In the field of digital technology R&D, AI is becoming deeply embedded in researchers' digital innovation processes, enhancing innovation efficiency through human-AI collaboration. However, growing concerns have emerged over excessive dependence on AI. In some firms, such dependence has led to creative homogenization, core data leaks, and new product failures. This raises a critical question: does AI dependence stimulate or undermine R&D personnel's digital innovation vitality? Existing research has two main limitations. First, most studies examine how AI dependence affects traditional innovation activities, with limited attention to digital innovation, where non-rivalry and other features may amplify or alter its effects. Second, prior research treats AI dependence as a homogeneous construct, overlooking its internal heterogeneity and failing to distinguish between exploitative and exploratory digital innovation. Moreover, existing studies tend to focus on single mediation pathways, with limited insight into the chain through which motivational differences translate into cognitive and behavioral changes. To address these gaps, this study conceptualizes AI dependence as two dimensions: functional and existential dependence. Drawing on self-determination and conservation of resources theories, it incorporates information confrontation and knowledge integration as potential chain pathway and develops a "motivation-cognition-capability-behavior" framework. It examines how different forms of AI dependence affect R&D personnel's exploitative and exploratory digital innovation, while introducing AI literacy as a contextual factor and examining its moderating role. The study aims to provide theoretical insights and practical implications for digital innovation R&D management in the age of AI. The study yields four main findings. First, functional dependence significantly positively promotes both forms of digital innovation, whereas existential dependence inhibits them. Second, information confrontation intensity and knowledge recombination capability mediate the relationship between AI dependence and ambidextrous digital innovation, forming a sequential "cognition-capability" pathway. Third, AI literacy positively moderates the relationship between functional dependence and information confrontation intensity, thereby strengthening the chain mediation effect. However, its moderating effect on the existential dependence pathway is not significant. Fourth, subgroup analysis based on AI use shows that AI tool diversity strengthens the effect between functional dependence and exploratory digital innovation, but not other pathways. This study contributes to the literature in four ways. First, it deepens our understanding of how AI dependence affects ambidextrous digital innovation. By distinguishing between functional and existential dependence, it reveals their differential effects on ambidextrous digital innovation, contributing to the "empowerment vs. disempowerment" debate. Second, it extends self-determination theory. Within this framework, it identifies the sequential mediating roles of information confrontation intensity and knowledge recombination capability, explaining how motivational differences translate into cognitive and behavioral outcomes. Third, it clarifies the contextual boundaries of AI dependence effects. By examining the moderating effect of AI literacy and the subgroup-specific effect of AI use quantity, it finds that both effects are significantly asymmetric, suggesting that the boundary conditions are path-specific and context-specific. Fourth, it enriches research on the antecedents of ambidextrous digital innovation at the individual level by focusing on AI dependence and revealing how it shapes R&D personnel's exploitative and exploratory digital innovation.

The construction of world-important talent centers and innovation hubs is a strategic priority of China's talent-power strategy, yet how national talent endowments can be transformed into high innovation performance remains unresolved. Based on national innovation systems theory and the talent innovation and entrepreneurship ecosystem theory, this study develops a four-layer model of the national talent innovation and entrepreneurship ecosystem, comprising talent resources, innovation-supporting entities, key innovation resources, and an open and livable environment, and examines, from a configurational perspective, how these conditions jointly shape ecosystem effectiveness. This study applies fuzzy-set qualitative comparative analysis (fsQCA) to a cross-sectional sample of 94 countries drawn from the 2025 Global Innovation Index, the 2025 Global Talent Competitiveness Index, and the 2025 International Statistical Yearbook. Seven antecedent conditions are operationalized: talent resources, knowledge innovation entities, technological innovation entities, financial resource support, government policy support, openness to the outside world, and livability. The outcome is measured by the knowledge-creation score of the Global Innovation Index, capturing a country's talent-driven innovation effectiveness. The findings unfold in three steps. First, necessity analysis reveals that no single condition is necessary for high effectiveness, whereas the absence of technological innovation entities emerges as a quasi-necessary condition for low effectiveness, indicating that the lack of such innovation carriers is the most prevalent deficiency among underperforming countries. Second, configurational analysis identifies two equifinal pathways to high effectiveness, with an overall consistency of 0.898 and coverage of 0.568. Path S1, a talent-supported "knowledge innovation entity, government policy, open and livable environment" mode, characterizes 25 mostly small and medium-sized advanced economies such as Finland and Singapore, in which strong universities and stable policies combine with openness and livability to sustain innovation. Path S2, a talent-supported "technological innovation entity, key innovation resources, open and livable environment" mode, characterizes 12 large developed economies such as the United States, Germany, and the United Kingdom, in which dense corporate innovation carriers couple with developed financial systems and global talent networks. Talent resources, external openness, and livability constitute core conditions shared by both pathways, while knowledge innovation entities and the technological-innovation and financial-resource bundle exhibit a mutually substitutive, cross-configurational distribution. Third, the analysis of low-effectiveness configurations yields five paths classifiable into an innovation-foundation-weakness pattern, anchored in the joint absence of talent resources and technological innovation entities, and a resource and actor mismatch pattern, in which strong financial resources coexist with missing innovation entities and inadequate livability, as exemplified by resource-dependent economies such as Saudi Arabia. Robustness checks across alternative consistency thresholds, PRI thresholds, and calibration anchors confirm the stability of these results. This study makes two theoretical contributions. It extends innovation ecosystem research from the organizational and regional levels to the national level, demonstrating that the effectiveness of national talent ecosystems arises from multi-actor synergy rather than from the endowment of any single factor. It also uncovers the contingent role of technological innovation entities, challenging the practice of equating firm size with innovation capacity. The findings also carry practical implications. Countries should coordinate talent development, external openness, and livable-environment construction as an integrated undertaking, choose differentiated development modes according to their endowments, and avoid mistaking capital investment for innovation-capacity building when fundamental actors are missing. The results offer policy references for China's efforts to build world-important talent centers and innovation hubs, and for other economies seeking to enhance their national innovation ecosystems.

Against the backdrop of China's national strategy to promote integrated innovation between large enterprises and small and medium-sized enterprises (SMEs), asymmetric innovation cooperation has emerged as a core carrier of industrial innovation synergy. This cooperation mode is defined by distinct power and information asymmetry: leading enterprises dominate decision-making with abundant resources and highly public information, while follower SMEs hold localized technological comparative advantages yet face low observability of their core innovation capabilities. Nevertheless, the pre-cooperation screening mechanism under such an asymmetric structure remains largely underexplored. Existing studies from resource complementarity, technological proximity, and institutional perspectives often yield conflicting linear conclusions and assume equal evaluation power, which deviates significantly from real-world practices. To fill this research gap, this study draws on signaling theory and constructs a "detection-decoding" dual-process model to investigate how follower firms' technology differentiation influences the formation of asymmetric innovation cooperation through signal quality and signal visibility. This study adopts a sample of Chinese A-share listed companies covering the period from 2012-2023. The core explanatory variable of technology differentiation is measured using patent abstract data and natural language processing techniques. A professional technical terminology lexicon is built, the TF-IDF method is applied to construct text vectors of enterprise technology portfolios, and each firm's relative uniqueness in the overall industry technology space is calculated, which is an approach that captures micro-level technological path differences more precisely than traditional IPC classification methods. The conditional Logit model serves as the baseline analytical framework, and pseudo-cooperation pairs are constructed via propensity score matching to estimate cooperation probability. Three moderating variables are introduced to test boundary conditions: signal environment (proxied by the number of investor online inquiries for follower firms), market overlap (calculated by business scope text similarity between pairs), and leading enterprises' innovation capability (measured by the output-input ratio of innovation activities). Robustness tests, including variable replacement, sample period adjustment, and pseudo-pair reduction, consistently support the baseline findings. The empirical results demonstrate three key findings. First, technology differentiation has a significant inverted U-shaped effect on asymmetric innovation cooperation. In the low-differentiation range, the positive screening effect of signal quality dominates: rising differentiation signals higher R&D commitment and unique innovation potential, thus increasing cooperation probability. After passing the optimal threshold, the negative attenuation effect of signal visibility takes over: excessive deviation from the mainstream technology track widens cognitive distance and amplifies environmental noise, making signals undetectable by leading enterprises and reducing cooperation likelihood. Second, the three moderating variables exert heterogeneous effects. A favorable signal environment alleviates the visibility attenuation of high-differentiation signals and flattens the inverted U-shaped curve. Higher market overlap both shortens signal transmission distance in the detection stage and reduces decoding difficulty in the decoding stage, strengthening the inverted U-shaped relationship and shifting the overall curve upward. Stronger innovation capability of leading enterprises improves signal decoding accuracy and risk tolerance, also reinforcing the inverted U-shaped relationship. Third, the core logic of asymmetric innovation cooperation screening lies in leading enterprises' signal discrimination and cognitive processing under high uncertainty, rather than simple resource complementarity matching. Many high-value SMEs fail to enter cooperation not due to insufficient technological value, but because their signals are not effectively detected or accurately decoded. This study makes three major theoretical contributions. First, it embeds signaling theory into asymmetric innovation cooperation, clarifying the signal-based essence of partner screening by leading enterprises and addressing existing scenario deviation. Second, the proposed "detection-decoding" dual-process model elevates signal visibility to a core dimension alongside signal quality, expanding the traditional cost-centric framework and opening the black box of receivers' cognitive processing. Third, it identifies the heterogeneous boundary roles of signal environment, market overlap, and receiver innovation capability, enriching the understanding of receiver heterogeneity. The findings offer practical implications for leading enterprises to optimize screening systems, for follower SMEs to balance technological differentiation and cognitive compatibility, and for policymakers to build signal dissemination platforms and reduce information asymmetry in industrial innovation ecosystems.

The deep integration of new-generation information technologies with the manufacturing industry, exemplified by industrial Internet platforms (IIPs), has become an important force driving industrial transformation and upgrading. Although leading enterprises have achieved efficiency gains through digital transformation, many small- and medium-sized manufacturing enterprises (SMEs) still struggle to translate IIP adoption into measurable improvements in production efficiency. This raises an important question: under what configurations of conditions can resource-constrained SMEs leverage IIPs to improve production efficiency? Answering this question is essential for promoting the high-quality development of the manufacturing sector. Existing studies have predominantly focused on large incumbents or IIP technical architecture, neglecting heterogeneous outcomes among SMEs. IIP adoption is a complex strategic process shaped by technological, organizational, and environmental conditions. Moving beyond linear “net-effect” logic, this study adopts a configurational perspective to identify multiple equifinal pathways through which IIPs improve SME production efficiency, and examines how platform technology, organizational characteristics, and external environment interact to generate high efficiency. Grounded in the technology – organization – environment (TOE) framework, this study constructs an analytical model with three dimensions. Platform technology is represented by IIP capability. User organization is reflected in SMEs’ profitability, resource allocation capability, and managerial digital orientation. Platform environment is characterized by the regional industrial Internet ecosystem and market competition intensity. To test the model, this study employs fuzzy-set qualitative comparative analysis (fsQCA). The sample consists of 477 Chinese SMEs that have adopted IIPs, identified from the “2024 Top 100 Industrial Internet Platforms” and the “2024 Top 50 Featured and Specialized Industrial Internet Platforms.” Data on antecedent conditions were collected for 2022 from the official websites of the platforms, corporate annual reports, and the CSMAR database, with expert scoring used to evaluate platform capability. The outcome variable, production efficiency in 2023, was measured using the SBM-DEA model based on an input – output indicator system. The one-period lag between antecedents and the outcome helps establish temporal precedence. The results reveal four configurations leading to high production efficiency, and these are classified into three archetypal paths. First, the platform resource-deepening pathway, represented by Configurations S1a and S1b, emphasizes the deep coupling between strong platform capability and strong firm-level resource allocation capability. It reflects an internally driven growth model in which SMEs use IIPs to optimize existing resources. In this pathway, a strong external ecosystem is not indispensable; the strategic core lies in the joint effect of platform capability and internal resource allocation capability. Second, the platform ecosystem inclusive pathway represented by Configuration S2, reflects a compensatory mechanism. Firms with relatively weak resource allocation capability can still achieve high production efficiency by embedding themselves in a well-developed IIP ecosystem. By accessing platform-based data, algorithms, industrial collaboration, and resource spillovers, such firms compensate for internal capability deficiencies. Third, the market pressure-driven pathway, represented by Configuration S3, indicates that intense market competition can serve as a key external trigger. SMEs with strong profitability and resource allocation capability can respond rapidly to competitive pressure and optimize internal processes. In this configuration, high platform capability is not necessary; efficiency improvement is mainly driven by organizational agility and financial resilience under competitive pressure. The analysis of non-high production efficiency configurations further corroborates these findings by identifying two constraints: strategic capability imbalance and resource poverty. This study makes three contributions. First, it shifts the focus from large enterprises to SMEs, enriching contextual understanding of IIP-empowered transformation. Second, it reveals the equifinal, configurational nature of IIP empowerment, demonstrating that no universally optimal pathway exists. Third, it highlights the active role of user firms, moving beyond the technology-centric view emphasizing platform functions alone. The findings also offer practical implications. Policymakers should shift from encouraging platform access to supporting effective platform use through differentiated policies, including data governance, model sharing, service vouchers, cloud credits, targeted subsidies, and talent incentives. SME managers should adopt a configurational mindset, identify their resource endowments and external conditions, and strategically combine internal capabilities with platform resources and environmental opportunities.

