Artificial-Intelligence-Enabled Innovation Ecosystems: A Novel Triple-Layer Framework for Micro, Small, and Medium-Sized Enterprises in the Chinese Apparel-Manufacturing Industry

Abstract

The rapid advancement of artificial intelligence (AI) in the traditional-apparel-manufacturing sector is accelerating innovation and transformation, as cutting-edge AI applications have been increasingly integrated into the industry in recent years. While China has made outstanding achievements in applying AI in the apparel-manufacturing sector, the adoption of AI by traditional apparel manufacturers has progressed slowly. This study aims to develop a sustainable triple-layer framework of an AI-enabled innovation ecosystem from grounded required AI capabilities and barriers to AI adoption, thereby generating the conceptual propositions for micro, small, and medium-sized Chinese apparel manufacturing. Through semi-structured interviews conducted with 20 organizations, this study qualitatively analyzes interviews with representatives from enterprises, universities, and apparel associations to determine the required AI capabilities and barriers to adopting AI. It proposes 13 propositions within a theoretical framework that addresses barriers and aligns multi-actor collaborations, ultimately forming a sustainable AI-enabled Triple-Layer Innovation Ecosystem Framework. This novel framework reflects the dynamic interplay between external knowledge absorption capacity and a firm’s internal innovation capacity, providing a theoretical foundation for understanding and advancing AI-driven innovation in the apparel-manufacturing sector.

1. Introduction

Artificial Intelligence (AI) is transforming low-tech industries [1], driving innovation and digitalization in traditional apparel manufacturing. Despite its large scale, apparel manufacturing remains labor-intensive and technologically underdeveloped [2,3,4]. This industry’s low-value products and intense price competition force firms to minimize costs, particularly labor expenses, to sustain profit margins. At the same time, fast-changing fashion trends necessitate highly flexible production, where manual labor enables quick adaptation to small-batch, customized orders. Moreover, the industry’s buyer-driven supply chain prioritizes outsourcing to low-cost regions, reinforcing cost-cutting imperatives.

As a leading global apparel manufacturer, China is pivotal in shaping global sourcing dynamics. Its dominance stems from its robust raw material supply, extensive production capacity, established industrial clusters, and an integrated supply chain [5,6]. Consequently, AI-driven innovations offer significant opportunities to enhance manufacturing efficiency and competitiveness through strategic AI adoption. To realize this transformation, China must implement comprehensive AI strategies to drive knowledge creation and technology spillover at a national scale [1]. Recognizing AI’s potential, the government of China has introduced policies to accelerate industrial upgrading and intelligent transformation. The “Accelerating the Transformation and Upgrading of the Traditional Manufacturing Industry Guidance” released by the Ministry of Industry and Information Technology of the People’s Republic of China (MIIT) under Union Regulation No. 258 (State Council, 2023) underscores AI integration as a key strategy for strengthening digital capabilities and global competitiveness. Despite ambitious policy initiatives, AI adoption remains challenging for China’s micro, small, and medium-sized enterprises (MSMEs), constituting 92% of the industry (China National Textile and Apparel Council, 2022). As defined by MIIT No. 300 (2011), these firms are hampered by financial constraints, inadequate technological infrastructure, and limited AI expertise [7]. Their labor-intensive operations and fragmented supply chains further hinder AI integration. Thus, while national policies advocate AI-driven transformation, practical implementation at the MSME level remains a significant hurdle.

The favorable policy environment has stimulated academic research on the roles of government, universities, and industries in fostering AI-driven innovation ecosystems [8,9]. Prior research has examined China’s AI policies, emphasizing how established enterprises leverage AI-driven transformations and Triple-Helix (TH) collaborations with universities and research institutions [8,9,10]. Studies have also explored AI adoption in traditional sectors for open innovation [11,12], yet they primarily focus on resource-rich firms, neglecting MSMEs, which form the backbone of China’s apparel-manufacturing industry. Moreover, the existing literature mainly frames innovation ecosystems as structured networks where firms strategically engage with external actors, such as research institutions and government agencies, to access technological resources and policy incentives [13]. However, less attention is paid to MSMEs, whose boundaries are becoming increasingly fluid due to AI-driven transformations. Unlike large firms, which can systematically integrate AI, MSMEs operate in an environment of uncertainty, limited absorptive capacity, and resource constraints, making technological adoption and supply chain restructuring particularly challenging. Their positioning within AI innovation networks remains underexplored. Furthermore, current attempts to integrate Triple-Helix models, dynamic capabilities, and innovation ecosystems in AI adoption research lack a coherent mechanism for supporting an open innovation ecosystem that effectively aligns internal organizational constraints with external collaboration dynamics [14,15]. This gap is particularly pronounced for apparel MSMEs, for which unique industry characteristics and complex operational realities amplify resistance to AI-driven transformation. Thus, this study aims to bridge these gaps by developing a framework that formulates propositions for the sustainable innovation ecosystem of MSMEs in China’s apparel-manufacturing sector. To achieve this objective, this research addresses the following three key questions:

RQ1: What are the emerging AI capabilities required for MSMEs?

RQ2: What are the key challenges hindering AI adoption in the Chinese manufacturing sector?

RQ3: How can an AI-enabled sustainable innovation ecosystem facilitate collaboration among enterprises, universities, associations, and governments?

By answering these questions, this study contributes to a more holistic understanding of how AI reshapes industry boundaries and firm interactions within the innovation ecosystem of MSME apparel manufacturers in China. We accomplish this by developing 13 propositions that provide theoretical and practical significances into the evolving AI-driven innovation landscape. Theoretically, by categorizing key required AI capabilities in the Chinese apparel-manufacturing sector and the factors hindering their AI adoption, this study provides a theoretical lens in a novel theoretical triple-layer framework for innovation ecosystems in order to reveal how open innovation within the apparel industry, universities, associations, and government entities helps entities leverage AI technologies for mutual benefit. Practically, the framework developed offers guidance on how traditional businesses improve collaboration with universities, associations, and government agencies to co-create values in the AI-enabled innovation ecosystem. Simultaneously, the research outcomes provide an innovation path for university talent cultivation in the AI-driven innovation context. Additionally, policymakers can use the findings to develop strategies that encourage the integration of AI into industry and innovation systems, contributing to the broader goal of sustainable economic development in the Chinese apparel-manufacturing sector, cementing practical policy measures for the transformation and upgrading of the apparel industry.

2. Research Background

2.1. Innovation Ecosystem

Innovation ecosystems have become a key research focus, integrating strategy, innovation, and entrepreneurship [16,17,18,19]. Despite varying definitions, they broadly refer to joint value creation among interconnected actors [20]. Granstrand and Holgersson (2020) [19,21] identified four core components, namely, actors, collaboration, activities, and institutions, further incorporating artifacts such as technologies and resources, aligning with open-innovation theory. Open-innovation ecosystems rely on external knowledge flows, wherein firms collaborate with universities, industry peers, and government agencies [22,23]. In Industry 4.0 and 5.0, AI and digital technologies accelerate knowledge sharing, enhance collaboration, and strengthen adaptability, emphasizing learning, external knowledge integration, and joint value creation. Building on this foundation, the next section explores how AI capabilities intersect with open-innovation systems.

