Research on the Impact of Enterprise Artificial Intelligence on Supply Chain Resilience: Empirical Evidence from Chinese Listed Companies

Abstract

Artificial intelligence (AI), as a strategic technology leading the current technological revolution and industrial transformation, functions as a pivotal catalyst for enhancing high-quality supply chain development and as the primary engine driving supply chains towards environmentally sustainable, low-carbon models. This study seeks to clarify how AI bolsters supply chain resilience through enhanced information transparency and dynamic capabilities, while examining the moderating influence of digital government in this context. Based on this, this study selected A-share listed companies from 2012 to 2023 as research samples. An entropy-based approach was utilized to develop a supply chain resilience indicator system. A two-way fixed-effects model was employed to analyze the mechanism by which business AI impacts supply chain resilience. Studies demonstrate that company artificial intelligence can markedly improve supply chain resilience. In this process, information transparency, innovative capacity, and absorptive capacity partially mediate the effect, while digital governance exerts a positive moderating influence. Heterogeneity studies indicate that artificial intelligence has a significantly greater favorable effect on supply chain resilience for high-tech corporations, manufacturing firms, growth-stage companies, mature-stage businesses, and chain master enterprises. The research findings not only reveal the impact and underlying mechanisms of enterprise artificial intelligence on supply chain resilience, offering a new perspective for systematically understanding the relationship between enterprise AI and supply chain resilience, but also provide key pathways and empirical evidence for leveraging digital technologies to build sustainable supply chains.

1. Introduction

The globe is presently experiencing a phase of increased uncertainty, characterized by the coexistence of opportunities and problems. The prevalence of black swan and gray rhino occurrences has heightened the possibility of supply chain disruptions, presenting new challenges to supply chain security and stability. The 2021 Suez Canal blockage resulted in weekly losses of $6 billion to $10 billion for global trade, diminishing annual trade growth by 0.4%. In 2022, the Russia-Ukraine conflict precipitated shortages of essential automotive components, compelling prominent German automakers to cease operations at local assembly factories and halt many manufacturing lines. In response to these issues, 14 nations, including the United States, ratified the inaugural multilateral supply chain agreement in 2023, indicating that supply chain resilience has emerged as a pivotal element in national policies and corporate competitiveness. In this context, investigating effective processes and strategies to bolster supply chain resilience is essential for ensuring supply chain security. This study examines the role of AI technology in enhancing supply chain resilience to mitigate unprecedented disruption risks in global supply chains, with the objective of offering theoretical support and practical recommendations for sustainable supply chain development.

Early research on supply chain resilience focused on defining its meaning. Soni et al. (2014) [1] pointed out that supply chain resilience is the ability of a supply chain system to maintain normal or even exceptional operations when faced with various internal and external shocks and pressures. Brusset (2016) [2] emphasized that supply chain resilience is the ability of a supply chain to resist interference and interruptions. Shi Daqian et al. (2025) [3] broke away from the single capability perspective and further decomposed it into supply chain proactive capability, supply chain reactive capability, and supply chain design capability, thereby providing a multi-dimensional analytical framework for subsequent research. Due to the complex hierarchical structure of supply chains and the deep involvement of supply chain entities, supply chains exhibit increasingly complex social holistic characteristics. Especially in the volatility, uncertainty, complexity, ambiguity (VUCA) context, various security risks occur frequently, and any vulnerability in the supply chain can quickly spread throughout the entire supply chain. Digital technology is seen as a key tool to mitigate this vulnerability. In this context, scholars have begun to focus on the enabling effects of digital technology on supply chain resilience. The digitization of supply chain networks enhances enterprises’ dynamic capabilities, shortens the production cycle in response to market demand changes, and improves supply chain agility [4]. Building on this, digital platforms improve the efficiency of resource circulation within supply chain networks, optimize decision-making behaviors of upstream and downstream enterprises, and enhance supply chain management performance [5], thereby coordinating relationships between upstream and downstream supply chain entities. Blockchain technology facilitates data sharing among supply chain members, enhancing supply chain resilience by ensuring operational visibility and transparency [6]. As a strategic emerging digital technology, artificial intelligence has become a future research direction for managing supply chain risks and restructuring supply chain resilience [7]. Artificial intelligence exerts the most profound influence on supply chain resilience compared to other technologies [8]. As part of the Industry 4.0 strategy, artificial intelligence helps integrate key areas such as supply chain network management and knowledge resource management [9], gradually penetrating all aspects of supply chain production and operations, and driving the adaptation and evolution of supply chain information systems [10]. In this process, artificial intelligence utilizes its information processing capabilities to enhance supply chain resilience [11], thereby improving both proactive and reactive resilience [12], and facilitating the progressive evolution of supply chain resilience from adaptive to sustainable, and ultimately to transformative resilience. Artificial intelligence enhances supply chain resilience at various levels by diminishing dependency, augmenting continuity, and increasing efficiency [13]. While artificial intelligence bolsters supply chain resilience, it simultaneously presents obstacles including employment displacement, privacy and security risks, and ethical dilemmas [14].

In conclusion, current research demonstrates considerable limitations. Currently, studies predominantly examine digital technology in general; however, empirical research specifically addressing the crucial area of artificial intelligence is limited. Secondly, as a binding external governance framework, digital governance may place structural limitations on the avenues through which artificial intelligence influences supply chain resilience through two channels: resource acquisition and institutional procedures. This regulating mechanism has not been adequately investigated. Therefore, what impact does AI have on supply chain resilience? What are the underlying mechanisms? Does digital governance exert a moderating effect in this process? Clarifying these questions not only provides empirical foundations for addressing practical challenges in supply chain resilience but also promotes the coordinated development of industrial and supply chains, holding significant strategic importance for achieving sustainable industrial and supply chain development.

This study’s marginal addition is the empirical examination of the factors that enable AI for supply chain resilience. It elucidates the intermediary functions of information openness and dynamic capacities, alongside the moderating influence of digital governance. Moreover, it elucidates the diverse effects of AI across various industries, business stages, and supply chain roles. This establishes rigorous theoretical and empirical underpinnings for comprehending the intricate link between AI and supply chain resilience. This study empirically investigates the influence of AI on supply chain resilience and its fundamental mechanisms, aiming to provide definitive answers to inquiries regarding AI’s capacity to enhance supply chain resilience and the methods by which it achieves this, thus augmenting the current body of research on AI and supply chain resilience. Secondly, it analyzes the moderating role of digital governance in the influence of AI on supply chain resilience, providing a novel research viewpoint for examining AI’s effects on this domain. Third, it analyzes the diverse impacts of AI on supply chain resilience, including industry characteristics, enterprise life cycle phases, and supply chain status. This elucidates the strong correlation between corporate AI adoption and its internal and external environments, enhancing our comprehension of the interplay between corporate AI and supply chain resilience. This research not only refines the theoretical framework linking AI and supply chain resilience but also provides empirical evidence for government policy formulation, holding significant practical implications for advancing AI development and building sustainable supply chains.

The subsequent structure of this paper is arranged as follows: Section 2 proposes research hypotheses through theoretical analysis; Section 3 explains the selection of variables and data sources and constructs an econometric model; Section 4 analyzes empirical results and conducts robustness tests; Section 5 discusses the practical implications of empirical results and proposes corresponding policy recommendations.