The widespread adoption of Generative Artificial Intelligence (GenAI) is fundamentally reshaping knowledge work. While GenAI enhances human capabilities, it also raises concerns about skill redundancy, role ambiguity, and heightened job demands, potentially hindering sustainable human-AI collaboration. Effective work design is crucial for unlocking the synergistic potential of human-AI teams. However, existing work design models, developed primarily for traditional human-human collaboration, tend to describe work characteristics through vague interpretations or one-dimensional listings. This approach fails to accurately capture the distinctions between knowledge workers and ordinary workers, nor does it encompass the multiple factors that influence work experience and outcomes in the GenAI context. To consolidate fragmented research and provide scientifically grounded guidance for organizations in transition, this study systematically identifies and conceptualizes the structure of work characteristics specific to knowledge workers collaborating with GenAI. Adopting a constructive grounded theory approach, the study builds theory inductively from rich qualitative data. The primary data came from semi-structured interviews with 20 knowledge workers, all of whom had at least three months of GenAI experience. During the interviews, leading questions with outcome biases were avoided, and reasonable compensation commensurate with their time was provided upon completion. To enhance data diversity, an open-ended questionnaire was used as a supplementary method, with question design aligned with the interview protocol. From this, 40 valid responses were rigorously screened. In total, over 300 000 words of valid textual data were obtained and subjected to multiple rounds of systematic coding using NVivo 15. Three-fourths of the data were used to construct the theoretical framework, and the remaining one-fourth served to test theoretical saturation. The findings reveal that the work characteristics of knowledge workers in the GenAI context form a multi-level model comprising four higher-order dimensions, which encompass twelve specific facets. Contextual characteristics include GenAI facility conditions and GenAI risk review. Task characteristics consist of GenAI-enhanced autonomy, feedback from GenAI, role overload, and role ambiguity. Knowledge characteristics encompass GenAI-driven professionalism, GenAI-related skill expansion, GenAI information processing demands, and non-routine problem solving. Social characteristics involve interpersonal support for human-AI collaboration and self-management support needs. In the analysis of the internal logic of the model, contextual characteristics provide the foundational infrastructure for the work system. Knowledge and task characteristics interact reciprocally, forming the core operational mechanism. Social characteristics ensure the system functions effectively by providing necessary support. Within the knowledge dimension, GenAI-driven professionalism acts as the central hub, which directly determines employees' proficiency in mastering GenAI-related skills, the accuracy and quality of their verification and interpretation of GenAI-generated information, and their competence in solving non-routine workplace problems. Within the task dimension, a dual-path dynamic exists: GenAI-enhanced autonomy and feedback constitute an empowering pathway, while role ambiguity and overload form a constraining pathway, with both interacting dynamically. The primary theoretical contribution of this study lies in developing a comprehensive, context-specific structural model that extends classic work design frameworks into the GenAI era. It advances the conceptualization of work characteristics beyond mere listings by elucidating their nuanced connotations and complex interdependencies within human-AI collaboration. For practice, the model guides managers in introducing knowledge workers to the new work characteristics emerging in the GenAI context, prompting them to reflect on their own strengths and weaknesses in relation to these characteristics, and assisting in the formulation of personalized development plans. Moreover, managers should recognize the inherent interconnections among the various dimensions of work characteristics and leverage their synergistic configurations to enhance both organizational performance and employee well-being. Future research should develop validated measurement scales for the identified dimensions and test the proposed relationships across diverse occupational and cultural settings.

Core technologies in key fields serve as the cornerstone of national competitiveness, fundamentally driving new growth momentum and innovation-led development. Consequently, technological blockades targeting key core technologies have intensified, significantly impeding China's pursuit of independent and controllable technological development. Against this backdrop, leveraging digital innovation to secure a competitive advantage in global manufacturing and close the frontier technology gap has become imperative for building a manufacturing powerhouse and accelerating highlevel technological selfreliance. Manufacturing enterprises need to actively engage in digital technology cooperation, accelerate the formation of a new open innovation model through the reform of data element marketization, and safeguard national scientific and technological security as well as the initiative in economic development. The breakthrough in core technologies lies at the heart of enhancing technological innovation capability. This study focuses on the core issue of "how the marketization of data elements promotes key core technology innovation in manufacturing enterprises," with the underlying logic that data element marketization drives digital transformation and reduces corporate R&D investment. To identify the causal effect, this study employs the staggered rollout of citylevel data trading platforms as a quasinatural experiment. The timing of these platforms' establishment is driven by regional digital infrastructure and policy readiness rather than firmspecific innovation prospects, providing plausibly exogenous variation well suited to a staggered difference-in-differences (DID) framework. Firms in cities with an operational platform are assigned to the treatment group, while those in cities without a platform serve as the control group, with treatment timing aligned to the year of platform inauguration. Within this framework, the study selects A-share listed manufacturing enterprises from 2014 to 2023 as the research sample. Financial and governance data are drawn from CSMAR, CNRDS, corporate annual reports, and the National Bureau of Statistics; patent data are sourced from the China National Intellectual Property Administration (CNIPA). All continuous variables are winsorized at the 1st and 99th percentiles, yielding 9 394 firm-year observations. The study then systematically analyzes and explores the influence mechanisms and pathways through which data element marketization affects manufacturing enterprises' breakthroughs in core technologies in key fields. Empirical analysis shows that promoting the marketization of data elements effectively eliminates information barriers and resource exclusivity, and significantly enhances manufacturing enterprises' key core technology innovation. Indepth mechanism tests reveal that digital transformation and upgrading, as well as enterprise R&D investment, are the key transmission channels for the above effects. Extended research also indicates that enterprises' absorptive capacity and their perception of environmental uncertainty strengthen, to a certain extent, the promoting effect of data element marketization on enterprises' key core technological innovation capability. This study focuses on the core question of how data element marketization promotes key core technology innovation in manufacturing enterprises, and systematically discusses it from both theoretical and empirical levels, carrying important academic value and practical significance. First, at the theoretical level, this study fills the gap in the existing literature on the relationship between data element marketization and manufacturing enterprises' key core technology innovation. Second, at the methodological level, this study breaks through the linear assumption of the "datainnovation" relationship in traditional research, introducing the moderating variables of enterprise absorptive capacity and environmental uncertainty perception, thereby enriching the analytical dimensions of the relationship between data elements and enterprise innovation. Finally, at the practical level, this study reveals the heterogeneous impacts of data element marketization on manufacturing enterprises' key core technology innovation, providing more targeted guidance for policy makers.

As large language models shift from content generation to task execution, a new product form has emerged in mobile terminals and personal computing environments: the system-level AI agent. Unlike plug-in assistants embedded in a single application, system-level agents operate at the operating-system layer. They can read screen content, identify interface states, invoke tools, and perform cross-application actions through simulated clicking, typing, and task chaining. This transition moves human-computer interaction from an application-centered mode to a task-centered mode. It also restructures the way data are accessed, circulated, and acted upon. Against this background, data security risks are no longer confined to one-off collection or isolated leakage. They become chain-based, cumulative, and processual, extending across data acquisition, data circulation, and task execution. The study is motivated by both industrial practice and regulatory signals. Products such as Doubao Mobile Assistant have demonstrated cross-application execution in searching, comparison shopping, ordering, and message coordination. OpenClaw has further shown how persistent memory, browser automation, file management, and script execution can be incorporated into open ecosystems. At the same time, regulators in China, including the MIIT threat and vulnerability information sharing platform (NVDB), CNCERT, and industry associations in the financial sector, have issued risk warnings concerning excessive privileges, weak default security configurations, prompt injection, malicious plugins, abnormal account control, and execution errors. These developments indicate that the problem has moved beyond a technical experiment and become a concrete governance issue. The research design is anchored in a lifecycle-based analytical framework. Drawing on the categories of the Data Security Law, the study reconstructs system-level AI agent data processing into three linked layers: acquisition, circulation, and execution. On this basis, the article combines doctrinal analysis with governance-oriented institutional analysis. It examines how system-level permissions, end-cloud collaboration, tool invocation, and continuous task execution reshape the boundaries of data collection and the structure of responsibility. It also reviews existing scholarship from three strands—technical security studies, personal-information and data-compliance research, and platform-governance studies—to show that current discussions remain fragmented and do not fully capture the integrated risk structure of system-level AI agents. The study reaches three main findings. First, system-level AI agents create a full-chain risk structure. At the acquisition layer, full-domain sensing and cross-application access turn local exposure into overall exposure and make over-collection and inadvertent collection more likely. At the circulation layer, end-cloud collaboration, plugin calls, and context retention generate hidden circulation risks, including re-identification after "desensitized" upload, contextual leakage, and ecological spillover across platforms. At the execution layer, automated agency amplifies the risks of unauthorized actions, cumulative deviations in multi-step task chains, and black-box loss of user correction capacity. Second, existing governance mechanisms are misaligned at three levels. Technically, the system is often operable but insufficiently visible, controllable, and verifiable. Normatively, responsibility attribution, the minimum-necessity principle, and informed consent rules are difficult to apply in multi-actor, cross-application, and continuous-execution environments. Economically, high compliance costs, strong incentives for data expansion, and the externalization of losses weaken firms' incentives to invest in sustained safety governance. Third, an effective response requires a three-tier governance framework rather than isolated fixes. The main contribution of this study lies in proposing a governance structure based on the sequence of rule-based boundary setting, technical embedding of constraints, and implementation support. On the rule side, the study argues for clarifying the boundaries of high-risk data, high-risk permissions, and high-risk tasks through the adjustment of the minimum-necessity principle, the optimization of informed consent, and the reconstruction of responsibility attribution. On the technical side, it proposes embedding those boundaries into system operation through visibility, controllability, and verifiability mechanisms, including layered tracking, dynamic markings, graded authorization, anomaly interruption, deletion proofs, and auditable records. On the implementation side, it recommends graded market access, pre-launch assessment, responsibility internalization, and risk-sharing mechanisms so that governance is not left to voluntary optimization. In this way, this study moves beyond traditional app-based or static compliance models and provides a more targeted framework for balancing operational efficiency with data security in the age of system-level AI agents.

Manufacturing is the core sector of the national economic system and a key indicator of a country's comprehensive competitiveness. Since the reform and opening-up, China's manufacturing industry has made remarkable progress, yet the issue of being "large but not strong" persists. Its long-formed growth path is notably extensive, with prominent problems including low energy efficiency and a large surplus of low-end production capacity. Innovation is undoubtedly the core driver of transformation, and green innovation serves as a crucial engine for the sector's green transition. Helping Chinese manufacturing enterprises build a sustainable green ecosystem and advance high-quality green innovation has become a critical path for safeguarding the security of China's manufacturing industry and achieving the carbon peaking and carbon neutrality goals. To facilitate corporate green transformation, the government has launched a series of credit support policies at the fiscal level, guiding banks and other financial institutions to optimize credit structures and provide financial guarantees for enterprises' green transition. As corporate credit utilization links financial resource allocation and innovation, and Short-Term Debt for Long-Term Use is a typical issue in Chinese enterprises' investment and financing, it is necessary to systematically examine how credit utilization influences green innovation quality. This study takes China's A-share listed manufacturing enterprises as the research object, with the sample period ranging from 2010 to 2021. The reason why the cutoff year is set at 2021 is that the indicator measuring green innovation quality adopted in this study is derived from the citation counts of enterprises' invention patents by other entities, and the citation information in this database is updated to 2024. Therefore, on the premise of ensuring a three-year observation window, the research sample is accordingly intercepted up to 2021. The main research procedures are as follows: first, constructing a double machine learning causal inference model to empirically test the impact of Short-Term Debt for Long-Term Use on the green innovation quality of manufacturing enterprises; second, focusing on the mechanism of Short-Term Debt for Long-Term Use affecting the green innovation quality of manufacturing enterprises, investigating the mediating effects of ESG performance, financing constraints and human capital structure, as well as the moderating effect of industry competition; finally, conducting heterogeneity analysis based on differences such as the city tier where different enterprises are located and enterprise attributes to put forward targeted policy recommendations. The study finds that Short-Term Debt for Long-Term Use significantly inhibits the green innovation quality of manufacturing enterprises, and this conclusion still holds after a series of robustness tests. Mechanism analysis shows that Short-Term Debt for Long-Term Use inhibits the green innovation quality of manufacturing enterprises by weakening corporate ESG performance, exacerbating financing constraints and reducing the level of human capital structure, while the degree of industry competition strengthens its inhibitory effect on green innovation quality. Heterogeneity analysis shows that the inhibitory effect of Short-Term Debt for Long-Term Use exerts a significant inhibitory effect on the green innovation quality of manufacturing enterprises in non-state-owned, non-heavily polluting and high-tier cities. Its inhibitory effect on manufacturing enterprises in heavily polluting cities and general-tier cities is relatively weak, while no significant impact is observed on state-owned manufacturing enterprises. This study helps deepen the understanding of the effects of financial resource allocation, expand the existing research perspectives on green innovation, and clarify the mechanism of corporate green innovation behavior. It proposes that, first, government should optimize the financing supply structure by increasing the share of medium- and long-term credit in manufacturing firms, while improving bank assessment mechanisms to mitigate the over-reliance on short-term debt. Second, ESG performance should be explicitly linked to credit pricing and quotas, complemented by green tax incentives and talent subsidies to support green transitions. Third, differentiated policies should target non-state-owned, non-heavy-polluting, and high-tier city firms to prevent low-quality green innovation, by establishing quality-oriented evaluation mechanisms tied to financing access.