2.2. AI Capabilities and Open-Innovation Ecosystems

Mikalef and Gupta [24] define AI capabilities as tangible and intangible resources that enhance creativity and performance. Building on the concept of innovation ecosystems being networks of interconnected actors driving joint value creation, AI has emerged as a critical enabler of ecosystem-wide innovation by enhancing knowledge exchange, decision-making, and competitive positioning [12]. Open-innovation ecosystems rely on AI capabilities to facilitate collaboration, accelerate digital transformation, and sustain a dynamic, innovation-driven economy [25]. To maintain competitiveness in volatile industries, firms are increasingly leveraging AI to drive continuous innovation [26,27]. The openness of these ecosystems allows firms to access a diverse range of tangible and intangible resources to accelerate their innovation processes [28]. Additionally, ecosystem-wide collaboration enhances value creation across supply chains, reinforcing the link between digital tools, innovation strategies, and firm performance [28]. In this context, AI is the backbone of modern innovation ecosystems, driving knowledge sharing, dynamic capabilities, and adaptability. Given the central role of AI in supporting ecosystem-wide learning and adaptability, the next section examines how institutional actors such as universities, industry, and government interact to enable and regulate AI adoption. This is addressed through the TH framework.

2.3. TH in Innovation Ecosystems

The TH model, which emphasizes university–industry–government interactions, is a fundamental framework for understanding innovation ecosystems [29]. It highlights how these three key actors collaborate and thus drive knowledge-based innovation, facilitate technology transfer, and shape entrepreneurial dynamics [30]. The TH model has been widely adopted to analyze stakeholder relationships within innovation ecosystems, supporting descriptive and prescriptive approaches to ecosystem development (Chinta & Sussan, 2018; Pique et al., 2018, as cited in [8]). The TH model emphasizes synergistic interactions, where knowledge exchange between academia, industry, and government fosters open innovation and strengthens economic growth [31]. Research suggests strong government–academia–industry linkages significantly enhance enterprises’ performance by fostering resource sharing, talent development, and innovation diffusion [32,33]. Governments act as regulators, funders, and facilitators of ecosystem development [34]. In emerging economies, governments play a crucial role in fostering TH ecosystems, where policies actively support open innovation, startup incubation, and knowledge-based economic growth [33].

2.4. Synthesis of the Existing Research Gaps

Building on the TH framework in innovation ecosystems, it is evident that AI adoption introduces new complexities in government–industry–university interactions, particularly in China’s apparel-manufacturing sector. While China has taken a leading role in AI development [8,35], a significant research gap remains regarding how government, universities, and apparel-manufacturing firms co-create value within AI-driven ecosystems. Furthermore, the origins of this research stem from the researchers’ genuine interest in how Chinese apparel manufacturers leverage AI capabilities to construct a sustainable innovation ecosystem. However, existing studies have yet to ground AI capability definitions within innovation ecosystem parameters, particularly in regard to the apparel-manufacturing sector. There is a lack of research providing specific propositions and a theoretical framework for addressing AI adoption barriers in this context. Thus, in contrast to previous studies that either focus on resource-rich large firms or treat AI adoption as an isolated technological event, we developed an integrative, multi-level framework grounded in the specific context of labor-intensive apparel MSMEs. The triple-layer framework not only identifies firm-level capabilities and barriers but also explicitly connects them through empirically derived collaboration mechanisms. This direct integration of capability needs, institutional constraints, and ecosystem enablers represents a novel theoretical contribution to the literature on AI adoption and innovation ecosystems in traditional manufacturing sectors. Importantly, unlike existing AI adoption models that treat barriers, capabilities, and collaboration in isolation, this study integrates these elements into a recursive, empirically grounded framework specific to the low-tech MSME context.

3. Methodology

3.1. Data Collection Procedure

We employed semi-structured interviews as the primary data collection method, allowing for an in-depth exploration of AI adoption challenges and capabilities in China’s apparel-manufacturing sector. A grounded theory-driven analytical approach was adopted to ensure a rigorous and context-sensitive understanding of the mechanisms underpinning innovation ecosystems was obtained.

We adhered to ethical research standards in the interview process, with the participants’ confidentiality ensured, and they were notified that they could withdraw at any stage [36] (Bhattacherjee, 2012). Audio records of the semi-structured interviews were securely stored, and personal data was used solely for academic purposes, ensuring an honest presentation of findings.

A combination of purposive, snowball, and theoretical sampling was employed to capture diverse perspectives from apparel industry managers, university researchers, and association representatives [37,38]. Initially, the purposive sampling targeted participants with relevant expertise in AI adoption and innovation ecosystems. Given the limited accessibility of apparel-manufacturing managers, snowball sampling facilitated additional recruitment through industry associations (Coleman, 1958, as cited in [38]). Theoretical sampling, applied iteratively, refined emerging concepts and expanded participation across key apparel clusters in the Yangtze River Delta, the Pearl River Delta, and Northern China (Glaser & Strauss, 1967, as cited in [38]). This integrated approach ensured theoretical alignment and qualitative depth in understanding AI-driven innovation in MSMEs [37].

This study began with defining research objectives and designing interview questions aligned with three key research questions. A flexible semi-structured interview guide was developed to balance thematic structure with adaptability. Before formal data collection, ethical research protocol approval was obtained. To ensure clarity, pilot interviews with three participants (one industry manager, one academic, and one association leader) were conducted before formal data collection. A total of 15 participants were interviewed (see

Table 1. Preliminary interviewee information.

Interviewee ID.	Organizational Affiliations	Positions	Working Experience	Firm Location
A1	Associations	Secretary general	16–20 years	North
A2		Director	16–20 years	North
I1	Industries	Production manager	>20 years	North
I2		OEM business manager	10–15 years	North
I3		Fabric supplier	16–20 years	YRD
I4		ODM business manager	10–15 years	PRD
I5		Customer (retail)	10–15 years	PRD
I6		ODM business manager	10–15 years	North
I7		Customer (retail)	16–20 years	North
I8		OBM business manager	16–20 years	North
U1	Universities	Dean/professor	>20 years	North
U2		Professor	>20 years	North
U3		Dean/professor	16–20 years	YRD
U4		Specialized course instructor	10–15 years	North
U5		Specialized course instructor	10–15 years	YRD

Legend: A = association; I = industry; U = university; YRD = Yangtze River Delta; PRD = Pearl River Delta; OBM = original equipment manufacturer; OEM = original equipment manufacturer; ODM = original design manufacturer.

): eight apparel-manufacturing managers from large enterprises and MSMEs across China’s key apparel clusters, namely, the Yangtze River Delta (advanced), the Pearl River Delta (advanced), and North China (traditional); five university teaching fellows engaged in research on AI, innovation, and apparel manufacturing; and two apparel association leaders serving as industry representatives. Participants were recruited through professional networks, including current and former colleagues and association deputies, with snowball sampling facilitating additional referrals [39]. All interviewees had over ten years of industry or academic experience. Interviews were conducted between June and September 2024, employing six face-to-face interviews (40–80 min) in offices and cafes and nine phone interviews (20–50 min), aligning with standard semi-structured interview durations [40]. Anonymity agreements were established, and informed consent was obtained from all participants. The interviews, conducted in Chinese, resulted in 99,448 transcribed Chinese characters, with key quotations translated into English during the coding process.

To ensure theoretical saturation, five additional semi-structured phone interviews were conducted after completing the open, axial, and selective coding process (see

Table 2. Preliminary Interviewwe Informants for Saturation Examination.