2. Theoretical Analysis and Hypothesis Formulation

2.1. Enterprise Artificial Intelligence and Supply Chain Resilience

In contrast to other digital technologies, AI, as a strategically emergent technology, has attributes such as profound penetration, extensive substitution, and significant innovation, which enhance the collaborative synergy between upstream and downstream entities within the supply chain. It enhances industries by providing collaborative operational capabilities, adaptable ecological networks, and intelligent economic activities, ensuring the stable functioning of the supply chain system and the generation of different values. The effectiveness of AI in information processing and information sharing is considered a key factor in building supply chain resilience [15]. It not only offers data support for risk identification and response throughout all supply chain segments but also functions as the foundational architecture facilitating the transition of supply chains into ecological supply networks. In the process of AI-driven transformation toward ecological supply networks, cross-border entities participate in economic activities through collaborative means. Intelligent platforms establish ecological partnerships among cross-border entities, strengthening inter-enterprise communication and cooperation to enhance information processing efficiency and drive supply chain value creation [16]. Additionally, leveraging intelligent technologies, AI can enable end-to-end collaboration across the entire supply chain. During the manufacturing phase, AI, owing to its relative superiority over human labor in performing regular activities, might create a “machine-replacing-human” substitution effect on corporate labor demand [17]. Intelligent systems efficiently complete repetitive tasks, reducing human involvement in the production process and driving the transformation of traditional production processes toward automation and intelligence. Additionally, AI can evaluate the value of different alternative solutions to achieve the most optimal expected outcomes [18]. In inventory management, AI-based customer profiling reduces inventory buildup and stockouts, thereby cutting warehousing costs. In logistics and transportation, companies utilize AI to design intelligent logistics systems that provide real-time updates on transportation status, connecting production and consumption endpoints to enhance logistics management efficiency. In the marketing phase, AI empowers the entire process across four dimensions—product, price, promotion, and channel—to help companies achieve better marketing outcomes [19]. Specifically, in the product dimension, AI technology helps companies shorten product development and testing cycles and optimize product design [20]; in the price dimension, intelligent algorithms can accurately construct consumer profiles to achieve dynamic, personalized pricing [21,22]; in promotions, by intelligently analyzing consumer preferences, intelligent technology can help marketers generate high-quality marketing copy and targeted marketing content [23,24]. In terms of channels, artificial intelligence provides new avenues for connecting upstream and downstream enterprises. Interactive intelligent technologies, with their significant advantages in pattern recognition and real-time decision-making, can enhance the engagement and loyalty of downstream enterprises [25]. With AI technology, enterprises can integrate information from multiple industries and stages to achieve collaborative operations across all supply chain stages, thereby enhancing supply chain resilience. Based on this, the following research hypothesis is proposed:

Hypothesis 1.

Enterprise artificial intelligence can enhance supply chain resilience.

2.2. The Mediating Role of Information Transparency on Enterprise Artificial Intelligence and Supply Chain Resilience

Information transparency, a fundamental aspect of information sharing and collaborative decision-making, bolsters supply chain resilience via three avenues: improving data visibility, establishing trust mechanisms, and refining dynamic collaboration. In enterprises, intelligent technologies expedite information exchange among departments, successfully reducing problems like information distortion and misrepresentation, facilitating data sharing, and improving supply chain management and optimization capabilities [26]. Data visibility has been enhanced. Management’s ability to exploit asymmetric information for non-compliant operations has been weakened, thereby curbing opportunistic behavior, reducing transactional uncertainty, strengthening trust among supply chain stakeholders, and improving supply chain resilience. The “spotlight effect” of AI technologies makes corporate management practices and market performance more visible to external markets. External markets can now monitor enterprises in real time with unprecedented depth and breadth. In this context, information barriers between upstream and downstream supply chain participants are effectively dismantled. Enterprises can establish comprehensive communication systems, accelerate the dissemination of supply chain information, and mitigate internal information asymmetry within the supply chain. Toorajipour et al. (2021) [27] found that when suppliers and manufacturers share information data, the bullwhip effect is effectively mitigated, which helps promote supply balance and resource balance. Additionally, a transparent supply chain network promotes collaboration between upstream and downstream players in the industrial chain, enabling companies to formulate resource allocation plans based on supply chain status and demand changes, dynamically adjust resource allocation, enhance supply chain dynamic collaboration capabilities, and thereby strengthen supply chain resilience. Based on this, the following research hypothesis is proposed:

Hypothesis 2.

Information transparency plays an intermediary role between artificial intelligence and supply chain resilience.

2.3. The Mediating Role of Dynamic Capabilities on Enterprise Artificial Intelligence and Supply Chain Resilience

Dynamic capabilities are crucial for building supply chain resilience as they enable enterprises to establish sustainable competitive advantages in uncertain environments. Wang and Ahmed (2007) [28] argue that dynamic capabilities consist of three elements: innovation capability, absorption capability, and adaptability. Innovation capability refers to an enterprise’s ability to develop new products, explore new markets, and align its resources with market demands. Absorption capability refers to an enterprise’s ability to identify, absorb, transform, apply, and create knowledge to solve problems. Adaptive capacity refers to a company’s ability to adjust internal elements such as organizational structure to adapt to external environments. As an emerging strategic digital technology, artificial intelligence can enhance supply chain resilience by strengthening these three elements.

Specifically, artificial intelligence possesses information capture and resource integration capabilities. It not only provides data elements for corporate innovation but also enhances a company’s ability to integrate and utilize innovative resources such as technology and knowledge, thereby improving its innovation capabilities in areas such as technology, processes, and organization. Technological innovation helps enterprises identify risks in advance through technical means and formulate countermeasures, thereby effectively enhancing the supply chain’s “shock resistance” capabilities. Process innovation equips firms with enhanced agility, allowing rapid adjustments to products and services in reaction to market demand fluctuations, thus mitigating supply-demand imbalances. Organizational innovation deepens the breadth and depth of cooperation among upstream and downstream enterprises in the supply chain, fortifies supply chain collaboration, and improves supply chain resilience. In a digital ecosystem, artificial intelligence can address the “time lag” and “partiality” issues in external knowledge acquisition by expanding information boundaries and improving knowledge processing efficiency, thereby strengthening absorption capacity. Absorption capacity helps enterprises more efficiently convert external knowledge into internal capabilities, enabling them to more sensitively perceive changes in market demand, promptly adjust business strategies, and avoid potential risks. More importantly, absorption capacity enables companies to learn from crises and further transform these lessons into practical innovation strategies [29], promoting technological synergy across the supply chain and enhancing supply chain stability. Moreover, AI utilizes real-time data acquisition and astute market demand analysis to assist companies in mitigating the delays inherent in conventional decision-making, allowing for swift responses to abrupt fluctuations in demand, policy disruptions, and other external environmental shifts, thus improving their adaptability. Adaptability allows organizations to rapidly modify their operational plans in response to industry trends, technical innovations, and market fluctuations, while enhancing production factors to save costs and ensure adequate backup resources to avert supply chain interruptions.

Hypothesis 3a.

Innovation ability plays an intermediary role between artificial intelligence and supply chain resilience.

Hypothesis 3b.

Absorptive capacity plays a mediating role between artificial intelligence and supply chain resilience.

Hypothesis 3c.

Adaptability plays a mediating role between artificial intelligence and supply chain resilience.

2.4. The Moderating Role of Digital Governance on Enterprise Artificial Intelligence and Supply Chain Resilience

Digital governance is an important means for governments to enhance their governance capabilities by leveraging digital technologies and data resources, thereby reducing institutional transaction costs, optimizing the regional business environment, and enhancing governance efficiency. From the perspective of supply chain resilience, this efficiency is achieved through data integration and collaborative mechanisms, thereby reinforcing the role of artificial intelligence in promoting supply chain resilience. On the one hand, the construction of public databases is a core component of government digital governance. Opening up and sharing such data helps reduce information asymmetry among stakeholders [30], lower institutional transaction costs, and thereby reduce collaborative friction among supply chain entities. On the other hand, government digital governance achieves cross-departmental collaborative supervision and integration of government and enterprise data resources through the construction of online platforms [31], effectively lowering technical barriers to AI application for enterprises, optimizing the regional business environment, promoting enterprise communication and collaboration, breaking down data silos between enterprises, and enabling AI to efficiently integrate data across the entire supply chain, thereby enhancing risk prediction capabilities. Therefore, in regions with higher levels of digital governance, lower institutional transaction costs and more standardized business environments are more conducive to leveraging the role of AI in enhancing supply chain resilience. Based on this, the following research hypothesis is proposed:

Hypothesis 4.