The rapid iteration and full-scenario penetration of artificial intelligence (AI) have made endogenous value misalignment a critical global governance challenge. While major economies have adopted risk-based classification frameworks to balance innovation and risk control, China's existing governance systems primarily focus on physical safety and personal rights violations, failing to address the concealment, transmissibility and systemic impact of value misalignment risks. This structural flaw has produced a persistent regulatory paradox: high-risk systems remain under-regulated while low-risk applications face disproportionate compliance burdens, with legal frameworks and technical mechanisms operating in mutual isolation. This study aims to construct a precise AI governance framework tailored for value alignment, using risk classification as the core analytical lens and dual-dimensional legal-technical regulation as the implementation path. By employing normative analysis, comparative study and case analysis, the study examines regulatory practices in the EU and the US to identify transferable insights, and draws on authoritative data from the China Academy of Information and Communications Technology (CAICT) 2024 AI Governance Blue Book and the Cyberspace Administration of China (CAC) 2025 "Qinglang" Special Action Report to analyze compliance capacity gaps between leading enterprises and SMEs, as well as limitations of current regulatory mechanisms. The study first theoretically justifies the inherent coupling between risk classification and value alignment governance, demonstrating that a proportionality-based tiered model is the fundamental solution to current regulatory imbalances. It then identifies core obstacles: at the legal level, the absence of a unified national risk classification framework, fragmented sectoral regulations and homogeneous compliance requirements; at the technical level, the lack of standardized tiered specifications, uneven enterprise capabilities and insufficient full-lifecycle monitoring. The root cause is the structural exclusion of value misalignment risk from core evaluation dimensions in existing AI risk classification systems. To address this gap, the study develops a novel four-level AI risk classification framework for value alignment, based on a dual-dimensional indicator system measuring both the harm and probability of value misalignment. The framework defines clear criteria, typical scenarios and core requirements for each risk level, ranging from legally prohibited unacceptable risks to industry self-regulated low risks. Three supporting mechanisms of dominant risk determination, overlapping obligation application and dynamic adjustment with flexible exemptions ensure the operability of the system. Building on this framework, the study designs a tiered dual-dimensional governance system. For legal regulation, it establishes gradient compliance obligations: rigid prohibitions and full-chain accountability for unacceptable risks, full-lifecycle strict regulation and mandatory assessment for high-risk systems, targeted behavioral regulation and transparency obligations for medium-risk systems, and industry self-regulation with exemptions for low-risk systems. For technical regulation, it proposes differentiated full-lifecycle control measures, from closed physical isolation for unacceptable risks to voluntary compliance guidelines for low-risk applications. Finally, four synergistic mechanisms facilitate bidirectional interaction between legal and technical governance: standard coordination, information sharing, responsibility matching and remedy integration. The analysis yields three conclusions. First, a value alignment-oriented risk classification system is a prerequisite for transcending the "one-size-fits-all" approach and achieving regulatory precision. Second, the bidirectional co-construction of legal and technical regulation serves as the central mechanism for realizing AI value alignment, with legal norms defining operational boundaries and technical measures ensuring effective implementation. Third, adhering to proportionality in tiered governance is essential for balancing AI innovation and risk prevention. This study advances three key contributions. First, it innovatively incorporates value misalignment as the core evaluation dimension of AI risk classification, bridging the gap between general risk frameworks and value alignment objectives. Second, it establishes a differentiated dual-dimensional regulatory system anchored in risk classification, providing a systematic solution to current governance imbalances. Third, it clarifies the synergistic logic between legal and technical regulation and constructs four supporting mechanisms, forming a complete implementable governance framework. Limitations include the lack of in-depth industry-specific empirical research and quantitative validation of classification indicators, which will be addressed in future work.

How latecomer countries can achieve technological catch-up has long been a key concern across various sectors of society. Against the backdrop of the deepening division of labor in global innovation and the increasingly networked nature of innovation activities, proactively integrating into the global innovation network (GIN) has become an important pathway for latecomer firms in open economies to break through technological bottlenecks, enhance innovation capabilities, and achieve leapfrog catch-up. Given this historical context, systematically exploring how Chinese enterprises can leverage the GIN to facilitate technological catch-up holds substantial theoretical and practical significance. Enterprises entering international markets often face high entry costs and resource constraints; as a result, not all firms can participate directly in global innovation activities or benefit from them. In this context, for the vast number of domestic firms unable to directly engage with the GIN, effectively acquiring advanced international knowledge and continuously improving their innovation capabilities are critical to their survival, development, and competitiveness. Currently, the domestic production system is becoming increasingly complex with a deepening division of labor, and supply chain networks have emerged as important carriers of knowledge diffusion. This leads to the core question of this paper: Can "star firms" that pioneer integration into the GIN propel local firms toward technological catch-up through supply chain linkages? What are the underlying mechanisms of this effect? Drawing on a dataset of 160 million patent records and a longitudinal complex network analysis of the GIN from 2009 to 2023, this study empirically investigates the impact of customer integration into the GIN on suppliers' technological catch-up, using data from Chinese A-share listed firms. The results show that customers' integration into the GIN significantly promotes their suppliers' technological catch-up. Mechanism analysis indicates that knowledge spillover effects, improved information accessibility, and enhanced supply chain stability play important roles in this process. Heterogeneity analysis reveals that this promoting effect is more pronounced in firms with high dynamic capabilities, lower information asymmetry, and higher supply chain concentration. The contributions of this study are mainly reflected in the following aspects: First, by taking the perspective of the GIN, it extends the research boundary of technological catch-up by latecomer firms in open economies. Second, from a supply chain spillover perspective, it reveals the mechanism through which customers' integration into the GIN promotes suppliers' technological catch-up, offering rich theoretical implications. Third, using massive global patent data, it measures, at a fine-grained level, the degree of firms' integration into the GIN and their technological catch-up performance, providing a new reference for related empirical research. The policy implications of this study are as follows: First, policymakers should accelerate enterprise integration into the GIN by encouraging local firms to establish strategic cooperation with foreign enterprises possessing core technologies, thereby enhancing their capacity to access and leverage global innovation resources. Second, they should promote synergistic development between local supply chain networks and the GIN by leveraging the unified national market to break down administrative barriers and industry monopolies and by deploying digital technologies to remove information barriers and facilitate knowledge diffusion; this enables enterprises to integrate indirectly into the global innovation system through local supply chains, thereby helping more local firms achieve technological catch-up. Third, they should adopt firm-specific policies and provide precise guidance tailored to enterprise characteristics. This includes encouraging firms to systematically develop dynamic capabilities to convert open resources in the GIN into tangible innovation outcomes, accelerating digital infrastructure construction to break down information barriers across supply chain links, and guiding highly concentrated enterprises to prevent external risks from over-reliance on core partners and to deepen technological cooperation, which transforms stable supply chain relationships into sustained momentum for technological catch-up.

As one of the new productive forces in the era of digital intelligence, computing power is regarded as an important internal driving force to promote the digital transformation of economy and society. With the in-depth application of computing resources, the development model of key areas such as artificial intelligence, government services, and biopharmaceuticals has been reshaped, further changing the production and lifestyle. However, while empowering digital construction, computing resources also bring monopoly worries that cannot be ignored. Against this backdrop, this paper explores the theoretical connotation, risk patterns, and governance challenges of computing power resource monopoly, aiming to build an inclusive, effective, and pragmatic governance order. This study first clarifies the theoretical connotation of computing power resource monopoly, categorizing it into "computing power production resource monopoly" and "computing power scheduling resource monopoly". In terms of the generation mechanism of computing power resource monopoly, the study identifies that the high input, technical standards and ecological lock-in effects of the computing power industry and the characteristics of economies of scale have become key driving factors. Building on this, the study analyzes how these factors, combined with the integrated control of software and hardware by leading enterprises, create high entry barriers and ecological lock-in effects. The incompatibility of computing power underlying technology and the competition restriction of infrastructure often manifests in anti-competitive behaviors like tying, M&A, and predatory pricing. When the data monopoly advantage is superimposed on the computing power monopoly advantage, the leading enterprises can further enhance the ecological user stickiness. Finally, the paper discusses the practical dilemmas in anti-monopoly governance, noting that traditional definitions of relevant markets and dominant positions are ill-suited for this emerging field, making it difficult to balance industrial development with competition policy. In the world's largest computing power consumer market, it is imperative to strengthen the anti-monopoly of computing power resources. On the basis of clarifying the basic concept and generation logic of computing power resources, this paper combines the specific risk patterns and practical dilemmas brought about by the dynamics of computing power competition, and proposes a targeted normative path : First, "inclusive and prudent" regulation should serve as the core governance principle to address the Collingridge dilemma and balance computing power innovation with oversight. As an emerging industry, computing power is highly susceptible to technological iteration, and the post-correction logic of traditional antitrust often fails to respond to dynamic competition risks in a timely manner. Under the concept of inclusive prudence, anti-monopoly efforts should scientifically establish an analytical framework for defining the relevant market and identifying market dominance, gradually building a preventive governance system. Second, coordination between industrial and competition policies must be promoted to ensure the fairness of market access and factor flow through fair competition reviews. In terms of specific industrial policy guidance, the upstream sector should focus on independent research and development and establish a fair licensing mechanism for results, the middle sector should focus on the construction of dispatching platforms, and the downstream sector should focus on promoting ecological multi-coordination. Third, to establish a public-private cooperative paradigm for anti-monopoly oversight, effective computing power regulation requires a multi-governance model involving government, enterprises, and society. The government should safeguard international competitiveness, implement differentiated domestic supervision, and enforce mandatory interoperability rules. Industry associations need to play a self-discipline function to promote standard unification and resource coordination. While establishing a compliance mechanism, computing power enterprises should assume social responsibilities according to their market control ability.