Interviewee ID.	Organizational Affiliations	Positions	Working Experience	Locations
I9	Industries	ODM business manager	10–15 years	North
I10		OEM production manager	>20 years	YDT
I11		OBM CEO	>20 years	YDT
U6	Universities	Specialized course instructor	10–15 years	North
U7		Specialized course instructor	10–15 years	North

Legend: A = association; I = industry; U = university; YRD = Yangtze River Delta; PRD = Pearl River Delta; OBM = original equipment manufacturer; OEM = original equipment manufacturer; ODM = original design manufacturer; CEO = chief executive officer.

). These interviews, lasting 30–60 min, confirmed that no new conceptual categories emerged, proving the key themes were saturated [41]. Interviewees’ names were kept anonymous, and each one provided informed consent. A single researcher conducted the interviews in October 2024. All interviews were audio-recorded in Chinese, transcribed in 42,853 Chinese characters.

3.2. Coding for Data Analysis

Grounded-theory coding techniques were applied to systematically analyze interview data and construct an AI-enabled innovation ecosystem framework. In this approach, raw data are transformed into meaningful insights by identifying patterns, relationships, and theoretical constructs [42]. Through open, axial, and selective coding processes, data were manually coded using NVivo (Version 15.0.0) to ensure rigor and transparency [43].

The first phase, open coding, involved a line-by-line examination of interview transcripts to extract AI capabilities and adoption barriers. Emerging concepts were categorized using in vivo codes (field-specific terminology) and broader theoretical constructs [42]. Next, axial coding was used to refine these first-order categories by establishing relationships between AI capabilities, adoption barriers, and ecosystem collaboration (Strauss, 1987, as cited in [42]). To enhance validity, themes were adjusted based on literature insights and interview data (Kumar et al., 1993, as cited in [1]). Finally, selective coding was used to integrate the core themes into aggregate dimensions, representing higher-level theoretical abstractions [42]. This structured approach ensured that AI capabilities and adoption barriers were categorized systematically, aligning with this study’s research objectives. By employing an iterative coding process, we obtained findings that provide a comprehensive and theoretically grounded understanding of AI adoption among China’s apparel MSMEs.

4. Findings

4.1. Overview of the Structure of the Data

This section introduces the structure of the data derived from grounded theory coding, serving as the foundation for identifying AI capabilities and adoption barriers in subsequent findings. Figure 1 and Figure 2 illustrate the complete data structure, which is based on the results of open, axial, and selective coding. As a final selective-coding step, we developed theories about the logic and linkages across aggregate dimensions, second-order themes, and first-order categories [44]. We sought to generate the required AI capabilities and the challenges when adopting AI in manufacturing and production processes, thus building an innovation ecosystem where the collaborating actors leverage resources in activities based on China’s institutional (regarding policies and standards) context through contrasting lines of insight from the interviews. We presented the initial results to two key informants to validate the results through analysis. Constant comparison and memo writing were conducted during these processes, and data saturation was determined to be relevant after the first follow-up [42,45,46,47,48,49].

Drawing on empirical data from the 15 interviews, we identified and conceptualized core AI capabilities in apparel-manufacturing practices, incorporating insights not only from managers, suppliers, and customers but also educators and apparel association leaders with expertise in the fashion industry. While foundational AI capabilities have been established in previous studies, our findings focus on unexplored capabilities, addressing gaps in AI applications within apparel production processes. To identify AI capabilities for apparel production, we initially focused on interview questions about current issues in traditional apparel production and then expanded the questions to include AI technology awareness and the role of AI in addressing these issues, thereby exploring barriers and challenges to AI adoption. We present our findings and analysis in two parts: one relating to AI capabilities, and the other dealing with the barriers to AI adoption.

4.2. Required AI Capabilities

The key AI capabilities required in the Chinese apparel-manufacturing sector were categorized into adaptive production and augmented human–AI collaboration capability. The most frequently mentioned capabilities include order management and ensuring short lead times, producing small-batches with a quick turnaround, increasing work efficiency, fabric and trimming selection, and AI–human co-work. These capabilities enable firms to enhance their operational efficiency, improve their production flexibility, and optimize collaboration between human expertise and AI technologies, aligning with the increasing demands for speed, customization, and sustainability in the apparel sector.

4.2.1. Adaptive Production Capability

Adaptive production capacity forms the foundation for AI integration in apparel manufacturing, allowing firms to respond to the dynamic requirements of small-batch production, rapid order fulfillment, and fluctuating customer preferences. Adaptive production is particularly crucial for MSMEs due to their resource constraints and frequent requirement for small-batch, customized orders. AI adoption in this industry remains in an early stage [1], so the initial focus was on practical enhancements in order scheduling, efficiency, and quality control. The ability to manage flexible order scheduling is particularly vital, as evidenced by the frequent mention of order management and short lead times, both ranking highest in importance among industry respondents. These mechanisms ensure that manufacturers can effectively handle small-batch orders, for which rapid responsiveness is a key competitive factor. Predictive AI plays an essential role in this process by anticipating material needs, thereby streamlining resource allocation and reducing delays. As one respondent explained, “In production, AI can predict my needs, such as how many linings and trimming materials I require and their quantities when orders come in” (I8). This predictive capability allows firms to prepare for fluctuations in customer demands and facilitates more efficient production planning.

In addition to scheduling, dynamic priority adjustment and real-time response mechanisms support manufacturers in adapting to evolving customer requirements. AI enhances decision-making flexibility, allowing manufacturers to adjust production parameters based on real-time data. For instance, the ability to modify garment styles based on last-minute customer requests is a key challenge that AI can help businesses overcome. A respondent from the industry highlighted this aspect, stating, “For trench coats or jackets, only basic parameters like shoulder width and length are needed, and minor adjustments can be made… if today’s customer orders a long, loose trench coat but tomorrow needs a cinched waist style, that’s definitely not feasible without AI” (I6). This illustrates how AI-driven flexible production systems enable real-time customization without disrupting workflow, reinforcing the industry’s ability to meet diverse and rapidly changing demands.

Beyond production flexibility, efficiency and quality control emerged as fundamental needs within adaptive production. AI significantly improves work efficiency by automating labor-intensive tasks, particularly cutting and assembling garments. Respondent I1 emphasized this transformation: “A cutting machine can handle over 100 complex pieces a day, while manually cutting 10 sets a day is already impressive”, underscoring AI’s potential to enhance productivity by accelerating repetitive tasks, allowing human workers to focus on more complex design and quality-related aspects. Despite these improvements, challenges remain in balancing precision and practical flexibility in AI-driven quality control systems. AI systems tend to eliminate all errors, yet some minor variations in materials and stitching are acceptable within industry standards. One respondent illustrated this challenge: “The machine is too precise. In reality, some error is acceptable, but AI machines might eliminate all errors. Whether we’re making fabrics or clothes, achieving zero error is impossible” (I13). Thus, flexible quality control systems are essential to accommodate these nuances while ensuring that production remains both efficient and adaptable.

Another critical component of adaptive production is fabric and trimming selection, which poses challenges due to the vast array of material options available. AI helps automate this complex selection process by recommending materials based on customer preferences and past orders. One respondent explained, “For lining and trimming materials, every task involves different selections. We need to choose from thousands of options based on customer preferences and create combinations” (I4). This highlights the importance of AI in navigating extensive material databases, ensuring that manufacturers can efficiently manage customization while maintaining consistency in product quality. These AI-driven mechanisms, including flexible scheduling, production efficiency, and quality optimization, collectively establish the foundation of adaptive production, enabling manufacturers to optimize operational workflows and dynamically respond to the evolving complexities and demands of the apparel industry.