Digital governance plays a positive moderating role between artificial intelligence and supply chain resilience.

In summary, this paper introduces information transparency, dynamic capabilities and digital governance and constructs the structural framework of enterprise artificial intelligence and supply chain resilience, as shown in Figure 1.

3. Research Methodology

3.1. Construction of the Evaluation System for Supply Chain Resilience

Supply chain resilience is a multifaceted and intricate notion. When supply chains face external shocks such as sudden events, market fluctuations, and natural disasters, they need to have a diversified supplier network and effective risk warning and response mechanisms to ensure normal operations or minimize losses. This enables them to maintain their critical functions and service levels even in challenging circumstances. After a supply chain suffers severe disruption or interruption, strong organizational resilience and resource allocation capabilities can help enterprises quickly restore their original operational levels or even achieve higher-level restructuring and optimization, thereby restoring supply chain continuity and integrity in the shortest possible time. As such, relying on a single metric is insufficient to accurately and reasonably measure supply chain resilience. According to the definition of the risk lifecycle in risk management theory, resilience building must span the three stages of prevention, resistance, and recovery. Drawing on the resilience building framework proposed by Zhao et al. (2023) [32], supply chain resilience is decomposed into three dimensions: pre-impact prevention capability, during-impact resistance capability, and post-impact recovery capability.

First, in terms of supply chain resilience, proactive measures can reduce the likelihood of supply chain disruptions. Supply chain transparency, supplier concentration, customer concentration, and supply-demand matching are used as metrics. Supply chain transparency reflects the adequacy of information sharing between upstream and downstream entities. The higher the transparency, the stronger the risk identification and collaborative response efficiency. It is measured by the proportion of transaction amounts from the top five suppliers and customers that explicitly disclose transaction information. Supplier concentration serves as an inverse indicator of preventive capability, reflecting the supply chain’s reliance on upstream core suppliers. Higher supplier concentration indicates stronger path dependence on a few suppliers. If core suppliers experience disruptions, the supply chain faces significantly increased risks. This is measured by the ratio of procurement expenditures from the top five suppliers to total procurement expenditures. Customer concentration reflects the degree of concentration at the downstream demand end. High customer concentration indicates that the company is overly reliant on a few core customers. This high dependence amplifies the uncertainty risks at the downstream demand end and increases the potential probability of supply chain disruptions. It is measured by the ratio of sales from the top five customers to annual total sales. Supply-demand matching emphasizes the adaptability of the supply chain’s supply capacity to market demand, with the core objective of reducing supply-demand imbalances and lowering the risk of supply disruptions caused by resource misallocation. The smaller the deviation between production fluctuations and demand fluctuations, the higher the supply-demand matching degree, resulting in lower risks of sudden disruptions caused by oversupply or undersupply, and stronger preventive capabilities. This is measured by the deviation between production fluctuations and demand fluctuations [33].

Secondly, in terms of supply chain resilience, the ability of the supply chain system to maintain stable operations during shocks is key to risk mitigation. Financial strength, human capital, accumulated redundant resources, and technical support are used as metrics for measurement. Financial strength serves as the material foundation for supply chain risk mitigation, providing enterprises with the flexibility to respond to shocks. The larger a company’s total assets, the more emergency resources it can obtain through collateralization or other means, thereby reducing the probability of supply chain disruptions. This is measured by taking the logarithm of total assets. Human capital is not only an important resource for companies to gain a competitive edge in the market but also the vehicle for implementing risk response strategies. Highly educated personnel typically possess stronger emergency coordination and problem-solving capabilities, directly impacting the efficiency of strategy implementation. This is measured by the proportion of personnel with a bachelor’s degree or higher. Sedimentary redundant resources serve as strategic reserves for supply chain risk mitigation. They can be rapidly activated during crises to provide buffer time for the supply chain, thereby reducing disruption risks. These resources function as pre-established buffers against sudden shortages. Their establishment and maintenance require additional investment, measured by the ratio of management expenses to operating revenue. Technical support is a system that leverages digital tools and related technologies to enhance operational efficiency and risk management capabilities. Technical personnel are the primary users of digital tools, and a higher proportion of technical personnel leads to better technical empowerment and resilience. This is measured by the proportion of technical personnel.

Finally, in terms of supply chain resilience, the core lies in maintaining optimal operational performance quickly after a supply chain disruption [34], with financial relationships, supply chain recovery speed, and performance volatility selected as metrics for measuring supply chain resilience. Financial relationships reflect the intensity of liquidity constraints faced by enterprises after supply chain disruptions. If downstream customers occupy a large portion of upstream suppliers’ funds, suppliers may face significant accounts receivable pressure, leading to a shortage of free cash flow available for repairing damaged capacity, thereby delaying the supply chain recovery process and weakening supply chain resilience. This is measured using the logarithm of the ratio of accounts receivable to revenue. The faster the supply chain recovers, the shorter the time it takes for the supply chain to return to normal operations after an interruption, enabling losses to be contained within a shorter cycle. This is measured using inventory turnover rate. The degree of deviation in business performance effectively reflects the dynamic changes on both the supply and demand sides of the supply chain and, to some extent, reflects supply chain resilience. Following the approach of Martin (2012) [35], a sensitivity index (SI) for relative changes is constructed to measure performance volatility. The formula is as follows:

SI_i,t = (Inx_i,t − Inx_i,t−1) − (InX_i,t − InX_i,t−1)

(1)

X represents the total business revenue of the industry in which company, and x represents the main business revenue of company. The lower the performance deviation index, the smaller the fluctuations in operating performance and the stronger the recovery capacity.

To systematically quantify supply chain resilience, we first standardize the data based on the 11 indicators across the three dimensions mentioned above. We then use the entropy method to determine the weights of the indicators and ultimately construct a comprehensive supply chain resilience index to measure supply chain resilience. The specific indicator system is shown in

Table 1. Measurement indicators of supply chain resilience.

Target Layer	First-Level Indicators	Secondary Indicators	Specific Indicators	Weight	Attributes
Supply Chain Resilience	Supply Chain Prevention Capability	supply chain transparency	The proportion of the transaction volume of large suppliers and customers with clear disclosure names to the total transaction volume of the top five suppliers and customers.	0.293	+
		supplier concentration	Ratio of top five suppliers’ purchases to total annual purchases	0.015	−
		customer concentration	Ratio of top five customers’ sales to total annual sales	0.015	−
		Supply and demand matching degree	Log(abs(Net inventory value − L.Net inventory value)/L.Net inventory value)	0.019	−
	Supply Chain Resistance Capability	Financial strength	Logarithm of total assets	0.030	+
		Human capital	Proportion of people with bachelor’s degree or above	0.062	+
		Precipitated redundancy resources	The ratio of management cost to operating income	0.064	+
		Technical support	Proportion of technical personnel	0.070	+
	Supply Chain Resilience Capability	Financial relationship	The ratio of accounts receivable to operating income is logarithm	0.040	−
		Recovery speed	turnover of inventory	0.370	+
		Performance volatility	SIi,t = (Inxi,t − Inxi,t−1) − (InXi,t − InXi,t−1)	0.019	+

.