In the digital economy, platform algorithmic governance has become a critical determinant of users' sustained participation. While platforms enhance matching efficiency, issues such as algorithmic discrimination, opaque decision-making, and perceived unfairness have triggered public skepticism and undermined user trust, thereby threatening platforms' long-term viability. Existing research on algorithmic governance remains fragmented, largely focusing on single-dimensional factors such as transparency or efficiency, while overlooking the conjunctural mechanisms through which multi-dimensional governance attributes shape users' fairness perceptions and subsequently influence sustained participation. To address this gap, this study integrates the Stimulus-Organism-Response (SOR) paradigm with the Technology-Organization-Environment (TOE) framework to construct a complex mediation model. It positions algorithmic governance attributes(spanning technological, organizational, and environmental dimensions)as external stimuli that drive fairness perception and ultimately elicit sustained participation as the behavioral response. The research aims to identify heterogeneous configurational pathways, examine the mediating role of fairness perception, and analyze variations in direct and indirect effects under different intensities of external constraints. A mixed-methods design combining fuzzy-set qualitative comparative analysis and regression analysis was adopted. Data were collected through a structured online questionnaire on the Credamo platform, yielding 600 valid responses from active users of major Chinese e-commerce, social media, and short-video platforms. Regression models subsequently tested the direct effects of configuration membership scores on sustained participation and the mediating role of fairness perception, controlling for demographic variables. The results reveal that no single governance attribute constitutes a necessary condition for high fairness perception, confirming the conjunctural nature of effective algorithmic governance. Six sufficient configurations were identified and parsimoniously synthesized into four distinct types. Justice-Responsibility configurations feature high data bias control and algorithmic accountability as core conditions, with low transparency compensated through internal justice mechanisms, as observed in Meituan's dispatch system and Kuaishou's content recommendation. Transparency-Accountability configurations combine high data bias control and accountability cores with peripheral transparency and low controllability, exemplified by Douyin and Bilibili. The Stability-Constraint configuration integrates high data bias control, controllability, and dual external pressures (regulation and social opinion) as cores, reinforced by transparency, as seen in Pinduoduo under stringent oversight. Externally Oriented configurations are driven by high transparency alongside dual external constraints and low controllability, demonstrated by JD. com's rule disclosure practices. Regression analyses confirm that all four governance types exert significant positive effects on sustained participation, with the Stability-Constraint type displaying the strongest direct effect. Fairness perception significantly mediates these relationships across all types, serving as the central psychological bridge. Heterogeneity is evident with respect to external constraints: configurations operating under strong dual external pressures generate pronounced dual promotion effects through both direct activation of trust and system dependence and indirect fairness-mediated pathways, whereas those relying primarily on single external drivers operate predominantly through the indirect route. The study makes four theoretical contributions. It extends SOR theory into platform algorithmic governance by demonstrating fairness perception as a multi-dimensional mediator that transforms conjunctural TOE stimuli into behavioral outcomes. It advances TOE framework applications by revealing synergistic technological-organizational-environmental interactions rather than isolated effects. The fsQCA-regression mixed approach enhances methodological rigor for examining complex mediation in digital platform research. It also shifts algorithmic governance scholarship from efficiency-centric or single-attribute perspectives toward a justice-oriented configurational lens. The findings indicate that platforms should prioritize the synergistic integration of data bias control mechanisms and robust accountability systems, as this combination reliably elevates fairness perceptions even under conditions of lower transparency. Transparency interfaces must be carefully designed to avoid cognitive overload. Managers can further strengthen user retention by aligning internal governance with external regulatory compliance and social responsiveness, particularly under strong institutional pressures. Policymakers are advised to foster synergistic regulatory-social supervision frameworks that incentivize fairness-enhancing algorithmic practices.

Accelerating the transformation of scientific and technological achievements into real productivity and strengthening basic research capabilities are powerful drivers for the growth of emerging and future industries, and are essential requirements for cultivating and developing new quality productivity. The 2024 Central Economic Work Conference identified technology? led innovation to advance new quality productivity as a key task, and the 15th Five-Year Plan further called for more efficient transformation and application of scientific and technological achievements. The transformation of such achievements serves as a bridge linking research with markets and innovation with industries, and is therefore crucial for enhancing new quality productivity and promoting high-quality development. However, China's innovation system still suffers from a low commercialization rate of research outputs and inefficient technology transfer channels. With the rapid development of the digital economy, data factors have become a key driver of economic growth and technological innovation. As core infrastructure of the data factor market, data trading platforms bridge supply and demand, increase the supply of high-quality data, and facilitate the cross-entity circulation of data factors, thereby alleviating information asymmetry, reducing transaction costs, and accelerating knowledge spillovers. Thus, the platforms provide new momentum for technology transfer. Hence, in-depth investigation of the effects and mechanisms through which data factor marketization influences technology transfer holds significant theoretical value and practical implications for promoting high-quality development and fostering new quality productivity. Using panel data from 216 Chinese cities spanning 2011-2023, this study treats the establishment of data trading platforms as a quasi-natural experiment and employs a staggered difference-in-differences model to examine the impact of data factor marketization on technology transfer. The findings show that data factor marketization significantly promotes technology transfer. This conclusion remains robust after a series of robustness checks, including excluding the influence of other policies, accounting for heterogeneous treatment effects, using alternative variable measures, excluding municipalities directly under the central government, winsorizing extreme values, excluding the impact of the COVID-19 pandemic, and employing instrumental variable estimation. Mechanism analysis reveals that data factor marketization facilitates technology transfer primarily through three channels: alleviating information asymmetry, enhancing industry-university-research collaboration, and reducing R&D costs. Heterogeneity analysis further indicates that the positive effect of data factor marketization on technology transfer is more pronounced in regions with better information infrastructure, more developed institutional environment for data, stronger intellectual property protection, and higher levels of fintech development. Building on these findings, this study offers policy implications including accelerating the deployment of data trading platforms, promoting the precise matching and efficient utilization of data resources in R&D, and tailoring differentiated strategies for data factor market development to local conditions. The marginal contributions of this study are threefold. First, in contrast to existing studies that focus on the impact of public data on technology transfer, this study focuses on the role of data factor marketization, which deals primarily with commercial data. This expands the research boundaries of both the economic effects of data factor marketization and the drivers of technology transfer, offering a new perspective on how data factor marketization promotes technology transfer. Second, by identifying three mechanisms - alleviating information asymmetry, enhancing industry-university-research collaboration, and reducing R&D costs - this study reveals how data factor marketization empowers the transformation of scientific and technological achievements, providing novel empirical evidence on the underlying pathways. Third, through heterogeneity analysis along the dimensions of data infrastructure, intellectual property protection, and fintech development, this study identifies the key conditions under which data factor marketization takes effect, providing policy insights for precisely matching the development of the data factor market with regional development environments.

Cross-border flows of R&D factors are a vital support for fostering new quality productive forces. In recent years, however, escalating external technological restrictions and intensifying geopolitical rivalry have disrupted the global pattern of R&D factor mobility. Against this backdrop, standard connectivity under the Belt and Road Initiative(BRI) has increasingly become an important institutional interface for removing barriers to cross-border R&D factor flows and optimizing the global allocation of science and technology innovation resources. Existing studies have largely examined standard cooperation and cross-border R&D factor flows separately, focusing mainly on the trade-creating effects of standards coordination, technology diffusion, and international standards certification. Systematic and direct empirical evidence on how standard connectivity affects cross-border R&D factor flows in the BRI context, however, remains limited. Drawing on the knowledge spillover perspective and the multidimensional proximity framework, this study employs text-mining techniques to construct an indicator of standard connectivity for BRI partner countries based on project-related news texts published on the official Belt and Road Portal of China. This indicator is then matched with the OECD Inter-Country Input-Output Tables (OECD-ICIO) to examine the impact of standard connectivity on cross-border R&D factor flows. Given that the latest OECD-ICIO (2025 edition) covers data up to 2022, and taking into account the availability of BRI project-related data, the sample period is set from 2011 to 2022. Ultimately, the analysis yields a balanced panel of 696 observations across 58 countries. The empirical results show that standard connectivity significantly promotes cross-border R&D factor flows between China and its partner countries, thereby strengthening bilateral innovation linkages. From a directional perspective, the effect is mainly reflected in facilitating the outward flow of China's R&D factors to partner countries, while its direct linear effect on the inward flow of partner countries' R&D factors into China is not statistically significant. Further analysis reveals that the effect of standard connectivity is conditioned by multidimensional proximity in an asymmetric manner. Heterogeneity analysis shows that the promoting effect is more pronounced for countries along the 21st Century Maritime Silk Road, whereas it is not significant for countries along the Silk Road Economic Belt. Threshold analysis further indicates that the economic size of partner countries constitutes an important boundary condition for the effect of standard connectivity on R&D factor inflows into China. Specifically, using Hansen's panel threshold approach, the study identifies a significant threshold effect of partner countries' economic size on R&D factor inflows into China. Below this threshold, standard connectivity exerts a negative effect on partner-to-China R&D inflows; above the threshold, the effect turns significantly positive. This nonlinear reversal explains the baseline insignificance: smaller economies face binding constraints in R&D supply capacity and standard adaptation costs, whereas larger economies possess the absorptive capacity to leverage harmonized institutional interfaces and accelerate R&D resource allocation toward China. This study makes three main contributions. First, it extends research on the economic consequences of standard connectivity beyond trade performance, export quality, and technology diffusion to the domain of cross-border R&D factor flows, thereby filling an important gap in the literature at the intersection of standard connectivity and cross-border innovation resource allocation. Second, drawing on knowledge spillover theory and the multidimensional proximity framework, it identifies the contingent mechanisms through which standard connectivity affects cross-border R&D factor flows and highlights the heterogeneous effects on outward and inward R&D flows. Third, by constructing a country-year measure of standard connectivity from BRI cooperation project news texts, this study offers a context-sensitive quantitative approach for capturing institutional "soft connectivity" under the BRI, going beyond conventional measures that primarily rely on firm-level data, single-industry standards databases, or formal standards records dominated by developed countries.

In the context of the digital economy, platform ecosystems have become a critical organizational form for firms to transcend boundaries, integrate resources, and co-create value. Yet whether platform ecological embeddedness necessarily improves complementor performance remains contested. Existing research has primarily focused on platform owners' ecosystem orchestration strategies, with limited attention to how complementors acquire and transform knowledge through ecosystem participation. Moreover, most studies adopt a static perspective on collaboration networks, overlooking how partner dynamics shape knowledge flows. Drawing on open innovation theory, this study addresses three questions: Does platform ecological embeddedness enhance complementor performance? What knowledge mechanisms explain this effect? How do partner dynamics moderate the process? To answer these questions, this study draws on open innovation theory to examine how platform ecological embeddedness affects complementor performance, the mechanisms through which knowledge flows operate, and the boundary conditions shaping these relationships. In particular, it investigates the mediating roles of explicit and tacit knowledge flows, as well as the moderating effects of partner stability and partner expansion. Using a panel dataset of Chinese listed firms connected to national-level cross-industry and cross-domain industrial Internet platforms from 2018 to 2023, this study constructs a final sample of 3 064 firms and 12 899 firm-year observations. The results show that (1) platform ecological embeddedness significantly improves complementor performance. This effect is more pronounced among state-owned enterprises, firms with lower financial constraints, and those whose top managers possess digital backgrounds. (2) With respect to the underlying mechanisms, platform ecological embeddedness promotes complementor performance indirectly by fostering both explicit and tacit knowledge flows. Platform ecosystems facilitate explicit knowledge flows through modular architectures, open interfaces, technical standards, process documents, and patent-based knowledge diffusion. They also promote tacit knowledge flows through repeated interaction, project-based collaboration, trust accumulation, and shared contextual understanding. (3) Partner dynamics shape these mechanisms in different ways: whereas partner stability weakens the relationship between platform ecological embeddedness and explicit knowledge flow, it strengthens the relationship with tacit knowledge flow. Conversely, partner expansion shows the opposite pattern. This study makes several contributions. First, it shifts the focus of platform ecosystem research from platform owners and ecosystem orchestration to complementors' value creation and performance improvement. Existing studies have paid considerable attention to how platform owners build, govern, and expand ecosystems, but less is known about how complementors benefit from ecosystem participation. By examining complementors' performance outcomes, this study enriches the understanding of value appropriation in platform ecosystems. Second, this study extends open innovation theory to the platform ecosystem context. Rather than treating knowledge flow as a single mechanism, it distinguishes between explicit and tacit knowledge flows and shows that platform ecosystems support both codified knowledge diffusion and experience-based knowledge transfer. This distinction clarifies the micro-level knowledge foundation of platform-enabled performance improvement. Third, this study incorporates partner dynamics into the analysis of platform ecological embeddedness. By showing the differentiated roles of partner stability and partner expansion, it highlights that collaborative networks are not static background conditions, but dynamic strategic arrangements that shape firms' knowledge acquisition and transformation processes. The findings also have practical implications. Complementors would benefit from embedding in platform ecosystems to access external resources and knowledge, but they should avoid treating platform participation as a purely technical connection. Instead, they need to develop dynamic collaboration strategies. Stable core partnerships should be maintained to support trust-based tacit knowledge exchange and capability accumulation, while new partnerships should be expanded to access heterogeneous explicit knowledge and emerging opportunities. Platform owners and policymakers should improve interface openness, rule transparency, developer support, and ecosystem governance mechanisms, thereby reducing knowledge-flow frictions and enhancing collaborative innovation within platform ecosystems.