While these capabilities provide clear operational advantages, they may also lead to unintended environmental consequences. By enabling rapid product development and small-batch customization, AI technologies can accelerate fashion turnover and intensify consumption behaviors. This dynamic potentially contradicts global efforts toward sustainable consumption and circular economy models. For instance, the European Parliamentary Research Service has advocated for reducing individual clothing purchases to five items per year to mitigate environmental impacts. To ensure alignment with sustainability goals, AI adoption should be accompanied by responsible innovation strategies such as the development of rental platforms, integration into green supply chains, and support for zero-waste design practices.

4.2.2. Augmented Human–AI Collaboration Capability

Unlike large enterprises, MSMEs often lack dedicated R&D teams, making human–AI collaboration especially significant. Beyond adaptive production, the apparel industry requires augmented human–AI collaboration capacity, emphasizing the synergy between human expertise and AI-driven automation. This capability aligns with the Industry 5.0 paradigm, where AI supports rather than replaces human labor, ensuring that production processes retain flexibility, craftsmanship, and efficiency [50]. A key aspect of this capability is AI–human cooperation, particularly in processes where human supervision remains indispensable. While AI enhances efficiency in structured workflows, tasks such as sewing and handling delicate materials still require human intervention. As respondent I6 noted, “While certain processes can intermittently or fully automate in mature workflows, human supervision is ultimately still required”. This underscores the industry’s recognition that fully autonomous systems are impractical for tasks demanding adaptive decision-making and specialized expertise. Similarly, AI-driven pattern-cutting automation reduces the time required for precision-driven tasks, freeing skilled workers to focus on creative and value-added activities. One industry participant illustrated this impact: “Typically, a pattern maker takes a full day to finish a set of samples, including revisions. For tailored pieces, it takes at least 4–6 h after taking measurements. Automating sample creation and pattern cutting significantly speeds up the process” (I1). These examples highlight how AI optimizes repetitive processes, improving production efficiency while preserving the human element in decision-making.

Complex production processes also necessitate human involvement, particularly in material selection and customized product development. AI plays a role in optimizing raw material procurement, but human expertise is crucial in managing material properties that require deep technical knowledge. As one respondent noted, “The customer categories involve deep-level raw material development, including chemical and structural types, and require a large number of operators” (I2). This illustrates that while AI can automate certain decision-making aspects, human oversight remains essential for complex, high-stakes manufacturing tasks. In addition to direct production roles, AI-driven sustainable supply chain management is a crucial component of augmented collaboration. AI enhances sustainability by reducing labor costs, optimizing supplier selection, and facilitating environmentally friendly production practices [6]. One participant highlighted the economic benefits of AI in manufacturing, stating, “AI replaces manual labor and reduces costs” (I3). Furthermore, AI assists in supplier sourcing, allowing firms to identify specialized subcontractors (“satellite factories”) based on their expertise. As respondent I6 explained, “I need to locate ‘satellite factories,’ each with its own strengths”, emphasizing AI’s role in creating an agile and adaptive supply chain. Additionally, AI supports green manufacturing initiatives, ensuring that production processes align with environmental sustainability goals. As one participant observed, “They are now undergoing green upgrades, increasingly meeting the needs of customers and end-users” (I7). These advancements highlight how AI facilitates cost-effective, sustainable operations, integrating efficiency with responsible production practices.

Another essential component of augmented AI–human collaboration is big data training and analytics, which enhance production planning and quality control. AI enables data-driven decision-making, allowing manufacturers to predict demand and optimize production workflows. As respondent I6 described, “The template library contains 100 patterns, which can be used for machine learning and computation. Once key parameters like shoulder width and waist circumference are input, the angle can be directly cut, and sewing naturally follows”. This capability standardizes production processes, reducing variability while enhancing precision in customized orders. These interview findings illustrate how augmented human–AI collaboration represents a fundamental shift in apparel manufacturing, where AI enhances efficiency, quality, and sustainability while retaining human expertise for specialized, complex tasks. This balanced approach ensures that AI enhances rather than replaces human labor, ultimately fostering a more adaptive, efficient, and innovative manufacturing ecosystem.

While AI enhances operational efficiency by automating routine processes, its broader social implications deserve careful consideration. In labor-intensive MSMEs, automation may reduce the demand for low-skilled workers, potentially leading to workforce displacement. This shift highlights the importance of coordinated policy and organizational responses. Without proactive retraining programs and employment transition support, the benefits of AI may come at the cost of increased social burdens. Therefore, AI adoption should be approached not only as a technical upgrade but also as a socio-economic transformation that requires governments and enterprises to jointly address reskilling and social responsibility.

4.3. Factors Hindering AI Adoption

The primary barriers hindering AI adoption in the Chinese apparel-manufacturing sector were categorized into industry factors, university factors, and government factors. These barriers reflect systemic challenges deeply embedded within the industry’s structural and operational framework. Notably, the complexity of garment design, unawareness of AI among top managers, and reliance on intermediary organizations emerged as the most frequently cited obstacles, underscoring the technological, financial, managerial, and policy-driven constraints that collectively impede the widespread integration of AI. The interplay of these factors creates a reinforcing cycle, wherein technical limitations, economic pressures, and institutional gaps mutually exacerbate the challenges associated with AI implementation.