3.2. Sample Selection and Data Sources

Based on the availability of research content and data, Chinese A-share listed companies from 2012 to 2023 were selected as the research sample. The research data was obtained from the CSMAR database and the Giant Tide Information Network. To enhance the reliability of the research conclusions, the following treatments were applied to the sample: (1) Companies classified as ST, *ST, or PT during the sample period were excluded; (2) Financial industry companies were excluded; (3) Missing values for key variables were excluded. After processing, 10,489 sample data points were obtained. To eliminate the impact of outliers, all continuous variables were trimmed at the 1% and 99% percentiles. Data analysis was conducted using Stata 17.0 software.

3.3. Variable Definition and Index Construction

3.3.1. Dependent Variable: Supply Chain Resilience (SCR)

In this study, we measure the level of supply chain resilience by constructing a multi-dimensional evaluation index system for supply chain resilience.

3.3.2. Independent Variable: Enterprise Artificial Intelligence (AI)

The development of enterprise AI depends on the decision-making orientation of management, and the description of AI in the annual reports of listed companies to some extent reflects the development orientation of enterprise AI. The more AI-related words appear in the annual report, the stronger the company’s willingness to develop AI and the higher its level of AI development. Therefore, referring to the research of Zhang and Wei (2024) [36], we use AI-related keywords in annual reports to measure AI development levels. First, we use text analysis to extract AI-related keywords from listed companies’ annual reports. Second, we add 1 to the frequency of AI-related keywords and take the natural logarithm as a measure of AI development levels.

3.3.3. Mediating Variable: Information Transparency (DA) and Dynamic Capabilities

Supply chains are built on a complex network of division of labor, with transactions between companies serving as the connecting link, enabling each company to leverage its strengths within the supply chain network. Increased corporate transparency enhances the efficacy of the trust mechanism within the supply chain network, hence fostering sustainable development of the supply chain. Referring to the research by Liu et al. (2015) [37], research report attention is used to measure information transparency, with higher research report attention indicating stronger company information transparency.

Current research predominantly employs questionnaire surveys to assess dynamic capabilities; however, cross-sectional data cannot fully reflect the evolution of a company’s dynamic capabilities. Therefore, drawing on the research of Yang (2023) [38], dynamic capabilities are measured from three dimensions: innovation capability, absorption capability, and adaptability. Innovation capability (IC) is measured using the logarithm of annual R&D expenditure. Absorption capability (AC) is measured using the ratio of annual R&D expenditure to revenue. Adaptive capability (YC) is measured using the coefficient of variation of three types of expenditure: R&D, capital, and advertising.

3.3.4. Moderating Variable: Digital Governance (DG)

Digital governance is a process in which multiple stakeholders use digital technology to coordinate and optimize public services and improve governance efficiency. The fundamental nature of the Information for the People pilot policy aligns closely with the key attributes of digital governance. This represents the practical application of digital governance within public services. Therefore, The Information for the People pilot policy serves as a variable in digital governance. If the city in which the enterprise is situated enacted this policy in 2014 or subsequently, the value is 1; if not, the value is 0.

3.3.5. Control Variables

Referring to relevant literature, the following control variables were selected: debt-to-equity ratio (Lev), equity balance (OC), management shareholding ratio (MO), property rights nature (SOE), return on assets (ROA), financial distress (ZCORE), and revenue growth rate (RGR). In addition, industry and year fixed effects were controlled for. Variable definitions are shown in

Table 2. Variable definitions.

	Code	Definition
Dependent Variable	SCR	As mentioned above
Independent variables	AI	ln(word frequency + 1)
Mediating Variable	DA	The number of tracking analysis of research papers
	IC	ln(R&D investment)
	AC	R&D expenditure/operating income
	YC	The coefficient of variation of the three main expenditures of R&D, capital and advertising
Moderating variable	DG	Information benefiting people policy
Control variables	Lev	total liabilities/total assets
	OC	The proportion of the second to fifth largest shareholder/the proportion of the largest shareholder
	MO	The number of shares held by directors and supervisors/the total number of shares
	SOE	State-owned enterprises are assigned a value of 1, otherwise it is 0.
	ROA	Net profit/total assets
	ZSCORE	ZScore model
	RGR	(Current operating income − previous operating income)/previous operating income
	Industry	Industry dummy variable
	Year	Year dummy variable

.

3.4. Model Design

To explore the mechanism of enterprise artificial intelligence’s role in supply chain resilience, this paper constructs the following model:

SCR_i,t = α₀ + α₁AI_i,t + α₂Control_i,t + ∑Industry + ∑Year + ϵ_i,t

(2)

Among them, the subscript i represents the enterprise, t represents the year, and SCR_i,t represents the supply chain resilience of enterprise i in the tth year; AI_i,t represents the development level of artificial intelligence in the tth year of enterprise i; Control_i,t represents control variables.

4. Empirical Results and Analysis

4.1. Descriptive Statistics

Table 3. Descriptive statistics.

	Number	Mean	S.D.	Min	Max
SCR	10,489	0.151	0.070	0.068	0.455
AI	10,489	1.120	1.363	0.000	4.836
DA	10,489	20.560	24.543	1.000	118.000
IC	10,489	9.123	1.385	5.255	12.933
AC	10,489	0.056	0.058	0.000	0.310
YC	10,489	−1.060	0.317	−1.732	−0.258
DG	10,489	0.551	0.497	0.000	1.000
Lev	10,489	0.398	0.190	0.056	0.897
OC	10,489	0.774	0.590	0.034	2.773
MO	10,489	16.722	19.978	0.000	67.760
SOE	10,489	0.257	0.437	0.000	1.000
ROA	10,489	0.046	0.057	−0.265	0.192
ZSCORE	10,489	5.395	5.847	−0.082	36.305
RGR	10,489	0.309	0.676	−0.666	4.997

reports the descriptive statistics of the main variables. Among them, the maximum value of supply chain resilience is 0.455, the minimum value is 0.068, and the standard deviation is 0.070, indicating that there are differences in supply chain resilience among different enterprises, and the overall development level is relatively low, with significant room for improvement. The maximum and minimum values for enterprise artificial intelligence are 4.836 and 0.000, respectively, and they exhibit the characteristic of “large standard deviation and small mean,” indicating significant differences in artificial intelligence levels among different enterprises. The descriptive statistical results for the remaining variables are consistent with existing research.

4.2. Benchmark Regression Analysis

Using Model (1) to test the impact of artificial intelligence on supply chain resilience,

Table 4. Benchmark regression.

	(1) SCR	(2) SCR
AI	0.0034 *** (0.001)	0.0035 *** (0.001)
Lev		−0.0049 (0.007)
OC		0.0020 (0.002)
MO		−0.0001 (0.000)
SOE		0.0073 ** (0.003)
ROA		−0.0481 *** (0.016)
ZSCORE		0.0010 *** (0.000)
RGR		0.0022 (0.002)
Industry	YES	YES
Year	YES	YES
Cons	0.1476 *** (0.002)	0.1431 *** (0.004)
N	10,489	10,489
R2	0.256	0.264

Note: *** p < 0.01, ** p < 0.05; ( ) is the robust standard error clustered to the enterprise level. The variables are defined as indicated in Table 2.

presents the results of the benchmark regression analysis. Column (1) controls only for year and industry, with the coefficient for artificial intelligence at 0.0034, significantly positive at the 1% level. Column (2) includes additional control variables, with the coefficient for artificial intelligence at 0.0035, still significantly positive at the 1% level, indicating that the development of artificial intelligence in enterprises has a significant promotional effect on supply chain resilience. From an economic perspective, an increase of 1 unit in corporate AI is associated with an approximate 0.35% enhancement in supply chain resilience. Therefore, both statistically and economically, corporate AI enhances supply chain resilience, and research hypothesis H1 is supported. Enterprise artificial intelligence methodically enhances supply chain capabilities across three dimensions—forecasting, resilience, and recovery—assisting firms in developing robust supply chains. As global uncertainties intensify, companies proficient in using AI are more effectively positioned to bolster supply chain resilience and achieve high-quality supply chain development.