With the continuous expansion of urban agglomerations and the optimization of their internal structures, the infrastructure, industrial systems, and development environment of the intelligent economy have been progressively restructured under the combined effects of resource allocation and agglomeration. As a result, the intelligent economy exhibits a sustained upward development trajectory over time. Driven by key production factors and digital technologies, such as artificial intelligence, big data, and cloud computing, the intelligent economy is characterized by multi-factor coupling, multi-agent collaboration, and dynamic evolution. Its development is reflected not only in quantitative growth but also in structural optimization and systemic upgrading. Existing literature has extensively explored the connotations and components of the intelligent economy. A small but growing body of research has begun to examine its developmental trends building upon the digital economy, with preliminary attempts made to construct measurement systems for empirical verification. However, comprehensive, systematic, and comparable measurement studies regarding the intelligent economy remain scarce, and a widely accepted statistical standard or indicator framework has yet to be established. Against this backdrop, it is essential to theoretically characterize the evolutionary patterns of intelligent economy development within urban agglomerations and to reveal their underlying mechanisms. This study adopts a broad conceptualization of the intelligent economy and introduces the WSR framework to construct an analytical structure. Intelligent economy development is decomposed into three interrelated dimensions: foundational support, operational mechanisms, and agent behavior. Specifically, the Wuli dimension corresponds to digital infrastructure and technological conditions, the Shili dimension captures industrial systems and factor allocation mechanisms, and the Renli dimension reflects government regulation and enterprise behavior. Within this framework, accounting for incorporating factor interactions and spatial spillover effects, this study develops a theoretical model to explain the formation mechanism of the wave-like ascending pattern and derives its mathematical expression. The final sample comprises 198 prefecture-level cities. Methodologically, a comprehensive evaluation index system is constructed, covering foundational support, factor input, industrial development, and integrated application. The entropy method is used to measure the level of intelligent economy development across these urban agglomerations from 2014 to 2023. A coupling intensity index and a threshold model are further employed to examine the collaborative development between core and peripheral cities. In addition, numerical simulations based on MATLAB are conducted to validate the proposed wave-like ascending pattern. The results indicate the following findings: (1) the overall level of intelligent economy development shows a steady upward trend, with a spatial gradient extending from eastern coastal regions to central and western regions, reflecting disparities in infrastructure, data resources, and innovation capacity; (2) a significant threshold effect exists, whereby core cities exert strong spillover effects after surpassing development thresholds, while peripheral cities remain constrained and require policy and resource support; (3) intelligent economy development exhibits a nonlinear upward trend with cyclical fluctuations, namely a wave-like ascending pattern. Variations in fluctuation amplitude and frequency across urban agglomerations reflect differences in innovation cycles, industrial restructuring, and policy regulation, revealing heterogeneity in development resilience and adaptability. Thus, in order to reinforce the leading role of core cities and promote regional synergy, it is essential to leverage the advantages of core hubs to radiate innovation and resources to surrounding areas, thereby fostering balanced development within urban agglomerations. Furthermore, to accelerate infrastructure development in peripheral and lagging cities, targeted investments in digital foundations and enhanced policy support should be channeled to bridge regional gaps and strengthen joint development capabilities. Finally, differentiated dynamic regulation mechanisms should be implemented according to regional characteristics to optimize resilience and ensure sustainable growth, while deepening the integration of intelligent technologies with traditional industries to boost comprehensive economic competitiveness.

Intensifying global competition and shortened technology cycles have established innovation as a cornerstone for the sustainable development of technology-based small and medium-sized enterprises (SMEs). Resource scarcity represents a fundamental constraint for these firms, compelling them to seek efficient innovation pathways. Micro-innovation, characterized by low cost, minimal risk, and rapid results, has consequently emerged as their strategic preference. However, the subjective perception of this scarcity can trigger divergent strategic responses, for it can either stimulate proactive innovation or induce defensive conservatism. Some studies suggest that the structural strain induced by the perception of resource scarcity can heighten an organization's willingness to take risks, thereby leading to greater innovation investment. However, whether this positive effect can be sustained requires further investigation. When an organization perceives that its resources are nearing depletion, it may enter a defensive mode, causing its innovation strategy to become more conservative or even irrational. Consequently, the relationship between resource scarcity perception and innovative behavior is likely non-linear. Current academic research primarily focuses on the impact of objective resource conditions, with less attention given to subjective perceptions and to the internal organizational capabilities that these perceptions activate. Furthermore, existing studies often adopt a linear perspective, overlooking potentially complex non-linear relationships and the critical moderating role of organizational psychological resources. Grounded in Conservation of Resources (COR) theory, this study addresses this gap by constructing an integrated model that introduces organizational improvisation capability as a mediator and negative emotion management capability as a moderator. The study empirically tests this model using data from 315 valid questionnaires collected from technology-based SMEs across various provinces in China, employing hierarchical regression and bootstrapping methods. First, this study examines the direct relationship between resource scarcity perception and micro-innovation. The results reveal a significant inverted U-shaped relationship. This finding highlights the dual nature of resource scarcity perception: at moderate levels, it heightens organizational vigilance and promotes micro-innovation as a resource investment strategy; however, when perception exceeds a critical threshold, it triggers defensive resource hoarding, thereby stifling micro-innovation vitality. Second, by incorporating organizational improvisation capability into the framework, the study analyzes its specific mediating role. The findings indicate that resource scarcity perception significantly enhances organizational improvisation capability. Moreover, organizational improvisation capability itself exhibits a significant inverted U-shaped relationship with micro-innovation, acting as a core mediator. Scarcity pressure drives firms to develop improvisational skills, which, within an optimal range, facilitate micro-innovation through flexible resource recombination and iterative learning. Conversely, excessive improvisation leads to cognitive overload and strategic fragmentation, ultimately hindering the sustained accumulation of micro-innovations. Bootstrapping analysis confirms the significance of this mediating pathway. Finally, the study investigates the moderating role of negative emotion management capability. Our analysis confirms that this capability significantly mitigates the inverted U-shaped impact of resource scarcity perception on micro-innovation. A high level of negative emotion management provides a psychological buffer against the strain of intense scarcity perception, helps maintain cognitive flexibility and psychological safety, and thus mitigates the decline in micro-innovation observed at high perception levels. In conclusion, this study refines the conceptual understanding of resource scarcity perception, clarifying that its influence is not linear but follows an inverted U-shaped curve governed by the principles of resource investment and defense. It identifies organizational improvisation capability as a double-edged sword and confirms the critical moderating function of negative emotion management capability. These findings delineate the specific pathways through which perception translates into action, extending existing research on the antecedents of micro-innovation and the mechanisms linking resource scarcity perception to innovation outcomes. The study offers valuable insights for technology-based SMEs in China and similar contexts, guiding them on how to effectively manage scarcity perceptions, cultivate dynamic capabilities, and build emotional resilience within resource-constrained environments.

Green innovation, serving as a pivotal driver of industrial upgrading and high-quality economic development, has emerged as a strategic imperative for firms seeking to attain sustainable competitive advantage. Against the backdrop of China's "dual carbon" goals, stimulating firms' intrinsic motivation for green innovation represents a crucial strategic imperative for achieving high-quality development and economic transformation. As an emerging strategic orientation embedded with environmental sustainability concepts, green entrepreneurial orientation (GEO) serves as a pivotal endogenous driver enabling enterprises to overcome path dependencies and pursue high-quality green innovation. However, existing literature predominantly relies on questionnaire-based measurements of GEO, which constrains large-scale longitudinal empirical analysis, while systematic exploration of the micro-transmission mechanisms and boundary conditions through which GEO translates into substantive green innovation remains insufficient. To address these gaps, this study rigorously investigates the impact of GEO on corporate green innovation, unpacking the underlying mechanism black box by introducing green intellectual capital as a mediator and multi-dimensional stakeholder pressure as a moderator. Grounded in the natural-resource-based view, intellectual capital theory, and stakeholder theory, this study selects Chinese A-share listed companies from 2013 to 2023 as the research sample. To overcome the limitations of conventional subjective measurements, the study employs machine learning and text analysis techniques. Specifically, it applies the Jieba module and the Word2Vec skip-gram model with negative sampling to construct a GEO dictionary from the Management Discussion and Analysis (MD&A) sections of corporate annual reports, thereby establishing an objective and robust GEO measurement scheme. The study further develops multiple linear regression models incorporating two-way fixed effects and adopts an instrumental variable (IV) strategy leveraging the entrepreneurial activity index of CEOs' birthplaces to mitigate potential endogeneity concerns. The empirical findings yield several noteworthy contributions. First, GEO exerts a significantly positive effect on corporate green innovation, confirming that embedding sustainability concepts into strategic decision-making effectively directs resources toward green technology advancement. Second, stakeholder pressure exerts a significant positive moderating effect on this relationship. Government regulatory pressure, supply chain collaborative pressure, and investor capital pressure all positively moderate this relationship. Third, mechanism analyses reveal that GEO enhances green innovation primarily through fostering green intellectual capital accumulation, operating via three parallel yet interconnected pathways: green human capital, green structural capital, and green relational capital. Notably, synergistic interaction effects are identified among these dimensions, particularly between structural and relational capital. Fourth, heterogeneity analyses demonstrate that the promotional effect of GEO is more pronounced in non-state-owned enterprises which are more sensitive to market-oriented green advantages and in pollution-intensive industries that face more stringent environmental regulatory pressures. Finally, extended analyses confirm that GEO significantly enhances firms' green total factor productivity (GTFP) through green innovation, highlighting its long-term economic and environmental implications. The theoretical contributions of this paper are threefold. First, it proposes a novel, objective GEO measurement scheme grounded in annual report text analysis, providing a replicable methodological foundation for large-sample future research. Second, by incorporating the stock perspective of green intellectual capital, it delineates a refined micro-transmission pathway from GEO to green innovation, extending beyond the conventional dynamic capability framework. Third, it constructs a three-dimensional moderating framework of stakeholder pressure, broadening the contextual boundary conditions of the GEO - innovation relationship. These findings carry important practical implications: policymakers are advised to implement differentiated support instruments tailored to firm heterogeneity and to establish synergistic multi-stakeholder mechanisms that provide robust external guarantees for corporate green transformation.

Artificial intelligence(AI) has become a strategic technology shaping industrial transformation, national competitiveness, and global governance. However, rapid AI-related patent growth does not necessarily reflect improved innovation capability. Corporate AI innovation differs from conventional innovation in its strong dependence on high-quality data, interdisciplinary scientific and technological talent, computing capacity, and application scenarios. This study examines whether the policy for designating National Technological Innovation Demonstration Enterprises(NTIDEs) can promote both the quantity and quality of corporate AI innovation and generate external spillovers beyond designated firms. Using an unbalanced panel of Chinese A-share listed companies on the Shanghai and Shenzhen stock exchanges from 2007 to 2024, the study treats NTIDE designation as a quasi-natural experiment and estimates its effects with a staggered difference-in-differences model with firm and year fixed effects. AI innovation quantity is measured by the logarithm of AI-related invention patent applications identified through patent-abstract keyword matching, and quality by patent knowledge breadth. The analysis further examines two mechanisms of data utilization and talent concentration, and evaluates spillovers across industries, supply chains, and geographic space. Multiple identification and robustness procedures are employed, including instrumental-variable estimation, propensity-score-matching difference-in-differences, the Heckman two-stage method, parallel-trend and placebo tests, alternative measures, exclusion of potentially confounding AI pilot policies, and estimators robust to heterogeneous treatment effects. The results show that NTIDE designation significantly increases both the quantity and quality of corporate AI innovation, producing a clear dual improvement in innovation scale and technological quality. The findings remain stable across alternative specifications and endogeneity tests, indicating that the policy stimulates more than short-term patent expansion. Mechanism analysis shows that designation improves firms' data utilization and strengthens their capacity to attract and concentrate scientific and technological talent. The policy therefore operates through the factor foundations critical to AI innovation rather than merely through general R&D increases or symbolic recognition. The spillover analysis reveals substantial differences across transmission channels. At the industry level, the presence of a designated enterprise has no significant effect on peer firms' AI patent counts but significantly improves their innovation quality, suggesting that industry spillovers operate through technological demonstration, organizational learning, and competitive pressure rather than simple imitation or patent-volume expansion. At the supply-chain level, spillovers are strongly directional: when a designated enterprise acts as a supplier, its AI-enabled products and solutions significantly promote both the quantity and quality of AI innovation among customer firms, whereas no significant spillover to suppliers is observed when it acts as a customer. AI innovation therefore diffuses from the technology-supply side to the application side. At the regional level, geographic proximity alone does not produce significant spillovers, implying that AI knowledge diffusion depends more on substantive technological and economic linkages than on physical distance. Further analysis identifies important boundary conditions. The positive effects are concentrated among firms with stronger prior AI investment foundations and in high-technology industries, highlighting the importance of absorptive capacity and industrial compatibility. The effects are also more pronounced in regions with relatively weak computing infrastructure, suggesting that designation can partially compensate for local constraints through policy attention and resource access. In addition, the policy significantly enhances the continuity of corporate AI innovation, demonstrating sustained rather than one-off effects. This study contributes to the literature in three respects. First, it evaluates an innovation-oriented designation policy across AI innovation quantity, quality, and continuity. Second, it identifies data utilization and talent concentration as AI-specific mechanisms. Third, it clarifies the networked and directional nature of policy spillovers while demonstrating the limitations of geographically based diffusion. The findings suggest that future policy design should strengthen dynamic evaluation, emphasize innovation quality and sustained output, support enterprise data governance and high-level technical talent, and amplify demonstration effects through industry platforms and supply-chain collaboration.

