The Impact Mechanism of AI Technology on Enterprise Innovation Resilience

Abstract

Amid the rapid advancement of artificial intelligence (AI) and increasing environmental uncertainty, enterprises are facing unprecedented challenges in sustaining innovation. As a key enabler of digital transformation, AI enhances resource allocation efficiency and knowledge acquisition, offering new avenues for continuous innovation under dynamic conditions. Innovation resilience—defined as a firm’s ability to maintain and restore innovation activities during external shocks—has emerged as a critical indicator of organizational adaptability. Leveraging its advantages in data processing, process optimization, and organizational learning, AI is increasingly regarded as a pivotal driver of innovation resilience. This study develops a theoretical framework linking AI technology, dynamic capabilities, and innovation resilience. Using panel data from Chinese A-share listed companies between 2013 and 2023, we conduct an empirical analysis with a two-way fixed effects model. The results reveal that AI technology significantly enhances innovation resilience; dynamic capabilities partially mediate this relationship; and financial constraints positively moderate the effect of AI on innovation resilience. By adopting a dual perspective of technological enablement and capability construction, this research uncovers the internal mechanism through which AI fosters resilient innovation and provides practical insights for enterprises seeking capability upgrading under resource limitations.

1. Introduction

In an era of intensifying global uncertainty and rapid technological change, firms are increasingly exposed to disruptive shocks and transformative pressures [1]. Events such as the COVID-19 pandemic, geopolitical tensions (e.g., the China–U.S. technology decoupling), and structural shifts in global supply chains have fundamentally reshaped the logic of corporate innovation. These challenges demand more than efficiency—they call for resilience. This transformation echoes Schumpeter’s (2013) conception of innovation as a process of creative destruction and renewal, where firms must continuously reconfigure their innovation activities to survive and thrive [2].

This emphasis on dynamic adaptability is especially salient for firms operating in institutionally complex and resource-constrained contexts. In emerging markets such as China, where regulatory volatility, financing constraints, and technological dependence are prominent, the ability to maintain, recover, and transform innovation capabilities under adversity becomes central to long-term competitiveness. Against this backdrop, the concept of innovation resilience—defined as a firm’s capacity to sustain, restore, and adapt innovation processes in response to disruption—has gained significant traction among scholars and practitioners [3]. Unlike traditional measures of innovation success focused on output quantity or R&D input, innovation resilience captures a firm’s robustness and sustainability in volatile environments.

In parallel, artificial intelligence (AI) has emerged as a transformative general-purpose technology, enabling firms to enhance decision-making, automate processes, and respond dynamically to environmental turbulence [4]. Broadly defined, AI encompasses computational methods such as machine learning, natural language processing, and data-driven inference that simulate human intelligence [5]. The rapid proliferation of AI across industries has sparked significant academic interest in its innovation implications. Existing research has highlighted multiple pathways through which AI influences innovation. Studies show that AI enhances innovation efficiency by optimizing R&D processes, improving data analytics, and strengthening knowledge integration [5,6,7]. Others emphasize AI’s role in transforming business models and enabling collaborative innovation ecosystems, such as AI-driven interfirm networks and platform-based co-creation [8,9]. AI has also been widely examined as a key driver of digital transformation, contributing to intelligent manufacturing, service upgrading, and cross-sector innovation [10,11].

However, most prior research tends to adopt an efficiency-oriented lens, focusing on how AI improves innovation output, cost structures, or speed. There remains limited understanding of how AI adoption contributes to innovation resilience—that is, a firm’s ability to sustain, adapt, and recover its innovation processes under disruption. In complex and uncertain environments, where volatility and resource constraints prevail, this resilience perspective is increasingly vital yet undertheorized.

To advance this agenda, we draw on dynamic capability theory as the core analytical framework. Dynamic capabilities refer to a firm’s ability to integrate, build, and reconfigure internal and external competencies to address rapidly changing environments [12]. We argue that AI can enhance dynamic capabilities—such as sensing opportunities, seizing them through resource reconfiguration, and transforming knowledge structures—thus enabling firms to adapt their innovation systems in response to external disruptions.

Furthermore, we recognize that the impact of AI on innovation resilience is highly contingent upon organizational and contextual factors. In particular, AI deployment is capital-intensive and involves uncertain returns, making it especially challenging for resource-constrained firms. In the Chinese context, small- and medium-sized enterprises (SMEs) often struggle with limited financial capacity, weak digital infrastructure, and insufficient absorptive capabilities. Therefore, this study also incorporates financial constraints as a boundary condition to assess how they moderate the relationship between AI adoption and innovation resilience.

Importantly, we do not treat China as a mere empirical site but as a theoretically significant context. As a leading emerging economy undergoing a digital transition amid geopolitical and institutional complexity, China provides a critical setting to explore how firms strategically adopt AI to enhance resilience. Drawing on a panel dataset of Chinese A-share listed firms from 2013 to 2023, this study empirically examines how AI adoption contributes to innovation resilience through dynamic capabilities, while considering financial constraints as a boundary condition. This dual focus allows us to refine theoretical understandings of resilience and provide practical guidance for firms navigating uncertainty in transitional economies.