4.3. Robustness Tests

4.3.1. Replace Independent Variables

Management Discussion and Analysis (MD&A) reflects investors’ understanding of a company’s future development direction. Consequently, we employ the logarithm of the total frequency of the term artificial intelligence in the MD&A as an indicator of the company’s level of artificial intelligence development and perform a robustness test. The test results are shown in

Table 5. Robustness test results.

	Replace Explanatory Variables		Replace the Explained Variable
	(1) SCR	(2) SCR	(3) SCR	(4) SCR
AI	0.0030 *** (0.001)	0.0023 *** (0.001)	−5.3176 ** (2.416)	0.0880 *** (0.010)
Lev	−0.0044 (0.007)	−0.0118 * (0.007)	−16.5932 (18.354)	−0.0953 (0.063)
OC	0.0021 (0.002)	0.0022 (0.002)	1.8699 (3.911)	0.0422 *** (0.016)
MO	−0.0001 (0.000)	−0.0000 (0.000)	0.2521 (0.159)	0.0010 ** (0.000)
SOE	0.0073 ** (0.003)	0.0055 ** (0.003)	−8.8623 (8.019)	0.1138 *** (0.026)
ROA	−0.0480 *** (0.016)	−0.0531 *** (0.017)	−371.2694 *** (67.902)	−0.7445 *** (0.159)
ZSCORE	0.0010 *** (0.000)	0.0011 *** (0.000)	3.0215 *** (0.666)	0.0176 *** (0.002)
RGR	0.0023 (0.002)	0.0026 (0.002)	30.6700 *** (4.532)	0.1389 *** (0.013)
Industry	YES	YES	YES	YES
Year	YES	YES	YES	YES
Cons	0.1439 *** (0.004)	0.1350 *** (0.005)	155.5298 *** (11.620)	−0.2459 *** (0.040)
N	10,489	9345	10,489	10,489
R2	0.264	0.272	0.381	0.622

Note: *** p < 0.01, ** p < 0.05, * p < 0.1; ( ) is the robust standard error clustered to the enterprise level. The variables are defined as indicated in Table 2.

Column (1), where the coefficient for artificial intelligence is 0.0030, which is significantly positive at the 1% level. Additionally, companies need to take specific actions in their actual operations to achieve tangible results. Therefore, investments in artificial intelligence intangible assets measure the effectiveness of artificial intelligence actions. The test results are shown in Column (2). The coefficient for artificial intelligence is 0.0023, which is significantly positive at the 1% level, indicating that artificial intelligence development actions significantly enhance supply chain resilience.

4.3.2. Replace the Dependent Variable

To further enhance the robustness of the research conclusions, inventory turnover days were used for testing. The shorter the inventory turnover days, the stronger the supply chain resilience. The test results are shown in Table 3. The coefficient for artificial intelligence is −5.3176, which is significantly negative at the 5% level. That is, the higher the level of artificial intelligence in an enterprise, the shorter the inventory turnover time, and the stronger the supply chain resilience, consistent with the test results mentioned above. AI significantly reduces inventory turnover time through precise forecasting and intelligent scheduling, hence minimizing capital investment and operating hazards. This efficient inventory system provides the supply chain with strong resilience to respond quickly to interruptions.

To examine whether the results of supply chain resilience measurement are influenced by the empowerment method, this study further employs principal component analysis for robustness testing based on the indicators constructed using the entropy method. The test results are shown in Column (4). The coefficient for artificial intelligence is 0.0880, which is significantly positive at the 1% level, consistent with the benchmark regression results. This indicates that the conclusions are robust.

4.4. Endogeneity Test

To address potential endogeneity issues in the research process, such as sample selection bias, omitted variables, and reverse causality, we employed Heckman two-stage random selection, interactive fixed effects, and lagged dependent variables for endogeneity treatment.

4.4.1. Heckman Two-Stage Procedure

The Heckman two-stage method is an important technique for addressing sample selection bias. This paper selects the average level of artificial intelligence development within the same province and industry as an exclusionary variable. On the one hand, companies within the same industry often face similar market environments and competitive landscapes, and the level of artificial intelligence development within the same province can influence the artificial intelligence development of companies within the same region, thereby meeting the endogeneity condition. On the other hand, the AI development levels of the same province and industry are determined by macroeconomic factors and do not directly affect the AI development of the firm itself, satisfying the exogeneity condition. First, the mean AI development levels of the same province and industry are selected as the instrumental variable and introduced into the first-stage Probit regression, with the inverse Mills ratio (IMR) calculated. Second, the inverse Mills ratio is included as a control variable in the second-stage regression model. The results of Heckman’s first-stage test indicate that the coefficient of the exclusion variable is 0.6754, which is significantly positive at the 1% level. The results of the second-stage test indicate that the coefficient of AI is 0.0032, which is significantly positive at the 1% level, and the inverse Mills ratio is not significant, indicating that there is no sample selection bias issue, and the research conclusions are reliable.

4.4.2. Increasing the Interaction Fixed Effect

To eliminate endogeneity issues caused by omitted variables, we further controlled for the fixed effects of the year-industry interaction. The test results are shown in

Table 6. Endogeneity test results.

	Heckman Two-Stage Procedure		Interactive Fixed Effects	Lagged Explanatory Variable
	(1) AI	(2) SCR	(3) SCR	(4) SCR
AI		0.0032 *** (0.001)	0.0037 *** (0.001)
IV	0.6754 *** (0.041)
IMR		−0.0065 (0.005)
L.AI				0.0024 * (0.001)
Lev	0.3483 *** (0.115)	−0.0043 (0.007)	−0.0065 (0.007)	−0.0101 (0.009)
OC	0.0540 ** (0.026)	0.0018 (0.002)	0.0021 (0.002)	0.0006 (0.002)
MO	−0.0009 (0.001)	−0.0001 (0.000)	−0.0001 (0.000)	0.0000 (0.000)
SOE	0.0257 (0.040)	0.0061 ** (0.003)	0.0072 ** (0.003)	0.0064 * (0.004)
ROA	0.5325 * (0.306)	−0.0505 *** (0.017)	−0.0557 *** (0.018)	−0.0435 ** (0.021)
ZSCORE	0.0024 (0.004)	0.0011 *** (0.000)	0.0010 *** (0.000)	0.0007 ** (0.000)
RGR	0.0658 ** (0.027)	0.0020 (0.002)	0.0024 (0.002)	0.0017 (0.002)
Industry	YES	YES	YES	YES
Year	YES	YES	YES	YES
Year # Industry	NO	NO	YES	NO
Cons	−1.1472 (1.390)	0.1483 *** (0.007)	0.1440 *** (0.005)	0.1522 *** (0.006)
N	10,310	10,310	10,489	5970
R2		0.251	0.310	0.291

Note: *** p < 0.01, ** p < 0.05, * p < 0.1; ( ) is the robust standard error clustered to the enterprise level. The variables are defined as indicated in Table 2.

(3). The coefficient of artificial intelligence is 0.0037, which is significantly positive at the 1% level, indicating that the above conclusions are robust.

4.4.3. Lagged Explanatory Variable

To mitigate potential endogeneity issues arising from reverse causality, the lagged explanatory variable was included in the regression. The test results are shown in Column (4) of Table 6. The coefficient for artificial intelligence is 0.0024 and is positively significant at the 10% level, indicating that even after accounting for reverse causality, artificial intelligence still exerts a significant positive impact on the supply chain.

4.5. Analysis of Mediating Effect

To verify the mediating effect of information transparency and dynamic capabilities on the channel, a recursive equation was used. The results are shown in

Table 7. Mediation effect test results.