With the deep integration of artificial intelligence into corporate R&D, decision-making, and innovation activities, innovation is increasingly shaped by the collaboration between human intelligence and artificial intelligence rather than by either human expertise or technological automation alone. This shift is particularly important for high-tech enterprises, which face rapid technological change, high R&D uncertainty, long innovation cycles, and increasing external shocks. Under such conditions, firms need not only innovation capability, but also innovation resilience, namely the ability to maintain innovation activities, recover from disruptions, and reconstruct innovation paths in uncertain environments. Human-AI collaboration provides a new mechanism through which high-tech enterprises can reorganize heterogeneous innovation resources and enhance innovation resilience. Drawing on resource orchestration theory, this study constructs a theoretical model of "human-AI collaboration—knowledge coupling—innovation resilience" and introduces digital agility as a boundary condition. Resource orchestration theory emphasizes that competitive advantage does not come simply from resource possession, but from the effective structuring, bundling, and leveraging of resources. Following this logic, human-AI collaboration is regarded as a process of resource structuring, in which human experience, contextual judgment, algorithmic capability, and data resources are jointly configured. Knowledge coupling represents the process of resource bundling, through which dispersed knowledge elements are connected, recombined, and transformed into usable innovation resources. Innovation resilience reflects the result of resource leveraging, showing whether firms can transform reconfigured resources into sustained innovation under uncertainty. This study further distinguishes between complementary knowledge coupling and substitutive knowledge coupling. Complementary knowledge coupling emphasizes the integration of different knowledge domains and the expansion of knowledge boundaries, while substitutive knowledge coupling focuses on replacing or reconstructing existing knowledge paths to reduce dependence on a single technological trajectory. Digital agility is introduced to explain when human-AI collaboration is more likely to enhance innovation resilience. Firms with high digital agility can sense changes more quickly, mobilize digital resources more effectively, and transform human-AI collaborative outputs into innovation actions more rapidly. The empirical analysis is based on two-wave paired survey data from 328 employees of Chinese high-tech enterprises. To reduce common method bias, the first-wave survey measured human-AI collaboration, knowledge coupling, and digital agility, while the second-wave survey measured innovation resilience. The sample covers several high-tech manufacturing industries, including pharmaceutical manufacturing, aerospace equipment, electronic and communication equipment, computer and office equipment, medical instruments, and information chemicals. The results show that human-AI collaboration significantly improves innovation resilience. Both complementary knowledge coupling and substitutive knowledge coupling play positive mediating roles between human-AI collaboration and innovation resilience. In addition, digital agility positively moderates the relationship between human-AI collaboration and innovation resilience, indicating that firms with stronger digital agility can better transform human-AI collaboration into resilient innovation outcomes. This study makes three contributions. First, it extends resource orchestration theory to the digital intelligence context by conceptualizing human-AI collaboration as a new form of strategic resource configuration. Second, it reveals the knowledge transformation mechanism through which human-AI collaboration enhances innovation resilience. Third, it identifies digital agility as an important organizational condition shaping the effectiveness of human-AI collaboration. The findings also offer practical implications for high-tech enterprises seeking to build human-AI collaborative advantages, strengthen knowledge integration, and enhance sustained innovation capability.

Innovation is a central driver of high-quality economic development and a key source of firms' long-term competitive advantage. Yet corporate innovation is typically characterized by high uncertainty, long investment cycles, weak collateralizability, and severe information asymmetry, which make innovative firms highly dependent on stable external financing. In China's bank-dominated financial system, traditional credit allocation has long faced problems such as fragmented information, repeated credit granting, and excessive lending among multiple banks. These problems may distort financial resource allocation and weaken the ability of credit markets to support long-term innovation. To address these issues, the China Banking and Insurance Regulatory Commission (CBIRC) introduced a pilot joint credit system in 2018, featuring a lead-bank mechanism, joint credit committees, unified credit limits, and financing ledgers. Using Chinese A-share listed firms from 2011 to 2024, this paper treats the implementation of the Joint Credit Granting System as a quasi-natural experiment and employs a propensity score matching difference-in-differences approach to examine its causal effect on corporate innovation. Innovation input is measured by R&D intensity, while innovation output is measured by patent applications. The results show that the Joint Credit Granting System significantly increases both R&D investment and patent output. In terms of economic magnitude, the policy raises R&D intensity by approximately 4.1% relative to the sample mean and increases expected patent output by about 14.0%. These findings remain robust after parallel trend tests, placebo tests based on 500 random assignments of pseudo-treated firms, alternative matching methods, alternative measures of key variables, different estimation models, lagged-control regressions, and additional controls for industry-specific trends and regional pilot effects. The evidence suggests that the innovation-promoting effect is unlikely to be driven by sample selection bias, omitted variables, model specification, or unobservable time trends. Mechanism analyses reveal three complementary channels. First, the system generates an information-sharing effect. By coordinating information among creditor banks, it reduces interbank information fragmentation, lowers information search costs, alleviates financing constraints, and improves firms' external information environment, thereby creating more stable credit expectations for long-term R&D projects. Second, it produces an innovation governance effect. Joint credit committees strengthen coordinated creditor monitoring, reduce managerial opportunism and agency conflicts, and restrain the inefficient use or diversion of innovation resources, thereby improving the allocation efficiency of financial resources. Third, it creates a collaborative innovation effect. Through reputational endorsement by multiple financial institutions, the system attracts governments, universities, research institutes, suppliers, and other external participants into firms' innovation networks, promoting joint patent applications and industry-university-research collaboration. Heterogeneity analyses show that the innovation-enhancing effect is stronger among manufacturing firms and firms located in regions with stronger intellectual property protection. Economic consequence tests further reveal that innovation induced by the system significantly improves firms' basic earnings per share. This paper contributes to the literature in several ways. First, it shifts the analysis of finance and corporate innovation from single bank-firm relationships to multi-bank coordinated credit governance, enriching financial intermediation theory from a collective governance perspective. Second, it expands research on the economic consequences of the Joint Credit Granting System by showing that this reform is not only a debt-discipline or risk-prevention mechanism, but also an innovation-enhancing institutional arrangement. Third, it opens the black box of how financial institutional reform affects firms' innovation decisions by identifying information sharing, governance monitoring, and collaborative innovation as interrelated mechanisms. Fourth, by linking joint credit granting, corporate innovation, and earnings performance, this paper provides micro-level evidence that supply-side financial reform can generate real economic value when credit governance is aligned with firms' long-term technological upgrading. These findings deepen our understanding of the institutional foundations of innovation financing and the role of coordinated credit governance in improving financial resource allocation.

Amid intensifying global economic uncertainty, profound domestic structural adjustment, and accelerating technological change, enterprises face the dual challenge of enhancing innovation capacity and preventing default risk. While innovation strengthens long-term competitiveness, it is inherently characterized by high investment, long cycles, and substantial uncertainty. When firms bear the risk of innovation failure independently, capital constraints may deepen and default risk may escalate. Default risk reflects not merely the accumulation of debt, but the interplay of asset value, debt servicing pressure, and cash flow stability. In the context of innovation-driven development, such risk is further linked to the uncertainties, information asymmetries, and resource misallocations inherent in innovation activities. Therefore, this study examines the risk-mitigation effect of university-industry-research collaborative innovation on corporate default risk and its transmission mechanisms. As firms face pressure to improve innovation capacity while maintaining financial stability, collaboration among enterprises, universities, and research institutes may contribute not only to innovation performance but also to corporate risk governance. Using Chinese A-share listed companies from 2013 to 2024 as the sample, this study develops a theoretical framework based on information asymmetry theory and agency cost theory. Corporate default risk is measured using a distance-to-default model. A two-way fixed-effects model is employed to estimate the impact of university-industry-research collaborative innovation on corporate default risk. To address potential endogeneity and sample-selection bias, the study applies an instrumental-variable approach and propensity score matching. To mitigate potential interference caused by the sensitivity of matching methods, this study adopts the entropy balancing method to re-estimate the model. This study further employs three additional methods to enhance the robustness of the conclusions: alternative variable measurement, exclusion of certain confounding factors, and the inclusion of high-dimensional fixed effects. The findings show that university-industry-research collaborative innovation significantly reduces corporate default risk. The mechanism analysis reveals that this effect operates through three channels. First, collaborative innovation generates a technological effect by improving total factor productivity. Access to knowledge, research expertise, and technological resources enhances knowledge integration, technological absorption, and resource allocation efficiency, thereby strengthening operational performance and financial resilience. Second, collaborative innovation produces a certification effect by alleviating financing constraints. Cooperation with universities and research institutes enhances the credibility of firms’ innovation activities, reduces information asymmetry between firms and external capital providers, and improves financing accessibility, thereby strengthening debt-servicing capacity. Third, collaborative innovation creates a governance effect by restraining inefficient investment and excessive debt. The involvement of research organizations introduces expertise and external constraints into corporate decision-making, helping reduce resource misallocation, curb aggressive investment, and limit excessive debt accumulation. Further analysis shows that the risk-mitigation effect is more pronounced among non-state-owned enterprises, firms in technology-intensive industries, and firms in regions with stronger intellectual property protection. Findings indicate that the effectiveness of collaborative innovation varies with ownership characteristics, technological attributes, and institutional environments. To optimize industry-university-research collaboration, a multi-stakeholder approach is recommended. First, enterprises should tailor strategies based on ownership; specifically, non-state-owned firms can leverage these partnerships to mitigate risks, improve information transparency, and reduce inefficient investments under financing constraints. Second, financial institutions should upgrade their credit risk assessment models to evaluate the quality and conversion efficiency of collaborative innovation, thereby offering better financial support to tech-intensive firms. Finally, governments should strengthen institutional environments by refining intellectual property protection and dispute resolution mechanisms, particularly in regions with weaker IP enforcement, to provide stable expectations and reduce innovation risks. This study extends research on the economic consequences of university-industry-research collaborative innovation from innovation performance to corporate risk governance. By identifying technological improvement, certification effect, and governance enhancement as key transmission channels, it provides micro-level evidence on how collaborative innovation affects financial stability and offers support for firms seeking to optimize collaborative innovation decisions and improve credit risk management.

In the digital economy era, data are profoundly transforming the innovation paradigm and have emerged as a vital strategic resource driving economic and social development as well as enterprises' innovation. Amid intensifying international technological competition, China's strategic goal of achieving high-level technological self-reliance has further highlighted the urgency of innovation in key core technologies (KCT innovation). However, compared with general innovation, efforts to tackle key core technologies are characterized by uncertain technological pathways, long R&D cycles, substantial capital requirements, and complex knowledge structures. Consequently, firms generally face extremely high R&D risks and trial-and-error costs, persistent liquidity constraints, and difficulties in cross-organizational collaboration. Although existing studies have explored the impact of data assets on general, green, and breakthrough innovations, systematic analysis remains scarce on how the specific attributes of data assets help resolve the dilemmas associated with KCT innovation. To this end, drawing on the resource-based view, signaling theory, and transaction cost theory, this study maps the static evaluation attributes of data assets onto dynamic functional characteristics for innovation empowerment, thereby constructing an analytical framework for examining the impact of data assets on KCT innovation. This study draws on Chinese A-share listed firms on the Shanghai and Shenzhen stock exchanges from 2013 to 2023 as the initial sample. Financial firms, ST firms, and observations with missing key variables are excluded, yielding a final panel dataset of 30 312 firm-year observations. Text mining techniques identify data asset-related terms in the annual reports of listed firms and construct a corporate data asset index. Enterprises' KCT innovation is measured based on the 1 047 key technologies covered in the Industrial Foundation Innovation and Development Directory (2021 Edition), with technical keywords extracted and matched with International Patent Classification (IPC) codes and corporate patent application data. A two-way fixed effects model tests the research hypotheses, with robust standard errors clustered at the firm level. The reliability of the conclusions is further verified through robustness tests using the instrumental variable method, entropy balancing matching, and a double machine learning model. The results indicate that data assets promote KCT innovation primarily through three pathways: the innovation efficiency enhancement effect, the innovation resource optimization effect, and the innovation network embedding effect. Further heterogeneity analysis finds that the positive enabling effect of data assets is more pronounced for firms that actively engage in industry-university-research collaboration, have greater access to patient capital, and operate in regions with robust computing infrastructure. Further analysis shows that data assets exhibit an asymmetric forward spillover effect within the supply chain network; that is, the data assets of focal firms significantly promote the KCT innovation of downstream customer firms, but have an insignificant impact on upstream supplier firms. This implies that product improvements, applied knowledge, and technical information generated by data assets may be more easily transmitted downstream along the production network to be absorbed and transformed by customer firms. Distinct from existing literature that primarily focuses on the input scale of innovation elements, this paper proceeds from the perspective of the morphological evolution of data elements from resources to assets, revealing the underlying logic through which data assetization helps mitigate R&D risks, financing constraints, and collaborative barriers, and constructing a corresponding theoretical analytical framework. Additionally, this paper further identifies boundary conditions such as industry-university-research collaboration, patient capital, computing infrastructure, and external technological restrictions, and examines the directional spillover effects of data assets within the supply chain. The research conclusions not only provide novel empirical evidence for understanding the microeconomic value of data assets but also offer practical insights for firms to effectively allocate new production factors, enhance independent innovation capabilities, and break through key core technology bottlenecks.