By doing so, this study contributes to the literature in three ways. First, it introduces innovation resilience as a distinct outcome of AI adoption, offering a conceptual shift from efficiency to adaptability. Second, it provides a dynamic capability-based mechanism for understanding AI’s role in enhancing resilience. Third, it highlights the constraining effects of financial resources and the importance of context-sensitive digital strategies. These insights are especially relevant for policymakers and managers seeking to promote AI-driven innovation transformation in emerging markets.

2. Theoretical Framework

2.1. AI Technology and Innovation Resilience

Innovation resilience refers to a firm’s flexibility, adaptability, and recovery capacity when facing environmental uncertainty [13]. It reflects the ability to rapidly adjust strategies, reconfigure innovation processes, and maintain ongoing technological development and organizational learning amid disruptions [14]. This capacity not only reflects a firm’s resistance to external shocks and its ability to stabilize its innovation system, but also indicates its potential to recover or even advance its innovation functions during crises.

As a general-purpose enabling technology, artificial intelligence (AI) is increasingly embedded in business operations, management, and innovation processes. With its advanced capabilities in data processing, real-time feedback, and machine learning, AI enables firms to make faster, more informed decisions in response to complex, rapidly changing conditions. The widespread adoption of AI enhances innovation resilience through several key mechanisms:

First, AI significantly improves firms’ ability to perceive and process information. Its powerful data mining and analytical capabilities allow firms to efficiently collect and interpret vast amounts of external information, helping them identify emerging opportunities and risks [15]. This strengthens their responsiveness and adaptability during crises, thereby enhancing innovation resilience [16].

Second, from a resource-based perspective, innovation resilience depends on the availability and flexibility of internal resources [17]. AI improves resource allocation efficiency and productivity through deep learning, predictive modeling, and process automation [18,19]. These gains allow firms to reallocate resources more effectively, ensuring that limited resources can be directed toward innovation recovery and system reconstruction during times of stress [20].

Third, AI transforms how firms access and manage knowledge. AI-driven digital platforms and collaborative systems reduce organizational boundaries and knowledge silos, facilitating cross-organizational innovation and knowledge flows [21]. This enables firms to absorb valuable technological, market, and managerial insights more rapidly, improving innovation responsiveness and systemic adaptability [1]. In addition, AI enhances risk identification and early warning systems, supporting more informed decision-making and reducing the likelihood of innovation disruptions caused by information delays or cognitive biases [22,23].

In summary, AI enhances firms’ information processing capabilities, optimizes resource allocation, and facilitates knowledge sharing, collectively supporting the continuity and evolution of innovation systems under external shocks. Based on this, we propose the following hypothesis:

H1: 

AI technology positively influences enterprise innovation resilience.

2.2. The Mediating Role of Dynamic Capabilities

Dynamic capabilities—comprising absorptive, innovative, and adaptive capacities—are essential for firms to sustain innovation in uncertain environments [24]. While AI offers powerful tools for information and resource processing, its impact on innovation resilience is not automatic. Rather, AI enhances innovation resilience by strengthening a firm’s dynamic capabilities, which serve as a key internal mechanism through which technology adoption is translated into systemic innovation capacity.

First, AI enhances firms’ absorptive capacity, which refers to the ability to identify, acquire, and integrate external knowledge. AI systems, through tools such as natural language processing, knowledge graphs, and big data analytics, help firms extract relevant insights from patents, academic literature, and market feedback [25]. This strengthens their awareness of technological and market trends and improves early warning capabilities. More importantly, absorptive capacity enables firms to learn from crises, internalize external knowledge, and translate it into actionable innovation strategies [26,27].

Second, AI strengthens innovation capacity, enabling firms to reconfigure systems and renew technologies more efficiently. AI is widely applied across the innovation value chain—for example, analyzing user needs through data mining, accelerating R&D with simulation algorithms, and facilitating smart manufacturing for flexible production [28]. These applications improve the speed and quality of technological development, helping firms rapidly adjust and avoid innovation system breakdowns during disruption.

Third, AI enhances firms’ adaptive capacity, allowing them to maintain strategic flexibility and organizational agility. This includes real-time monitoring of market changes, consumer preferences, and policy shifts through AI-powered forecasting systems [5]. Tools such as AI-driven decision support systems, scenario simulations, and algorithmic platforms improve cross-functional coordination and responsiveness, enabling firms to quickly recalibrate innovation strategies and maintain system stability amid volatility [29].

In sum, AI strengthens the core components of dynamic capabilities, which in turn serve as the mechanism through which AI enhances innovation resilience. Therefore, we propose the following hypothesis:

H2: 

Dynamic capabilities mediate the relationship between AI technology and innovation resilience.