	(1) DA	(2) SCR	(3) IC	(4) SCR	(5) AC	(6) SCR	(7) YC	(8) SCR
AI	2.0671 *** (0.368)	0.0033 *** (0.001)	0.1814 *** (0.018)	0.0029 *** (0.001)	0.0042 *** (0.001)	0.0028 *** (0.001)	0.0169 *** (0.004)	0.0035 *** (0.001)
DA		0.0001 * (0.000)
IC				0.0031 *** (0.001)
AC						0.1616 *** (0.029)
YC								0.0006 (0.003)
Lev	32.6038 *** (2.562)	−0.0074 (0.007)	2.0823 *** (0.147)	−0.0114 (0.007)	−0.0491 *** (0.005)	0.0030 (0.007)	−0.0840 *** (0.032)	−0.0049 (0.007)
OC	2.1462 *** (0.668)	0.0019 (0.002)	0.0820 ** (0.034)	0.0018 (0.002)	0.0031 ** (0.001)	0.0015 (0.002)	0.0022 (0.008)	0.0020 (0.002)
MO	−0.0409 * (0.021)	−0.0001 (0.000)	−0.0075 *** (0.001)	−0.0000 (0.000)	0.0001 *** (0.000)	−0.0001 * (0.000)	0.0004 (0.000)	−0.0001 (0.000)
SOE	−2.1827 * (1.146)	0.0075 *** (0.003)	0.4201 *** (0.061)	0.0060 ** (0.003)	−0.0010 (0.002)	0.0075 *** (0.003)	0.0240 * (0.013)	0.0073 ** (0.003)
ROA	162.2967 *** (7.438)	−0.0604 *** (0.017)	4.0800 *** (0.342)	−0.0609 *** (0.017)	−0.1960 *** (0.017)	−0.0165 (0.017)	0.0480 (0.074)	−0.0482 *** (0.016)
ZSCORE	0.4714 *** (0.077)	0.0010 *** (0.000)	−0.0066 (0.004)	0.0011 *** (0.000)	0.0019 *** (0.000)	0.0007 *** (0.000)	0.0020 ** (0.001)	0.0010 *** (0.000)
RGR	−0.6327 (0.593)	0.0023 (0.002)	−0.0658 ** (0.029)	0.0024 (0.002)	0.0052 *** (0.001)	0.0014 (0.001)	0.0033 (0.007)	0.0022 (0.002)
Industry	YES	YES	YES	YES	YES	YES	YES	YES
Year	YES	YES	YES	YES	YES	YES	YES	YES
Cons	−5.0010 *** (1.592)	0.1435 *** (0.004)	7.9128 *** (0.089)	0.1184 *** (0.008)	0.0636 *** (0.003)	0.1328 *** (0.005)	−1.0738 *** (0.021)	0.1438 *** (0.005)
N	10,489	10,489	10,489	10,489	10,489	10,489	10,489	10,489
R2	0.203	0.265	0.406	0.267	0.507	0.273	0.226	0.264

Note: *** p < 0.01, ** p < 0.05, * p < 0.1; ( ) is the robust standard error clustered to the enterprise level. The variables are defined as indicated in Table 2.

. Column (1) shows that artificial intelligence has a positive effect on information transparency at the 1% level. In column (2), which includes information transparency, the coefficient for artificial intelligence is 0.0033 and the coefficient for information transparency is 0.0001, both of which pass the statistical significance test. Compared with the baseline regression results, the regression coefficient of artificial intelligence decreases after including information transparency, indicating that the mediating effect of information transparency exists, and research hypothesis H2 is supported. Column (3) shows that AI has a positive impact on innovation capability at the 1% level; in column (4), which includes innovation capability, the coefficient for AI is 0.0029, and the coefficient for innovation capability is 0.0031, both of which pass the significance test at the 1% level. Compared with the benchmark regression results, the regression coefficient for AI decreases after including innovation capability, indicating that the mediating effect of innovation capability exists, and research hypothesis H3a is established. Column (5) shows that AI positively influences absorption capacity at the 1% level; in column (6) with absorption capacity included, the coefficient for AI is 0.0028, and the coefficient for absorption capacity is 0.1616, both passing the significance test at the 1% level. Compared with the benchmark regression results, the regression coefficient for AI decreases after including absorption capacity, indicating the existence of the mediating effect of absorption capacity, and research hypothesis H3b is established. Column (7) shows that artificial intelligence positively influences adaptability at the 1% level. In column (8), which includes adaptability, the coefficient for artificial intelligence is 0.0035, which is significantly positive at the 1% level, but adaptability does not pass the significance test, indicating that the mediating effect of adaptability does not exist, and research hypothesis H3c is not established.

To ensure the accuracy of the mediation effect testing mechanism, Bootstrap and Sobel tests were further employed. The regression results under the 95% confidence interval with bias correction, based on 1000 independent samples, are presented. The indirect effect of artificial intelligence on supply chain resilience via information transparency is 0.0001, with a confidence interval of [0.0000, 0.0002], which does not encompass 0. This suggests that information transparency serves as a partial mediator in the relationship between artificial intelligence and supply chain resilience. The indirect effect of artificial intelligence on supply chain resilience via innovation capability is 0.0005, with a confidence interval of [0.0023, 0.0007], which does not encompass 0. Innovation capability serves as a partial mediator in the relationship between artificial intelligence and supply chain resilience. The indirect effect of AI on supply chain resilience via absorption capacity is 0.0006, with a confidence interval of [0.0004, 0.0008], which does not encompass 0. Absorption capacity partially mediates the relationship between AI and supply chain resilience. The indirect effect of AI on supply chain resilience via adaptability is 0.0000, with a confidence interval encompassing 0. This suggests that the mediating effect of adaptability is absent. The Sobel test shows that the mediating effect of information transparency accounts for 4.5%, with Z = 2.666 (p = 0.008); the mediating effect of innovation capability accounts for 16.3%, with Z = 5.293 (p = 0.000); the mediating effect of absorptive capacity accounts for 19.5%, with Z = 7.261 (p = 0.000), indicating that information transparency, innovation capability, and absorptive capacity play a mediating role. The mediating effect of adaptability is 0.3%, with Z = 0.301 (p = 0.763), suggesting that adaptability does not serve as a mediator. The test results are consistent with the preceding text, and the research conclusions are reliable.

4.6. Analysis of Regulatory Effect

To investigate the impact of digital governance on the path of artificial intelligence enhancing supply chain resilience, a moderation effect test was further conducted. The test results are shown in

Table 8. The moderating effect of digital governance.

Variable	(1) SCR
AI	0.0029 *** (0.001)
DG	0.0032 (0.002)
AI × DG	0.0037 ** (0.002)
Lev	−0.0051 (0.007)
OC	0.0022 (0.002)
MO	−0.0001 (0.000)
SOE	0.0073 ** (0.003)
ROA	−0.0466 *** (0.016)
ZSCORE	0.0010 *** (0.000)
RGR	0.0022 (0.002)
Industry	YES
Year	YES
Cons	0.1462 *** (0.004)
N	10,489
R2	0.266

Note: *** p < 0.01, ** p < 0.05; ( ) is the robust standard error clustered to the enterprise level. The variables are defined as indicated in Table 2.

. The coefficient for enterprise artificial intelligence is 0.0029, indicating a significant positive effect at the 1% level. The interaction term coefficient between AI and digital governance is 0.0037, demonstrating a significant positive effect at the 5% level. This indicates that digital governance positively moderates the promotional effect of AI on supply chain resilience, meaning that the higher the level of digital governance, the stronger the promotional effect of AI on supply chain resilience.