AI enterprises urgently need to break through existing development paradigms and explore more application-oriented innovative models. Existing research and practice have demonstrated that scenarios, serving as a critical bridge between technology and market demand, can effectively address pain points in technology implementation and are becoming a pivotal element for AI enterprises to overcome high-growth predicaments. However, current literature still lacks in-depth analysis of the mechanism through which scenarios influence the growth of artificial intelligence enterprises. Particularly from a research perspective, mainstream views generally emphasize how enterprises should allocate and utilize existing resources to gain competitive advantages, drawing on resource orchestration or optimal differentiation theories. Nevertheless, the high growth of artificial intelligence enterprises does not merely depend on the scale of resource investment or the level of technological advancement; rather, it hinges on whether enterprises can identify and realize the application possibilities afforded by continuously evolving scenario practices, thereby further promoting the transformation of product value. Therefore, it is necessary to introduce an affordance perspective and explore, in stages, the dynamic mechanism through which scenario-driven artificial intelligence enterprises achieve high growth. Addressing this theoretical gap, this study focuses on the core issue of "how scenarios drive the high growth of AI enterprises". Following the logic of "affordance-affordance realization-growth performance", the study adopts an exploratory longitudinal single-case study method with Tianjin Huizhixingyuan Information Technology Co., Ltd. (hereinafter referred to as "Huizhixingyuan") as the research subject. As a pioneer in applying large language model (LLM) technology and AIGC products for government and enterprise sectors, Huizhixingyuan focuses on vertical LLMs for data agents. Its self-developed "Zhaoyao" large model has been extensively applied in five vertical domains: data, knowledge, government affairs, industry, and case handling. Huizhixingyuan has achieved remarkable revenue growth, entering a phase of rapid expansion and earning recognition as a high-growth enterprise. Through multi-channel data collection and the "first-order-second-order- aggregation" data coding method proposed by Gioia et al., this study constructs a theoretical framework of "contextual affordance- adaptive innovation- enterprise high growth". The findings are as follows: (1) The scenario affordance, including cognitive, interactive, and compatible affordance, drive artificial intelligence enterprises to engage in adaptive innovation. (2) The key mechanism for the high growth of artificial intelligence enterprises lies in the scenario-driven innovation path. Across the three stages of scenario anchoring, scenario refinement, and scenario amplification, this mechanism is presented as “targeted innovation driven by single scenario pain points- feedback innovation driven by cross-scenario extension- continuous innovation driven by multi-scenario integration”. (3) Under this mechanism, AI enterprises have demonstrated high growth performance by locking in the first-mover advantages, penetrating niche markets, and building high competitive barriers. Drawing on Affordance Actualization Theory, this study examines an AI enterprise that leverages large-model technology to deeply cultivate vertical-field scenarios, exploring the micro-level process of scenario-driven high growth and advancing micro-level theoretical development. First, it establishes an intrinsic link between scenario affordance and AI enterprise high growth, filling a theoretical gap in understanding how scenarios enable such growth. Second, from affordance and scenario perspectives, it delineates the specific content of scenario affordance, bridging Affordance Theory and Scenario Theory. Finally, it reveals the adaptive innovation process, encompassing targeted, feedback-driven, and continuous innovation, that enterprises demonstrate within specific scenarios, offering directions for the future development of Scenario-driven Innovation Theory.

New entrants capture niche markets via disruptive digital technologies, triggering a sharp rise in user migration risks. This transforms the long-accumulated stock data resources of incumbent platforms from core competencies into an inertia trap hindering their growth, while deep mining of stock data in existing businesses underpins the maintenance of current competitive advantages, over-reliance erodes agility in responding to new entrants. Dynamic capability refers to a higher-order competence that purposefully integrates, extends, and reconfigures existing resource portfolios based on a holistic understanding of internal and external environments and resources. It enables incumbent platforms to mitigate the predicament where existing resources and core competencies fail to align with market changes, and to generate sustained collaborative value across the industrial chain and competitive advantages. The theoretical nature of dynamic capability building lies in the effective redistribution of limited total resources between "responding to external changes" and "addressing internal conditions". New entrants lack scaled stock data and historical business constraints, so their dynamic capability building goal is to quickly capture market share based on initial endowments. In contrast, incumbent platforms must design innovative services to counter external new entrant shocks while enhancing internal stock data mining depth to consolidate existing service advantages. The logic of dynamic capability building for this "balancing internal and external" strategic contradiction differs significantly from that of new entrants. Yet, current process-focused studies often vaguely outline the entire lifecycle of platforms from inception, while framework-focused studies fail to elaborate on the demands and corresponding actions related to stock data, thus lacking sufficient theoretical explanations for incumbent platforms. This study follows theoretical sampling principles and selects the 1688 Platform as the research object. Given its highly complex industry context and sustained market leadership, 1688 Platform exhibits the archetypal characteristics of an incumbent platform with ongoing dynamic capabilities. By systematically tracing 1688 Platform's developmental trajectory and external critical events, the study establishes the evolution of business focus as the segmentation criterion. Three critical junctures are identified: (1) strategic transformation planning driven by policy shifts, (2) the official launch of the industrial belt map, and (3) the deployment of the self-developed large language model-based industrial knowledge graph. Guided by the triangulation principle, the study collects both primary and secondary data from multiple channels to support data analysis and theory building. Primary data comprise approximately 24.5 hours of semi-structured interview recordings gathered through three years of longitudinal fieldwork (2022—2025) with the executives, product managers, and ecosystem partners. Secondary data include internal strategic documents, meeting minutes, publicly available news reports, and corporate filings. Observational data accumulated during the consulting engagement further strengthen cross-source validation. The analysis identifies that, first, external competitive scenarios with new entrants and internal stock data constraints are the drivers of dynamic capability building. Second, the phased orchestration of external competition strategies and stock data strategies constitutes the process mechanism. During the three phases of capability generation, growth, and maturity, the platform builds dynamic capability via three mechanisms: compensatory orchestration, resonant orchestration, and leapfrog orchestration of the two strategies. Third, the building sequence is "sensing opportunities—reconfiguring stock data—seizing opportunities and new resources". The platform prioritizes reconstructing stock data into hierarchical forms with increasing value density to enable sharing across both exploratory and exploitative businesses. Fourth, the outcomes include "stock data aggregation capability, industrial knowledge connection capability, and industrial intelligence creation capability", which sequentially reflect the progressive process from basic data processing to high-order intelligence creation. The findings enrich dynamic capability building mechanisms for different actors in industrial digitalization contexts, reveal the core role of hierarchical stock data forms in dynamic capability building frameworks, and extend the application scope of Optimal Distinctiveness Theory to incumbent industrial internet platforms.

The formation and development of new quality productive forces urgently require institutional innovation as a driving force. As a significant measure in the reform of the science and technology innovation system, the "Jie Bang Gua Shuai" system aims to address key core technical bottlenecks and optimize the allocation of scientific and technological innovation resources. However, its impact mechanism on regional new quality productive forces remains to be thoroughly explored. From the theoretical perspective of productivity and production relations, the formation of new quality productive forces requires corresponding adjustments in production relations. The "Jie Bang Gua Shuai" system reconfigures the tripartite relationship among government, market, and innovation entities through its incentive and constraint mechanisms, representing a significant institutional exploration of production relations. This study seeks to reveal the complex causal relationships between the implementation of this system and the development of regional new quality productive forces, thereby providing both a theoretical basis and practical pathways for implementing the system according to local conditions to promote regional development of new quality productive forces. This study draws upon policy texts and statistical data covering 30 provincial-level regions in China over the period 2012 – 2021. It employs a mixed-method approach combining staggered Difference-in-Differences (DID) and fuzzy-set Qualitative Comparative Analysis (fsQCA). The DID method is used to test whether the "Jie Bang Gua Shuai" system can significantly enhance regional new quality productive forces and to examine its dynamic effects. Following the theory of incentive and constraint mechanisms, the study identifies six key operational factors from the implementation process of the system: demand collection method, competition mechanism, entry barriers, subsidy ratio, subsidy disbursement method, and subsidy supervision system. Using policy texts issued by provinces from 2017 to 2022, these antecedent variables are calibrated. Subsequently, fsQCA is employed to explore the complex interaction mechanisms among these six conditions and identify multiple configuration paths that drive the improvement of regional new quality productive forces. Regional new quality productive forces, the outcome variable, are measured using a comprehensive indicator system covering substantive elements and penetrating elements, calculated using the entropy method. The DID results indicate that the "Jie Bang Gua Shuai" system has significantly enhanced regional new quality productive forces. The policy effect is not only statistically significant but also demonstrates progressive and sustained characteristics. Event study analysis shows that the policy effect gradually releases over time, with coefficients increasing from 0.028 (p<0.05) in the first year of implementation to 0.118 (p<0.001) in the third year, confirming the system's long-term promoting effect. Necessity analysis reveals that no single operational factor constitutes a necessary condition for high-level new quality productive forces, underscoring the importance of configurational pathways. The fsQCA results identify four distinct driving pathways: (1) The competitive incentive path, which centers on open and competitive mechanisms, supplemented by high entry barriers, strict supervision, high subsidy ratios, and phased disbursement. This path stimulates market vitality and innovation agency through full competition. (2) The stable guidance path, characterized by government-determined needs, low entry barriers, low subsidy ratios, one-time disbursement, and budget-based supervision. This low-risk approach steadily promotes foundational improvements in new quality productive forces. (3) The targeted breakthrough path, focusing on government-led tackling of "bottleneck" technologies, combining low entry barriers with high subsidy ratios and multi-stage disbursement. This path provides stable resource support for long-term, high-risk research to overcome key technological constraints. (4) The transformation promotion path, centered on government-directed needs, combining high entry barriers with low subsidy ratios and the flexible "lump-sum" supervision system, aimed at accelerating the transformation and application of scientific and technological achievements into practical productive forces. This study advances quantitative research on new quality productive forces by empirically testing the "Jie Bang Gua Shuai" system's effects and mechanisms. It introduces an incentive-constraint framework explaining how six key mechanisms synergistically generate institutional effectiveness, overcoming prior reliance on descriptive methods. The mixed-method design employing staggered DID and fsQCA addresses single-method limitations in causal analysis, offering replicable tools for innovation policy research. Four identified implementation paths provide practical guidance for local adaptation.