4.3.1. Industry Factors

The barriers to AI adoption in the apparel-manufacturing industry arise from technical, financial, managerial, and competitive pressures, creating a reinforcing cycle that limits firms’ ability and willingness to integrate AI technologies. Among these barriers, the complexity of garment design emerged as the most significant technical constraint. Unlike standardized industries, where AI-driven automation has been successfully implemented, apparel production often requires intricate craftsmanship and adaptability to varying materials and designs, posing challenges for AI integration. As Respondent I4 pointed out, “They are limited to certain categories, such as knitwear, denim, or down jackets. These products are relatively simple and easy to operate”. This limitation significantly reduces AI’s applicability in high-complexity apparel production, particularly for garments requiring customized fits, layered construction, or delicate materials. Consequently, rather than transforming the industry, AI adoption remains restricted to specific production segments. Beyond technical constraints, financial limitations further inhibit AI adoption, particularly for MSMEs, which dominate the sector but often operate on narrow profit margins. The high initial investment required for AI technology and uncertainty regarding return on investment (ROI) deter firms from pursuing digital transformation. As Respondent I3 emphasized, “For small businesses like ours, it’s difficult to compete because the cost is too high”. Even in cases where AI technologies could theoretically improve efficiency, low profitability and unclear financial benefits make manufacturers reluctant to commit to significant investments. Respondent I6 reinforced this concern: “The issue of whether the return on investment can be achieved is real, as most people are unwilling to buy machines like ours, or even a 3D software solution”. These financial concerns are compounded by managerial barriers, particularly a lack of awareness and understanding of AI among top executives. Decision-makers, especially in risk-averse firms, often fail to recognize the strategic advantages AI can provide, leading to inertia in adoption. As Respondent I4 noted, “Many of our companies have not reached the management level required; they don’t understand how to use AI or even its potential”. Without knowledge of AI’s operational feasibility and long-term benefits, firm leadership remains hesitant to support AI implementation, further delaying technological progress. In addition to financial and managerial constraints, competitive pressures within the industry contribute to reactive, rather than proactive, AI adoption. The socio-cultural phenomenon of self-involvement (written in Chinese as内卷, with the Pinyin name Neijuan), wherein firms adopt AI not as an innovation strategy but as a defensive measure, creates an environment where technology is integrated superficially rather than effectively. As Respondent I4 explained, “It’s all due to intense internal competition”. This competitive landscape fosters short-term, incremental AI adoption, where firms seek to match competitors rather than strategically develop AI-driven capabilities that enhance productivity and efficiency. Price competition further exacerbates this issue, as firms are pressured to cut costs rather than invest in technological upgrades. Respondent I4 highlighted how pricing dynamics in online retail platforms such as Taobao further discourage AI investment: “Taobao has now introduced a price comparison system that offers many similar products at the time of payment. This mechanism disrupts the market; the Red Ocean strategy relies on price competition and somewhat undermines healthy market development”. Aside from economic and competitive pressures, demographic and workforce-related challenges further limit AI adoption. The aging workforce and lack of technical skills among apparel workers present significant obstacles, as AI implementation requires a workforce capable of operating and managing digital systems. Respondent I4 emphasized the industry’s shifting labor demographics: “With the 70s’ generation retiring, even fewer from the 80s’ generation are entering the industry. You must adopt intelligent solutions”. However, many workers lack the educational background required to transition to AI-assisted roles, making large-scale implementation difficult. Respondent I4 further noted that “Due to workers’ skill levels and knowledge reserves, many errors are frequent, and work progress is slow”. Finally, market instability adds another layer of uncertainty, discouraging firms from making long-term technological investments. The shifting retail landscape, particularly the decline of physical stores, creates unpredictability in demand, provoking firms to hesitate in committing resources to AI. Respondent I7 illustrated this concern: “Our difficulty lies in the sharp decrease in the number of stores, which leads to fewer customers”. These interconnected barriers, including technical complexity, financial constraints, managerial hesitation, competitive pressures, workforce limitations, and market volatility, reinforce the industry’s reliance on traditional labor-intensive production models rather than AI-driven digital transformation. Overcoming these challenges requires a coordinated approach involving financial incentives, managerial capacity-building, and workforce training initiatives, ensuring that AI adoption aligns with industry needs and operational realities.

4.3.2. University Factors

The role of universities in AI adoption within the apparel-manufacturing industry is constrained by misalignment between academic training and industry demands, a shortage of AI-skilled professionals, and limited university–industry collaboration. These factors collectively contribute to a talent bottleneck, where the workforce lacks the necessary expertise with which to support AI integration in apparel production. A critical issue is the lack of AI talent for emerging AI departments, highlighting the gap in educational programs tailored to AI applications in apparel manufacturing. Traditional design-focused curricula fail to incorporate interdisciplinary knowledge essential for AI-driven manufacturing, limiting graduates’ ability to apply AI solutions effectively. Respondent I3 emphasized the implications of this gap: “If AI is introduced, it could also attract highly knowledgeable talent into traditional industries. We don’t just need people skilled in making clothes but also engineers, those in industrial engineering, and professionals skilled in software system development”. However, current academic programs remain largely disconnected from industry needs, creating a workforce that is not adequately prepared for AI-integrated production environments. The mismatch between university curricula and industry requirements further exacerbates this issue. As Respondent I8 pointed out, “This doesn’t quite align with us because we focus more on specialties. If schools don’t offer these specialties, such as sportswear and performance fabrics, very few students can adapt”. This misalignment not only limits AI adoption at the enterprise level but also reduces the number of graduates equipped to develop and maintain AI systems within the apparel industry. Moreover, the lack of interdisciplinary talent presents another challenge. AI-driven innovation in apparel manufacturing requires knowledge spanning textile science, data analytics, industrial engineering, and automation, yet most university programs remain siloed within traditional disciplines. Respondent I2 highlighted this issue: “Natural fabrics with enzymes involve chemical knowledge and even food science. If you’re solely trained in fashion, you might know nothing about this. Fabrics are composed of chemical materials, making the field of fashion inherently interdisciplinary”. This narrow academic focus limits the industry’s ability to recruit talent capable of integrating AI with material science, supply chain optimization, and advanced production processes. Additionally, universities often fail to provide students with opportunities for practical exposure to AI technologies. This lack of access to intelligent equipment prevents students from developing hands-on experience with AI-driven systems, further widening the skills gap. As Respondent U4 noted, “If manufacturing truly transitions to being technology-driven, it will likely attract more scientific and technical talent and expose them to advanced technology and equipment”. Furthermore, university–industry collaborations remain limited, restricting opportunities for firms to engage with academic institutions in AI research and development. Existing partnerships mainly focus on garment design rather than AI integration, leaving apparel manufacturers without meaningful academic support for AI-driven production advancements. Respondent A2 observed that “The association is also exploring university-enterprise collaboration. I know that collaboration in Dalian mainly focuses on garment design and lacks technological influence”. Without structured collaboration between academia and industry, research on AI applications in the apparel sector remains fragmented, and firms lack access to trained professionals who understand both AI and apparel-manufacturing workflows. This disconnect reinforces the systemic gap between AI-driven technological advancements and practical industry implementation, further slowing AI adoption in apparel manufacturing. Therefore, addressing these barriers requires curriculum reform, interdisciplinary training programs, and stronger university–industry engagement, ensuring that the workforce is equipped with the skills necessary for AI-integrated production environments.

4.3.3. Government Factors

The Chinese government plays a central role in shaping AI adoption, yet policy inefficiencies, regional disparities, and a preference for supporting large enterprises over MSMEs present substantial obstacles. While national AI policies emphasize digital transformation, their implementation remains inconsistent across different regions and firm sizes, limiting their effectiveness in fostering AI adoption within the apparel industry. One of the primary issues is the reliance on intermediary organizations for AI policy implementation. Industry associations, which serve as conduits for government support, often lack the technical expertise or resources required to assist manufacturers in AI adoption. Instead, these organizations focus on administrative compliance rather than practical support. Respondent I5 illustrated this inefficiency: “They have such associations, but they mostly exist to fulfill policy-related tasks and address very few actual problems”. This results in a gap between policy objectives and real-world implementation, as firms struggle to navigate AI integration without localized technical and financial assistance. Additionally, regional disparities in AI support create unequal access to AI-related funding and training programs, disproportionately benefiting firms in economically advanced regions such as the Yangtze River Delta and Pearl River Delta while leaving manufacturers in Northern China with fewer opportunities. Respondent U1 described this regional divide: “Some southern companies are doing quite well in this regard, while in the north, there is relatively less involvement”. This imbalance limits the widespread scalability of AI adoption, as only firms in well-funded regions gain access to AI infrastructure, financial support, and skilled labor pools. Another structural issue is the government’s preference for supporting large enterprises over MSMEs. While AI adoption policies and funding initiatives exist, they primarily target large-scale firms rather than smaller apparel manufacturers, which comprise most of the industry. Respondent I1 underscored this disparity: “If you only have 30 orders, setting up an intelligent system for you is pointless”. This top-down policy bias sidelines MSMEs, which often lack the financial capacity to invest in AI without targeted incentives. Moreover, government funding initiatives focus on firms with pre-existing AI infrastructure, leaving smaller businesses without the foundational support required to transition to AI-driven production. Further compounding this issue is the perception of apparel manufacturing as a non-key industry in China’s broader AI development strategy. Government policies prioritize high-tech and engineering sectors, while apparel manufacturing, despite its economic significance, receives comparatively less AI-focused support. Respondent U1 noted that “Schools seem to approach it from a design perspective, with little official funding support. AI application in art fields is almost non-existent, but there’s a clear policy inclination toward engineering and AI-related fields”. This bias results in a lack of targeted AI initiatives for apparel manufacturers, reducing opportunities for sector-specific AI applications that could improve productivity, quality control, and supply chain efficiency. Additionally, the absence of financial backing for AI policies creates a disconnect between policy intent and practical implementation. While AI adoption is widely encouraged in government directives, the lack of direct subsidies or financial incentives limits MSMEs’ ability to afford AI technologies. Respondent I3 summarized this challenge: “There is some policy support for using AI, but no subsidies for now”. This limitation further reinforces the divide between large enterprises and MSMEs, as only firms with sufficient capital can independently fund AI integration.