5. Heterogeneity Analysis

5.1. Industry Characteristics

The high-tech industry is characterized by technological innovation and cutting-edge advancements, giving it a natural advantage in the application of emerging technologies. Therefore, referencing the “Classification of Strategic Emerging Industries (2012) (Trial)” [39], the research sample was divided into high-tech industries and non-high-tech industries to explore the heterogeneous impact of artificial intelligence on supply chain resilience. The test results are shown in

Table 9. Heterogeneity test of industry characteristics.

	High-Tech Industry	Non-High-Tech Industries	Manufacturing Industry	Non-Manufacturing Industry
	(1) SCR	(2) SCR	(3) SCR	(4) SCR
AI	0.0040 *** (0.001)	0.0020 (0.002)	0.0043 *** (0.001)	0.0022 (0.002)
Lev	−0.0025 (0.008)	−0.0065 (0.013)	−0.0003 (0.007)	−0.0168 (0.016)
OC	0.0036 * (0.002)	−0.0013 (0.003)	0.0052 *** (0.002)	−0.0052 (0.004)
MO	−0.0001 (0.000)	−0.0001 (0.000)	−0.0000 (0.000)	−0.0002 (0.000)
SOE	0.0060 * (0.003)	0.0098 * (0.006)	0.0093 *** (0.003)	0.0021 (0.007)
ROA	−0.0651 *** (0.019)	0.0063 (0.032)	−0.0642 *** (0.018)	−0.0039 (0.035)
ZSCORE	0.0011 *** (0.000)	0.0010 * (0.001)	0.0007 *** (0.000)	0.0023 *** (0.001)
RGR	0.0039 ** (0.002)	−0.0005 (0.002)	0.0064 *** (0.002)	−0.0021 (0.002)
Industry	YES	YES	YES	YES
Year	YES	YES	YES	YES
Cons	0.1391 *** (0.005)	0.1482 *** (0.009)	0.1243 *** (0.004)	0.1927 *** (0.012)
N	7274	3215	7545	2944
R2	0.260	0.279	0.112	0.214

Note: *** p < 0.01, ** p < 0.05, * p < 0.1; ( ) is the robust standard error clustered to the enterprise level. The variables are defined as indicated in Table 2.

, columns (1) and (2). In high-tech industries, the coefficient for artificial intelligence is 0.0040, indicating a significant positive effect at the 1% level. However, in non-high-tech industries, the promotional effect of artificial intelligence on supply chain resilience is not significant. A possible reason is that high-tech industries have a solid data foundation and high levels of informatization, providing a foundation for the development of artificial intelligence in enterprises. Additionally, high-tech industries have a high degree of acceptance and integration of AI technology, with strong reliance on AI in production and R&D processes, making it easier to establish advanced supply chain management systems. This allows AI’s enabling role to penetrate more deeply. In contrast, non-high-tech industries face higher technical barriers and relatively lagging information technology development, limiting the effectiveness of AI technology applications.

To further explore the differentiated impact of AI on supply chain resilience under industry heterogeneity, industries were categorized into manufacturing and non-manufacturing sectors based on the National Economic Industry Classification. The test results are shown in columns (3) and (4) of Table 9. In the manufacturing sector, the coefficient for AI is 0.0043, which is significantly positive at the 1% level. However, in the non-manufacturing sector, the promotional effect of AI is not significant. Possible reasons are that manufacturing supply chains are long and complex, with multi-stage dependencies, and they are highly vulnerable to factors such as natural disasters and logistics disruptions. AI can create management platforms that facilitate information sharing and intelligent collaboration, allowing for real-time monitoring of supply chain operations and adaptable modifications to maintain the stability of the supply chain system. Non-manufacturing supply chains depend more on information and service flows, which leads to reduced collaboration challenges and consequently limits the utility of AI in these supply chains. Furthermore, non-manufacturing sectors exhibit greater production substitutes, with supply chain resilience being more reliant on process flexibility or policy adaptability, resulting in a comparatively weaker correlation between AI and supply chain resilience.

5.2. Enterprise Life Cycle

Business development progresses through a lifecycle, exhibiting unique characteristics at each stage, which may lead to differing effects of artificial intelligence on supply chain resilience. Existing research employs various methods to classify the enterprise life cycle, with the composite score discrimination method being the most common. Therefore, drawing on the research of Dickinson (2011) [40], this study uses the composite score of four variables—revenue growth rate, retention rate, capital expenditure rate, and company age—to measure the enterprise life cycle. Specifically, the four variables are ranked and a composite score is calculated. The top 1/3 with the highest scores are classified as growth-stage companies, the middle 1/3 as mature-stage companies, and the bottom 1/3 as decline-stage companies. Virtual variables are used to define the enterprise life cycle, with 3 for growth stage, 2 for mature stage, and 1 for decline stage.

The test results are shown in columns (1) to (3) of

Table 10. Heterogeneity test of enterprise life cycle and supply chain status.

	Growth Period	Maturity Period	Decline Period	Chain Master Enterprise	Non-Chain Master Enterprise
	(1) SCR	(2) SCR	(3) SCR	(4) SCR	(5) SCR
AI	0.0034 ** (0.002)	0.0034 ** (0.001)	0.0029 ** (0.001)	0.0044 * (0.002)	0.0033 *** (0.001)
Lev	0.0043 (0.013)	−0.0092 (0.009)	−0.0081 (0.011)	−0.0119 (0.018)	−0.0059 (0.007)
OC	0.0021 (0.003)	0.0033 (0.002)	−0.0002 (0.003)	−0.0078 * (0.004)	0.0042 ** (0.002)
MO	0.0001 (0.000)	−0.0001 (0.000)	−0.0002 ** (0.000)	0.0000 (0.000)	−0.0001 (0.000)
SOE	0.0086 (0.006)	0.0056 (0.004)	0.0055 (0.004)	−0.0035 (0.005)	0.0093 *** (0.003)
ROA	−0.1342 *** (0.034)	−0.0606 *** (0.022)	0.0173 (0.026)	−0.1301 *** (0.041)	−0.0442 ** (0.018)
ZSCORE	0.0013 *** (0.000)	0.0006 ** (0.000)	0.0014 *** (0.000)	0.0017 ** (0.001)	0.0010 *** (0.000)
RGR	0.0046 * (0.003)	0.0011 (0.002)	0.0015 (0.002)	0.0019 (0.004)	0.0027 * (0.002)
Industry	YES	YES	YES	YES	YES
Year	YES	YES	YES	YES	YES
Cons	0.1320 *** (0.008)	0.1479 *** (0.006)	0.1475 *** (0.007)	0.1645 *** (0.013)	0.1398 *** (0.005)
N	2269	4089	4131	1913	8576
R2	0.296	0.303	0.277	0.367	0.259

Note: *** p < 0.01, ** p < 0.05, * p < 0.1; ( ) is the robust standard error clustered to the enterprise level. The variables are defined as indicated in Table 2.

. Companies in the growth and mature stages both have an AI coefficient of 0.0034, which is significantly positive at the 1% level, while the impact of AI on companies in the decline stage is not significant. A possible reason is that companies in the growth and mature stages have well-developed digital infrastructure, allowing AI technology to fully leverage its “enabling” role during this phase. Furthermore, firms in the growth and maturity stages typically invest greater resources in technological innovation. AI, as a pivotal technology, is likely to garner management attention and support, thereby establishing a robust technological foundation for the sustainable development of the supply chain. Companies in the decline phase have established specific technological path dependencies and entrenched effects. While AI technology has the potential to improve returns, the significant research and development costs lead companies to exercise caution in investing in new technologies, consequently influencing the extent of AI technology application.

