achieve sustainable outcomes.

6.3. Future Research Directions

Despite its contributions, this study has several limitations that suggest promising avenues for future research. First, our sample focuses on manufacturing firms in China’s Pearl River Delta—a technologically advanced region—limiting generalizability to less-digitalized or institutionally distinct contexts. Future studies should replicate this model in diverse geographic and industrial settings (e.g., inland provinces or emerging economies) using comparative designs to examine how national innovation systems and regulatory maturity shape GenAI adoption.

Second, while we include multiple subsectors (e.g., electronics, chemicals, textiles), we do not test for cross-sectoral differences due to subgroup size constraints. Given that high-tech industries may benefit more from novelty-centered BMI and process-heavy sectors from efficiency gains, future research should employ multi-group PLS-SEM or fsQCA to uncover boundary conditions based on automation level, R&D intensity, or environmental footprint.

Third, our reliance on self-reported data introduces potential perceptual bias. Future work should integrate objective indicators—such as ESG ratings, audited emission reports, or AI investment records—or adopt mixed-method approaches combining surveys with interviews or archival data to strengthen measurement validity.

Finally, although our analysis supports linear relationships, future studies could explore nonlinear dynamics, such as diminishing returns at high levels of GenAI use or “overregulation” effects where strict rules hinder innovation. Polynomial regression or response surface analysis could identify optimal regulatory thresholds.