4.4. Conceptual Framework Development

To consolidate the empirical findings presented in Section 4.2 and Section 4.3, we developed a three-layer conceptual framework that synthesizes how AI capabilities, adoption barriers, and external collaboration mechanisms interact to shape innovation ecosystems within China’s apparel MSMEs. This framework was constructed through a grounded-coding process and reflects patterns derived from interview data as well as alignment with the existing innovation ecosystem literature.

As illustrated in Figure 3, Layer 1 identifies the core AI capabilities required for MSMEs in the apparel industry, namely, the capacity for adaptive production and augmented human–AI collaboration. These capabilities are essential for improving operational agility, enhancing production efficiency, and enabling responsiveness to volatile consumer demands and complex supply chain configurations. Layer 2 captures the barriers that hinder AI adoption, categorized into industry-level, university-level, and government-level constraints. These include financial limitations, skill shortages, organizational inertia, misaligned academic training, and regional disparities in policy implementation. Layer 3 represents the collaborative mechanisms underpinning AI adoption in apparel MSMEs. It serves as the ecosystem’s central engine, addressing Layer 2 barriers (e.g., high adoption costs, skill shortages, and fragmented data-sharing) while enabling Layer 1 AI capabilities (adaptive production and human–AI collaboration). This layer is where actionable strategies and ecosystem-level coordination occur, supporting firm-level transformation.

In the following section, 13 propositions are developed to explain how the collaborative mechanisms represented in Layer 3 can overcome adoption barriers and support sustainable, AI-driven innovation in the apparel-manufacturing sector.

It has to be highlighted that this framework is particularly suited to the context of MSMEs, which operate under distinct constraints compared to large corporations. Unlike resource-rich enterprises that often have in-house R&D capabilities, flexible funding channels, and mature data systems, MSMEs must navigate AI adoption with limited financial, technical, and organizational capacity. Furthermore, their reliance on local policy environments, public support schemes, and institutional linkages makes them especially sensitive to systemic enablers such as government intervention, university collaboration, and industry associations. The three-layer framework reflects these interdependencies and provides a structure that aligns technological adoption with institutional support. This combination is essential for sustainable digital transformation in the MSME sector.

5. Discussion

This section primarily discusses the three research questions through the grounded findings.

5.1. Categorizing Required AI Capabilities and Barriers to Adopting AI in Chinese Apparel-Manufacturing MSMEs

This section addresses our first two research questions by refining the discussion of AI capabilities essential for MSMEs and the barriers to AI adoption. These insights further contribute to the innovation ecosystem framework, formed into two layers: Layer 1, which identifies AI capabilities required for MSMEs, and Layer 2, which examines the structural barriers impeding AI adoption.

Our findings reveal that AI is pivotal in enhancing MSMEs’ adaptive production capability and augmented human–AI collaboration, addressing their struggles with market volatility, skill shortages, and resource constraints. Adaptive production is crucial for MSMEs, given their reliance on small-batch, high-customization manufacturing. AI-driven predictive analytics and automated scheduling can mitigate inefficiencies caused by frequent design modifications, enabling dynamic adjustments to production workflows. By streamlining order management and material forecasting, AI allows firms to maintain agility in volatile markets. However, technological integration alone does not resolve industry complexities. The capacity for human–AI collaboration remains indispensable, as apparel manufacturing depends on human expertise in material handling, quality control, and customized garment production. While AI enhances precision in tasks like pattern cutting and fabric selection, it cannot fully replicate human adaptability in managing fabric inconsistencies and last-minute design changes. The respondents underscored AI’s role in reducing human error and improving operational efficiency, yet its effectiveness is contingent on a hybrid approach where AI complements human skills rather than replacing them. These two AI capabilities are interdependent. Adaptive production relies on AI tools embedded within human–AI collaboration systems, reinforcing the necessity for seamless integration of technology and human expertise to sustain innovation and efficiency in MSME operations.

Despite AI’s transformative potential, MSMEs face industry-, university-, and government-related constraints that hinder adoption. Industry constraints stem from high costs, technical complexity, and uncertainty regarding ROI. Many MSMEs lack the financial capacity for large-scale AI implementation, and even when resources are available, skepticism about AI’s applicability further discourages investment. The absence of standardized AI solutions tailored to apparel manufacturing exacerbates technical challenges, leaving firms to navigate adoption hurdles independently. University factors highlight persistent talent shortages and misalignment between academic training and industry needs. Evidence shows that current curricula emphasize design-oriented skills while neglecting AI applications in apparel production, so graduates lack the technical expertise required to integrate AI into manufacturing processes, forcing MSMEs to operate without the talent pool necessary for digital transformation. Government policies serve as both enablers and constraints. While national initiatives promote AI-driven industrial upgrading, regional disparities in implementation limit MSMEs’ access to targeted support. As a result, local policies tend to favor large enterprises, making it difficult for MSMEs to meet eligibility requirements for AI subsidies and further deepening structural inequalities within the innovation ecosystem.

Our results suggest that these barriers are interconnected, forming a reinforcing cycle. Talent shortages (pertaining to the university factor) compound technical adoption difficulties (pertaining to the industry factor), while limited financial support (pertaining to the government factor) prevents firms from effectively addressing either challenge. Addressing these constraints requires a coordinated approach in order to bridge technological gaps, strengthen industry–academia linkages, and implement inclusive AI policies that support MSMEs’ integration into the innovation ecosystem.

5.2. Developing an AI-Enabled Innovation Ecosystem Framework with Propositions for Chinese Apparel-Manufacturing MSMEs

The third research question examines how enterprises, universities, industry associations, and government collaborate to develop an AI-enabled innovation ecosystem in the Chinese manufacturing sector. To contextualize this framework, this section integrates insights from the innovation ecosystem and open-innovation literature, discussing AI’s role in facilitating cross-sector collaboration.