5.3. Supply Chain Status

As the leading enterprise in the supply chain, the chain master enterprise is the core of the supply chain system. Chain master enterprises can be identified from three dimensions: enterprise scale, market power, and industry barriers. Specifically, a virtual variable for enterprise scale (SZ) is constructed based on the median of year-end total assets. If an enterprise’s asset scale exceeds the median, SZ is set to 1; otherwise, it is set to 0. A virtual variable for market power (MP) is constructed based on the median of the ratio of an enterprise’s revenue to industry revenue. If market power exceeds the median, MP is set to 1; otherwise, it is set to 0. Based on the median of the industry Lerner index, a virtual variable for industry barriers (LN) is constructed. If the industry Lerner index is greater than the median, LN is set to 1; otherwise, it is set to 0. When a company exceeds the median in all three aspects—company size, market power, and industry barriers—it is identified as a chain master enterprise.

The test results are shown in columns (4) and (5) of Table 10. The AI coefficient for chain master enterprises is 0.0044, which is significantly positive at the 10% level; the AI coefficient for non-chain master enterprises is 0.0033, which is significantly positive at the 1% level. Compared to non-chain master enterprises, chain master enterprises have a greater impact on AI. A possible reason is that chain master enterprises hold a dominant position in the industrial chain and supply chain, enabling them to effectively coordinate and allocate upstream and downstream resources using AI. Furthermore, the extensive operations of chain master enterprises facilitate the distribution of research and development, as well as application costs associated with artificial intelligence, amplifying its technological value and thereby enhancing supply chain resilience.

6. Discussion and Conclusions

6.1. Discussion

This study, grounded in the reality that artificial intelligence (AI) is empowering China’s economic and social development, empirically examines the mechanism through which corporate AI influences supply chain resilience using data from A-share listed companies from 2012 to 2023. Findings reveal that AI significantly enhances supply chain resilience, with information transparency, innovation capability, and absorptive capacity acting as partial mediators. Digital governance positively moderates this relationship. Further analysis indicates that AI’s enabling effects are more pronounced in high-tech firms, manufacturing enterprises, growth-stage companies, mature-stage companies, and chain master enterprises.

The findings align closely with dynamic capability theory, demonstrating that AI, as a strategic digital asset, enhances supply chain resilience to mitigate disruptive shocks. This effect corroborates prior research by Wang (2022) [41] and Xu Y et al. (2025) [42], while further revealing the mechanism roles of information transparency, dynamic capabilities, and digital governance, thereby expanding theoretical boundaries.

The mediating role of information transparency highlights how AI technology enhances end-to-end supply chain visibility by integrating multi-source data and applying intelligent algorithms, creating a highly transparent information environment. This transparency enables AI systems to achieve more accurate demand forecasting, optimized holistic decision-making, and efficient cross-chain coordination, thereby systematically strengthening the supply chain’s capacity to prevent, withstand, and recover from disruptions.

Similarly, the importance of innovation and absorption capabilities demonstrates that AI, through intelligent analytical capabilities, enhances enterprises’ ability to identify and adapt to supply chain risks, enabling rapid responses to disruptions. Concurrently, AI technology stimulates organizational innovation by generating novel solutions, driving structural optimization and paradigm shifts in supply chains. This systematically transforms AI’s technological advantages into sustainable resilience benefits, consistent with Le (2024) [43].

Digital government, as a critical institutional environment factor, provides essential public data resources and computational support for enterprise AI systems through building government data open platforms and cross-border digital infrastructure. It also reduces policy uncertainty in AI applications via digital regulation and collaborative governance platforms. This broadens AI application scenarios, enabling enterprises to more effectively translate AI into supply chain resilience advantages, ultimately amplifying AI’s role in enhancing supply chain resilience.

In summary, artificial intelligence does not enhance supply chain resilience as an isolated technological solution but through complex interactions between internal mechanisms and external institutional support, with varying effectiveness across different corporate environments.

6.2. Theoretical and Practical Implications

From a theoretical perspective, this study reveals the intrinsic mechanism by which artificial intelligence enhances supply chain resilience through improved information transparency and dynamic capabilities. It establishes the role of digital government in boundary-enabling functions, providing new empirical evidence for applying dynamic capability theory and institutional theory in a digital context. Simultaneously, it offers significant insights for governments and enterprises in building sustainable supply chains.

Practically, this study provides enterprises with a clear pathway to enhance supply chain resilience by improving AI capabilities. It also offers decision support for governments to strengthen industrial and supply chain resilience through digital governance, holding significant practical value for safeguarding industrial chain security.

6.3. Policy Implications

In addition to providing a series of empirical evidence for enhancing supply chain resilience through artificial intelligence, the research conclusions offer the following insights for governments and businesses: From a government perspective, countries can strengthen top-level design and strategic planning by formulating a national strategy for building a smart supply chain powerhouse. By establishing artificial intelligence and supply chain resilience as core economic development strategies, countries can comprehensively enhance the supply chain’s ability to withstand various risks. Building on this foundation, local governments can align with the national long-term strategic vision, leverage regional economic characteristics and industrial foundations, and establish quantifiable regional and phased objectives. They can then formulate differentiated policy guidelines to promote the deep integration and application of AI across all supply chain segments. Within the framework of national strategic planning, governments should refine the policy and legal framework. They should implement digital governance to create a favorable institutional environment for AI development. Additionally, they should increase policy and financial support. The government can establish a special fiscal subsidy fund to support AI technology R&D and application demonstration projects in the supply chain sector, promote the commercialization of AI innovations, and fully leverage AI’s leading role. Tax incentives are also implemented to encourage enterprises to take the lead in AI development. Concurrently, digital infrastructure construction is coordinated. The development of 5G networks, data centers, and other digital infrastructure is accelerated to establish the hardware foundation for AI applications, thereby driving the deep integration and application of AI within supply chain systems and leveraging AI’s unique advantages in enhancing supply chain resilience.

From the corporate perspective, companies should accurately grasp the objective laws of AI development, accelerate AI technological innovation, strengthen AI application capabilities, and use intelligent applications to improve corporate management levels and drive business model innovation. Through AI technologies such as deep learning and big data analysis, companies can enhance the responsiveness and flexibility of supply chains. Additionally, companies should promote the construction of intelligent supply chain ecosystems to enhance the stability of supply and demand among upstream and downstream enterprises in the supply chain. By improving corporate transparency, real-time, secure information sharing between upstream and downstream enterprises and dynamic traceability of the supply chain can be achieved, thereby reducing potential risks caused by information asymmetry. Through cross-border supply chain collaboration and ecosystem-sharing mechanisms, the value co-creation of the supply chain ecosystem can be promoted. This enables the deep exploration of new production factors and the upgrading of traditional production factors through AI technology, achieving a full-chain upgrade and value sharing of the intelligent supply chain. This approach helps seize the opportunities of intelligent manufacturing and empower the sustainable development of the supply chain.

6.4. Limitations and Prospects

First, the indicator system of the supply chain resilience assessment framework remains incomplete and fails to adequately account for industry-specific characteristics. This may result in assessment outcomes lacking cross-industry comparability and relevance. Future research should focus on developing differentiated assessment systems tailored to specific industries to more accurately reflect actual resilience levels across sectors and provide enterprises with more actionable management insights.

Second, the causal pathways connecting artificial intelligence to supply chain resilience are intricate. Information transparency and dynamic capacities serve as merely partial mediation channels. Future study may incorporate multiple viewpoints, such as investigating managerial cognition and capacities to analyze how senior management teams affect AI technology adoption and its resilience-enhancing impacts.

Finally, this study did not sufficiently incorporate firm heterogeneity factors, such as ownership structure and regional differences, potentially limiting the generalizability of its conclusions. Therefore, subsequent research should conduct comparative and grouped studies of firms with different ownership structures and regional locations to derive more nuanced and explanatory findings. This will enable the provision of differentiated, precise strategic recommendations and policy support for various types of enterprises and government departments.