Abstract: As Chinese multinational corporations (MNCs) continue to deepen their international operations, repatriates have increasingly become important carriers through which overseas knowledge flows back into home-country organizations. However, repatriate knowledge transfer is often difficult to realize effectively in practice. This difficulty lies not only in the cross-regional and cross-border movement of knowledge, but more importantly in the fact that such knowledge is deeply embedded in the host-country institutional environment, business practices, and cultural context. As a result, this type of knowledge is highly tacit and context-dependent, making it difficult for domestic colleagues to directly understand, absorb, and apply. In this context, cross-cultural communication competence, as an important capability that enables individuals to identify cultural differences, adjust interaction patterns, and convey the meaning of information in multicultural settings, may become a useful factor in overcoming the barriers to repatriate knowledge transfer. While existing studies have demonstrated its positive effects on expatriates' adaptation and knowledge sharing, limited attention has been paid to its role at the repatriation stage. Moreover, prior research on repatriate knowledge transfer has focused mainly on individual characteristics or single mediating mechanisms, neglecting the joint roles of domestic colleagues as knowledge recipients and leaders as organizational support providers. Drawing on social exchange theory, social information processing theory, and followership theory, this study constructs a serial mediation model based on the "competence – trust – behavior" transmission logic. Using matched two-wave survey data collected from 205 repatriate-domestic colleague pairs, this study empirically examines the effect of cross-cultural communication competence on repatriate knowledge transfer, as well as the mediating roles of domestic colleagues' affective trust and leader empowering behavior, by means of hierarchical regression analysis. The empirical results yield three major findings. First, cross-cultural communication competence has a significant positive effect on repatriate knowledge transfer. This indicates that repatriates who are able to adjust their verbal and nonverbal expressions according to domestic colleagues' cognitive background and specific interaction contexts can make overseas knowledge easier for domestic colleagues to understand, absorb, and apply. In other words, stronger communication competence helps reduce communication friction and enhances the comprehensibility and applicability of overseas knowledge in the home-country organizational setting. Second, domestic colleagues' affective trust and leader empowering behavior each play significant parallel mediating roles between cross-cultural communication competence and repatriate knowledge transfer. On the one hand, stronger communication competence helps repatriates build affective trust with domestic colleagues, which in turn increases colleagues' willingness to receive, understand, and apply overseas knowledge. On the other hand, stronger communication competence also increases the likelihood that leaders provide empowering support, thereby creating favorable conditions for repatriate knowledge transfer. Third, domestic colleagues' affective trust and leader empowering behavior further constitute a significant serial mediation path. Specifically, cross-cultural communication competence enhances domestic colleagues' affective trust, which then promotes leader empowering behavior and ultimately facilitates repatriate knowledge transfer. These findings suggest that the realization of repatriate knowledge transfer is essentially the result of a continuous transmission process of 'competence-relationship-empowerment'. Compared with previous studies, this study contributes to the literature in two major respects. First, it extends the research context of cross-cultural communication competence from the expatriation stage to the repatriation stage, clarifying its role in promoting repatriate knowledge transfer and broadening its application context and connotation. Second, it integrates repatriates, domestic colleagues, and leaders into a continuous transmission framework, revealing the serial mediation of domestic colleagues' affective trust and leader empowering behavior between cross-cultural communication competence and repatriate knowledge transfer. This enriches the theoretical framework of repatriate knowledge transfer and provides practical implications for Chinese MNCs seeking to leverage repatriated knowledge and enhance global competitiveness.

With the continuous evolution of the digital environment driven by the integration of digital and intelligent technologies, patent technology value assessment is facing new challenges. The ongoing technological revolution and industrial transformation are reshaping industrial structures, innovation patterns, and resource allocation, with digitalintelligent fusion providing the technological foundation for emerging industries. However, these industries consist of multiple subsectors that differ significantly in technological paths, knowledge structures, and market conditions, and their reliance on digitalintelligent technologies is uneven. This environment alters the generation and diffusion mechanisms of new technologies, making patents a core strategic resource, while also necessitating a comprehensive overhaul of evaluation indicators, methods, and applications, thereby affecting the validity of assessment results. Traditional evaluation systems are often unable to accurately reflect the value of patent technologies across different sectors within emerging industries, as they rely on conventional indicators that lack dimensions related to technology space—a critical factor in the context of rapid technological convergence. The newgeneration information technology industry, characterized by deep digitalization, networking, and intelligence, exhibits frequent crosscombinations and knowledge recombination, making it an ideal case for studying these issues. Therefore, based on existing research on patent value assessment, this study develops a multidimensional patent technology value evaluation framework to improve assessment accuracy and support technological decisionmaking in the new-generation information technology industry. This study constructs a multidimensional patent technology value evaluation system tailored to the new generation information technology industry, grounded in the recognition that patent value in emerging industries arises not only from inherent innovative attributes but also from structural positions within knowledge networks and contextual contingencies of industrial application. The framework integrates four complementary dimensions: patent basic characteristics, patent technology quality characteristics, patent technology space characteristics, and industry context characteristics, encompassing 9 first level and 12 second level indicators in total. Specifically, the basic characteristics dimension includes knowledge synergy degree and technical expression richness. The quality dimension covers technological innovativeness, technological diversity, and technological competitiveness. The technology space dimension, derived from IPC cooccurrence networks, employs betweenness, closeness, and degree centrality to capture network control, enabling connectivity, and universal applicability. The industry context dimension incorporates industry coverage scope and intelligent ecosystem recombination capability (KF) to reflect adaptation and recombination potential in volatile industrial environments. To predict patent value, the study employs the CRITIC objective weighting method to determine indicator weights without subjective bias, combined with a BP neural network model that captures complex non linear relationships between indicators and value. Empirical testing using patent data from nine representative enterprises in the target industry validates the framework's applicability and effectiveness. The results indicate that the proposed framework demonstrates good evaluation performance and effectively reflects the multidimensional characteristics of patent value. The proposed system offers a more systematic tool for technology value identification and strategic decision making in the digital intelligent era. The introduction of technology space characteristics expands the analytical perspective of patent value assessment and facilitates the identification of patents occupying advantageous positions and possessing strong connectivity within technology networks. Compared with traditional approaches that mainly focus on technical attributes, the proposed framework captures the structural relationships among technologies and provides a more comprehensive understanding of patent value. Furthermore, given the high degree of technological recombination and integration in emerging technologies, the assessment of patent value should not rely solely on intrinsic technical characteristics but also consider structural features and industry context. By incorporating technology space characteristics into the evaluation framework and combining objective weighting with machine learning methods, this study provides a data-driven approach for identifying high-value patents and supporting technological decision-making. Overall, the study offers a new perspective for patent technology value assessment in the digital environment and provides a methodological reference for value evaluation in other emerging industries.

The 15th Five-Year Plan period marks a critical window for China to move beyond its position at the low and middle ends of global value chains and accelerate its emergence as a manufacturing powerhouse. While China has remained the world’s largest manufacturer for 16 consecutive years with a comprehensive industrial system, its manufacturing sector still faces structural constraints in value capture, core technology supply, organizational coordination, and converting context advantages into standards and rules. Amid the new technological revolution and industrial transformation, the imperative has shifted from expanding production capacity to transforming scale, industrial system, and application context advantages into sustainable capabilities for value creation. This study investigates the core question concerning how manufacturing-oriented artificial intelligence empowers value creation and drives China's manufacturing industry into medium-high tiers of global value chains. Methodologically, the study adopts a theoretical synthesis and framework-building approach. It integrates global value chain theory, high-value manufacturing theory, context-driven innovation theory, and research on AI-enabled manufacturing. It also draws on policy documents, macroeconomic and industrial data, and representative industrial cases to analyze the realistic foundation, structural dilemmas, mechanism of action, and strategic pathways for enhancing manufacturing value added. The study finds that the transition from scale expansion to value creation is not a simple continuation of capacity growth, equipment upgrading, or efficiency improvement. Rather, it is a systematic transformation of value sources, technological systems, organizational forms, and competitive rules. The manufacturing value creation paradigm in the digital-intelligent era takes the boost in added value as its core objective, artificial intelligence as its key technological engine, and context traction as its crucial mechanism. By embedding AI into core manufacturing links such as R&D and design, pilot testing and validation, production and manufacturing, marketing and services, and operations management, this paradigm promotes the coordinated upgrading of industrial contexts, industrial technologies, industrial organizations, and institutional support systems. As a result, it generates multiple forms of value, including technological value, product value, organizational value, context value, and rule-based value. The mechanism through which context-driven AI empowers manufacturing value creation lies in three interrelated functions of contexts. First, contexts provide a problem-definition mechanism by transforming dispersed, implicit, and complex industrial needs into clear technological tasks. Second, contexts provide a data-generation mechanism by continuously producing industrial data, process knowledge, and value relations across manufacturing activities. Third, contexts provide a value-verification mechanism by testing whether AI applications can truly shorten R&D cycles, improve pilot-scale conversion, improve production quality, reduce operating costs, extend service value, and support the formation of standards and rules. In this sense, contexts are not merely application spaces for technology deployment; they are institutionalized fields for value formation, evaluation, and diffusion. The paper further argues that context-driven AI-enabled value creation follows a dynamic evolutionary logic from efficiency upgrading to system restructuring and then to value definition. At the micro level, high-value, small-entry contexts help manufacturing firms address specific pain points such as quality inspection, process optimization, energy management, and flexible scheduling, thereby raising the bottom of the "smiling curve". At the meso level, industry-integrated contexts extend AI applications from production to the full value chain, including R&D, validation, manufacturing, services, and operations, thereby promoting system-wide industrial restructuring. At the macro level, industry-convergent contexts connect manufacturing with energy, transportation, cities, consumption, and other sectors, enabling Chinese manufacturing to move from value capture toward value definition through standards, system solutions, and global industrial ecosystems. The main contribution of this paper is threefold. First, it constructs an analytical framework for the manufacturing value creation paradigm in the digital-intelligent era and clarifies its distinction from the traditional scale-expansion paradigm. Second, it opens the "black box" of context-driven AI-enabled manufacturing value creation by explaining how AI capabilities are transformed into industrial value through core manufacturing links and coordinated factor upgrading. Third, it extends context-driven innovation theory by linking context evolution with global value chain restructuring and the reshaping of the smiling curve.

Promoting the deep integration of the digital economy with the real economy is a central pathway to high-quality economic development, and digital transformation has become a strategic priority for manufacturing firms. However, this transition is fraught with challenges. Manufacturers often grapple with limited transformation capabilities, high transformation costs, and prolonged adjustment periods before benefits materialize. Investment in intelligent equipment, industrial software, data infrastructure, and human capital exposes them to uncertainty and resource pressure. Understanding how public policy can ease these constraints is therefore both a theoretical and practical concern. Existing research on policy-driven corporate transformation has concentrated mainly on selective industrial policies, such as tax incentives and innovation subsidies, while functional industrial policies oriented toward long-term capability building have received much less attention. Intelligent manufacturing policy offers a useful setting to fill this gap. As a core component of China's Manufacturing Power Strategy, the Intelligent Manufacturing Pilot Demonstration Program guides firms toward intelligent production, digital management, and integrated manufacturing systems, providing a quasi-natural experiment for examining whether and how intelligent manufacturing policy drives enterprise digital transformation. Using Shanghai and Shenzhen A-share listed manufacturing firms from 2010 to 2023, this study treats the Intelligent Manufacturing Pilot Demonstration Program as a quasi-natural experiment and applies a staggered difference-in-differences (DID) design to identify its causal effect on enterprise digital transformation. Drawing upon the pilot lists released by the Ministry of Industry and Information Technology from 2015 to 2018, 119 pilot firms are identified as the treatment group and 1 277 non-pilot firms as the control group. The dependent variable is constructed through text mining of annual reports and measures firms' digital transformation level by aggregating word frequencies across five dimensions: artificial intelligence, big data, cloud computing, blockchain, and digital technology applications. Causal inference is strengthened through placebo tests, propensity score matching (PSM), and double machine learning (DML) estimation. The empirical results show that intelligent manufacturing policy significantly promotes enterprise digital transformation, and this finding remains robust across alternative tests and estimation strategies. The estimated effect is unlikely to be driven by random shocks, sample selection, or model specification, indicating that intelligent manufacturing policy channels firms' investment into digital technologies, intelligent production systems, and digitally oriented organizational capabilities. Mechanism analysis grounded in the resource-based view identifies two key channels through which the policy resolves resource conflicts in digital transformation. First, the policy alleviates financial constraints. Digital transformation requires large-scale resource input, and policy recognition improves access to financial resources while reducing investment pressure. Second, the policy optimizes human capital structure. Intelligent manufacturing requires employees with digital skills, technical expertise, and cross-functional collaboration capabilities, and policy support helps firms adjust workforce composition and attract high-skilled talent. Heterogeneity analysis shows that the driving effect is more pronounced among non-state-owned enterprises, technology-intensive firms, and firms located in eastern and western regions, indicating that ownership structure, technological attributes, and regional conditions jointly shape policy effectiveness. The policy also exhibits significant spatial spillover effects. It promotes digital transformation not only among pilot firms but also among non-pilot firms within the same city. This study makes several contributions. It provides new evidence for the theoretical framework linking industrial policy with micro-level enterprise transformation by examining a functional industrial policy oriented toward long-term capability building, rather than the selective instruments emphasized in prior work. Its main innovation lies in shifting the analytical focus from selective policy incentives to functional industrial policy, and in revealing both the internal resource mechanisms and external spatial spillovers through which intelligent manufacturing policy promotes digital transformation. The study also deepens theoretical understanding of policy empowerment by identifying financing constraint alleviation and human capital optimization as the underlying mechanisms within a resource-based framework. By documenting heterogeneous boundaries and spatial spillover effects, the study supports more targeted industrial policy and firms' optimization of transformation paths. For policymakers, intelligent manufacturing policy should emphasize regional demonstration, knowledge diffusion, and industrial coordination. For enterprises, participation in intelligent manufacturing initiatives should be viewed as an opportunity to build durable internal capabilities for digital transformation.