AI fosters interactions among key stakeholders by providing the technological infrastructure for knowledge exchange, resource coordination, and policy alignment [8,51]. These mechanisms align with Chesbrough’s (2003) open-innovation framework, where collaboration between external and internal actors drives technological advancement (Chesbrough, 2003, cited as in [52]). Building on Granstrand and Holgersson’s ([19], 2020) definition of innovation ecosystems—comprising actors, activities, artifacts, institutions, and relations—we formulate an AI-enabled innovation ecosystem framework grounded in empirical data. The 13 propositions developed in this study articulate the key interactions within this ecosystem, offering strategic directions for how AI-enabled collaboration can overcome systemic barriers and enhance firm-level innovation capacity. By strengthening firms’ knowledge absorption capacity, Layer 3 fosters the recognition, assimilation, and application of external AI-related knowledge [53]. Absorptive capacity emerges as a systemic outcome of ecosystem-wide collaboration, where government policies provide access to AI-related resources, universities supply technical expertise, and industry partnerships facilitate knowledge-sharing and competitive learning. This recursive interaction transforms external support mechanisms into firm-level AI capabilities, reinforcing a continuous cycle of technological advancement. The following sections examine the specific mechanisms (government policies, university contributions, and industry dynamics) that cultivate and leverage knowledge absorption capacity to operationalize the 13 propositions, overcoming AI adoption barriers and establishing a cohesive AI-driven innovation ecosystem for the Chinese apparel-manufacturing sector.

5.2.1. Mechanism 1 of Layer 3: Government Policies as Enablers for Collaboration

Government policies play a significant role in fostering AI adoption by shifting from a centralized policy-setting approach to enabling collaboration within the innovation ecosystem. China’s top-down policy has traditionally prioritized large-scale enterprises, exacerbating regional disparities in AI adoption [54]. While southern regions benefit from advanced infrastructure and industrial clusters, northern and less-developed areas face significant constraints in accessing financial and technical resources [55]. Establishing regional innovation hubs can mitigate these disparities by centralizing funding, technology, and expertise; facilitating cross-regional knowledge-sharing; and enhancing MSMEs’ competitiveness. Additionally, targeted funding mechanisms are crucial for supporting MSMEs, which struggle with high costs and limited access to subsidies. Unlike industries with longer product cycles, apparel MSMEs must adapt rapidly to changing trends [56], requiring flexible supply chain strategies and continuous investment [57]. Decentralized funding, including microgrants and low-interest loans, would enable MSMEs to experiment with AI adoption without disproportionate financial risks.

Local governments are also key facilitators of enterprise–university–industry collaboration. However, the government-related associations focus more on policy dissemination than fostering innovation. As MSMEs struggle to respond to fast-changing consumer trends, public–private partnerships can connect enterprises with academia and associations to promote joint innovation, including workshops, AI demonstration projects, and industry collaborations. This is especially relevant as the apparel sector transitions from original equipment manufacturer (OEM)/original design manufacturer (ODM) to original brand manufacturer (OBM) business models, requiring stronger branding, design capabilities, and consumer engagement strategies.

Despite their potential as intermediaries, industry associations remain underutilized in fostering AI adoption. Expanding their roles in capacity-building and knowledge-sharing would particularly benefit MSMEs that lack access to advanced technologies. Thus, redefining the government’s role as an enabler rather than a regulator is crucial to unlocking AI adoption. This leads to the following propositions:

P1. Establishing regional innovation hubs can reduce geographic disparities in AI adoption.

P2. Tailored funding mechanisms for MSMEs will enhance their AI adoption capacity.

P3. Public-private partnerships should align the goals of enterprises, universities, and associations.

P4. Local governments must act as facilitators of cross-sectoral collaboration.

P5. Industry associations should expand their roles to include capacity-building initiatives.

Beyond policy incentives and institutional arrangements, AI technologies also serve as functional enablers of integration within innovation ecosystems. In the context of Chinese apparel MSMEs, integration refers to the collaborative alignment of enterprises, universities, associations, and government through shared data systems, smart production platforms, and real-time communication interfaces. AI facilitates this alignment by supporting standardized digital infrastructures that enable cross-organizational coordination, such as automated supplier-matching algorithms, digital order tracking, and joint innovation databases. To ensure analytical rigor, our interview protocol included specific questions addressing the perceived changes in organizational boundaries, information exchange practices, and the role of AI in enabling inter-firm and cross-sector collaboration. These design choices were guided by this study’s aim to investigate not only the technological adoption of AI but also its institutional and relational consequences within the innovation ecosystem.

5.2.2. Mechanism 2 of Layer 3: Universities as Knowledge Hubs in Addressing Talent and Knowledge Gaps

Universities play a central role in addressing AI talent shortages in apparel manufacturing but remain underutilized due to misalignment between academic training and industry needs. Most university programs focus on traditional design and pattern-making, while the industry demands expertise in AI engineering, software development, and data analytics. Limited access to advanced manufacturing technologies in educational settings further delays graduates’ readiness for AI-related roles. To bridge these gaps, universities must reform curricula to integrate AI-specific interdisciplinary training and offer practical experience with intelligent manufacturing systems. However, current university–enterprise collaborations focus on short-term projects rather than long-term AI research. This prevents universities from addressing industry challenges like garment complexity and workforce readiness. Universities should participate in government-funded joint research and development (R&D) programs to co-develop AI solutions for industry-specific applications, including predictive analytics for supply chains and automated quality control systems. Additionally, establishing interdisciplinary AI innovation centers within universities would provide shared access to AI research and experimental facilities, fostering stronger collaboration across computer science, engineering, and apparel studies. Thus, transforming universities into active contributors to AI adoption requires systemic changes, leading to the following propositions:

P6. Universities must reform curricula to integrate interdisciplinary AI modules tailored to apparel manufacturing.

P7. Joint R&D programs should prioritize co-developing AI solutions for industry-specific challenges.

P8. Establishing interdisciplinary AI innovation centers will enhance AI-related research and training.

P9. University–enterprise partnerships should provide hands-on AI training to better prepare students for the workforce.

5.2.3. Mechanism 3 of Layer 3: Collaboration and Competition as Drivers of AI Adoption

AI adoption in apparel manufacturing is influenced by both collaboration and competitive pressures within the industry. Collaborative efforts are particularly evident in supply chain networks, where firms rely on AI-driven real-time inventory management and predictive analytics to optimize efficiency. Standardized data-sharing frameworks could enable better AI integration, ensuring seamless supply chain coordination and reducing inefficiencies. In addition to collaboration, competitive pressures also shape AI adoption. Firms that lead in AI adoption trigger a demonstration effect, prompting others to follow. However, intense price competition often prevents MSMEs from investing in AI, as low-profit margins force them to prioritize short-term cost reductions over long-term innovation. Balancing these competitive forces through industry-wide support mechanisms ensures market diversity and prevents MSME exclusion. Customer expectations also influence AI adoption, as AI-driven supply chain optimization enables faster delivery, higher-quality products, and greater customization. Firms leveraging AI for predictive analytics and automation can enhance customer responsiveness, strengthening long-term loyalty. However, disparities in AI access risk further widening the gap between large enterprises and MSMEs, requiring targeted policy and financial interventions to ensure equitable adoption. Therefore, the interplay between collaboration and competition necessitates systemic interventions, leading to the following propositions:

P10. Supply chain partners should implement standardized data-sharing protocols for AI integration.

P11. Firms leading in AI adoption should mentor smaller enterprises to promote equitable technology diffusion.

P12. To enhance efficiency, collaborative platforms should align supply chain operations with AI-enabled tools.

P13. Competitive pressures should be balanced with support mechanisms to prevent MSME marginalization.

5.2.4. Knowledge Absorptive Capacity as the Outcome of Layer 3 Interactions

The three mechanisms of government policies serving as enablers, universities serving as knowledge hubs, and the industry dynamics of collaboration and competition lay the foundation for the fourth mechanism, which is absorptive capacity—a firm’s ability to recognize the value of external knowledge, assimilate it, and apply it to create innovation [53,58]— which emerges as a systemic outcome of Layer 3 interactions. As we proposed, the Chinese government policies play a pivotal role by reducing structural barriers, such as geographic disparities and financial constraints, through regional AI innovation hubs and tailored subsidies for MSMEs. Hence, these initiatives provide firms access to external resources and technologies, creating opportunities for experimentation and engagement with AI solutions. As knowledge providers, universities bridge the gap between theoretical research and practical applications, equipping firms with the specialized skills and technical expertise required for AI adoption through joint R&D programs and interdisciplinary innovation centers. Meanwhile, industry dynamics foster knowledge absorption capacity by creating environments for shared learning and competitive benchmarking. Collaborative supply chain platforms and industry associations facilitate knowledge sharing, while competitive pressures incentivize firms to adopt and adapt AI innovations to maintain market relevance. These mechanisms enable firms to recognize, assimilate, and apply external knowledge, transforming barriers into adaptive capabilities. Thus, knowledge absorption capacity becomes both a product of Layer 3 interactions and a driver for sustained innovation within the ecosystem, reinforcing the recursive and dynamic nature of the innovation framework.

5.3. Theoretical Contributions

Within this framework, we have proposed 13 key propositions that describe the mechanisms through which AI facilitates collaboration across industries, universities, and government, thereby constructing the AI-enabled sustainable innovation ecosystem framework. The grounded innovation ecosystem framework developed through qualitative interviews offers significant theoretical contributions by advancing our understanding of how innovation ecosystems operate in emerging economies and traditional apparel manufacturing. While much of the existing literature on innovation ecosystems focuses on high-tech industries in developed economies, this framework broadens the scope by contextualizing AI adoption within a traditionally low-tech sector, emphasizing the unique pathways through which such industries can transform. A central contribution lies in its layered structure, which integrates required capabilities (Layer 1), barriers to adoption (Layer 2), and external collaborative mechanisms (Layer 3), offering a dynamic, multi-level perspective on how ecosystems evolve. This layered approach highlights the interplay between internal capabilities, such as adaptive production, and external systemic factors, such as workforce readiness and policy constraints, while also revealing the mechanisms by which external collaborations mitigate these challenges and enable knowledge absorption. Moreover, the framework reconceptualizes the role of governments in emerging economies, shifting from a top-down policy enforcement perspective to one that emphasizes enabling bottom-up initiatives and supporting decentralized innovation, especially for MSMEs. It also underscores the critical importance of interdisciplinary knowledge integration, revealing how ecosystems thrive by bridging disciplinary silos across fashion design, AI engineering, and supply chain management, facilitated by universities as key knowledge hubs. Furthermore, the framework positions knowledge absorption capacity not as a standalone firm-level construct but as a systemic outcome of collaborative mechanisms in Layer 3, illustrating how government policies, university contributions, and industry dynamics interact to enhance firms’ ability to assimilate and apply external knowledge. By contextualizing these interactions within China’s institutional environment, characterized by a blend of top-down state influence and market-driven dynamics, the framework provides a culturally specific lens that deepens theoretical insights into how ecosystems adapt in complex governance settings. Additionally, the framework enriches our understanding of the dual forces of collaboration and competition, showing how cooperation fosters shared learning and resource pooling. In contrast, competition drives innovation and benchmarking, achieving a balance that propels innovation.

5.4. Managerial Contributions

The sustainable innovation ecosystem framework developed in our study offers several key managerial contributions for stakeholders in the Chinese apparel-manufacturing sector. First, we provide actionable insights into how firms can leverage inter-organizational collaboration to overcome barriers to AI adoption. For managers, we highlight the importance of engaging with external actors, such as universities, suppliers, and the government, not only to access resources and knowledge but also to build long-term partnerships that enhance adaptive and collaborative AI capabilities. By understanding the roles and mechanisms outlined in Layer 3, firms can strategically position themselves within this ecosystem, proactively aligning their internal capabilities with external opportunities. Second, our framework underscores the need for managers to cultivate knowledge absorption capacity within their organizations. By fostering a culture of learning and openness, firms can better assimilate external knowledge and innovations, enabling more effective integration of AI technologies into their operations. Additionally, the emphasis on decentralized, bottom-up policy engagement suggests that managers should not passively rely on government support but actively advocate for their needs and participate in shaping supportive policies, particularly for MSMEs that often lack direct access to resources. Finally, our framework highlights the role of competitive benchmarking and collaborative experimentation in driving open innovation. Managers are encouraged to balance competitive pressures with opportunities for shared learning within supply chains and industry networks, fostering a culture where collaboration and competition coexist to accelerate innovation. By operationalizing these insights, managers can enhance their firms’ resilience and adaptability, ensuring sustainable growth in an increasingly AI-driven industrial landscape.

6. Conclusions, Limitations, and Future Research

In this study, we developed a framework and formulated propositions for the AI-enabled sustainable innovation ecosystem of MSMEs in the Chinese apparel-manufacturing sector, emphasizing the role of external collaborations in overcoming adoption barriers. The theoretical contributions of this research include advancing innovation ecosystem theory by bridging micro-level adoption determinants and macro-level systemic mechanisms. The framework enables firms to overcome barriers and build their knowledge absorption capacity through external collaborations, such as university–industry partnerships and supply chain integration. This study reconceptualizes the government’s role, shifting from top-down directives to facilitating collaborative innovation. From a managerial perspective, our findings highlight the importance of internal capability development alongside external ecosystem engagement. Also, policymakers should focus on creating regional innovation hubs, tailored funding mechanisms, and programs that address talent shortages and resource inequities. By aligning organizational efforts with ecosystem dynamics, stakeholders can accelerate AI adoption and foster sustainable innovation.

Despite its contributions, this study has several limitations. First, the grounded-theory approach requires extensive data collection and iterative coding, making theoretical saturation subjective. Future research should incorporate additional coding rounds and data triangulation to strengthen validity. Second, the sample size (15 interviews) may limit the diversity of insights, particularly regarding industry associations, as access was constrained. Expanding qualitative methods, including by incorporating focus groups, policy text analysis, and industry panel data, could provide a more comprehensive understanding of AI adoption dynamics. Third, while this framework offers theoretical insights, it has not yet been empirically tested within a firm. Future research should apply it in case studies or longitudinal projects to examine its effectiveness and how adoption barriers evolve. Survey-based quantitative studies can also test the relative influence of each enabling factor across different firm types or regional policy environments. Fourth, while the propositions (e.g., P1, P2, and P3) address China-specific challenges, they may be relevant to other industries facing regional disparities, financial constraints, and collaborative innovation needs. Future research should empirically validate these propositions in diverse sectors to assess their applicability. Also, given the breadth of stakeholders involved and the layered nature of the framework, the scope of these propositions may appear expansive. Future research could apply the model to more narrowly defined regional clusters or specific sub-sectors within apparel manufacturing to validate and refine the collaborative mechanisms proposed. Lastly, this framework has also not been tested in other developing economies (e.g., Vietnam, India, Indonesia, and Bangladesh), where MSMEs, government-driven industrial policies, and AI adoption initiatives share similarities with China. Comparative studies across these economies would help refine the framework, ensuring its adaptability to varying government support structures, industrial maturity, and digital transformation strategies. By addressing these limitations, future research can enhance the robustness, applicability, and empirical validation of AI-enabled innovation ecosystems in MSMEs beyond the Chinese apparel-manufacturing sector.