2.3. The Moderating Role of Financial Constraints

Financial constraints represent a key boundary condition that influences how firms engage in innovation [30]. When resources are limited, firms must make more selective and efficient decisions regarding technology adoption and innovation investment [31]. Given that AI implementation often involves high upfront costs, technical complexity, and long-term returns, financial constraints may moderate the impact of AI on innovation resilience.

On the one hand, under high financial constraints, firms are more likely to adopt a focused, value-driven approach to AI deployment. Limited resources force them to concentrate AI applications on core R&D processes or high-potential innovation areas, rather than dispersing investments across non-essential domains [32,33]. This selective use of AI enhances its marginal value and increases the likelihood of meaningful innovation outcomes, particularly under constrained conditions.

On the other hand, low financial constraints may weaken the impact of AI on innovation resilience. Firms with abundant resources are more prone to broad, unfocused AI investments and may pursue aggressive expansion strategies with less emphasis on resilience-building [34]. In such cases, AI may be underutilized or misaligned with core innovation goals. Additionally, excess slack can reduce the urgency to build flexible and responsive innovation systems, diminishing the resilience-enhancing effects of AI.

Based on this reasoning, we propose the following hypothesis:

H3: 

Financial constraints positively moderate the relationship between AI technology and innovation resilience—that is, the effect of AI on innovation resilience is stronger when financial constraints are higher.

3. Research Methodology and Data

3.1. Model Specification

3.1.1. Baseline Regression and Mediation Effect Models

To examine the impact of AI technology on enterprise innovation resilience, this study employs a two-way fixed effects model. The model is specified as follows:

$${Resi}_{it}={\beta }_0+{\beta }_1{AI}_{it}+{\beta }_2{Controls}_{it}+{Firm}_i+{Year}_t+{\epsilon }_{it}$$

(1)

$${Dynamic}_{it}={\phi }_0+{\phi }_1{AI}_{it}+{\phi }_2{Controls}_{it}+{Firm}_i+{Year}_t+{\epsilon }_{it}$$

(2)

$${Resi}_{it}=k_0+k_1{AI}_{it}+{k_{2}Dynamic}_{it}+k_3{Controls}_{it}+{Firm}_i+{Year}_t+{\epsilon }_{it}$$

(3)

Equation (1) represents the baseline regression, while Equations (2) and (3) are used to test the mediation effect. ${Resi}_{it}$ is the dependent variable, representing enterprise innovation resilience; ${AI}_{it}$ is the independent variable, denoting AI technology adoption; ${Dynamic}_{it}$ represents the firm’s dynamic capabilities; ${Controls}_{it}$ is a vector of control variables; ${Firm}_i$ and ${Year}_t$ represent firm and year fixed effects, respectively; $\beta$, $\phi$, $k$ are the coefficients of the corresponding variables; ${\epsilon }_{it}$ is the random error term.

If β1 is significantly positive, it indicates that AI technology has a significant impact on enterprise innovation resilience. If ${\phi }_1$ is significantly positive, it suggests a significant positive relationship between AI technology and dynamic capabilities. If both ${\phi }_1$ and $k_2$ are significant, and $k_1$ is also significant, this indicates a partial mediation effect. If both ${\phi }_1$ and $k_2$ are significant, but $k_1$ is not significant, this indicates a full mediation effect.

3.1.2. Moderation Effect Model

Based on the theoretical analysis, financial constraints are considered a moderating factor in the relationship between AI technology and enterprise innovation resilience. To test this hypothesis, the following model is constructed:

$${Resi}_{it}={\theta }_0+{\theta }_1{AI}_{it}+{\theta }_2{{KZ}_{it}+{\theta }_3{AI}_{it}\times {KZ}_{it}+{\theta }_{4}Controls}_{it}+{Firm}_i+{Year}_t+{\epsilon }_{it}$$

(4)

In this model, ${KZ}_{it}$ represents the firm’s financial constraints, and ${AI}_{it}\times {KZ}_{it}$ is the interaction term between AI technology and financial constraints. The other variables are consistent with those in Equation (1). Equation (4) is used to examine the moderating effect of financial constraints.

3.1.3. Variable Measurement and Descriptions

(1)

Dependent Variable

Following the approach of Zhao et al. (2024) [35], this study constructs a comprehensive index of innovation resilience using the entropy weight method. The index incorporates the ratio of R&D expenditure to total assets, the number of R&D personnel, the number of patent applications, and the logarithm of patent applications divided by R&D expenditure and multiply the result by 100.

(2)

Independent Variable

Drawing on Liu et al. (2025) [36], AI technology is measured based on the frequency of AI-related keywords appearing in firms’ annual reports. Specifically, the number of AI-related words is counted, and the logarithm of this number plus one is used as a proxy for the level of AI technology adoption.

(3)

Mediating Variable

Following the methodology of Yang et al. [37], dynamic capabilities are measured from three dimensions: innovation capability, absorptive capability, and adaptive capability. Innovation capability is assessed by standardizing and summing two indicators: R&D intensity and the proportion of technical personnel. Absorptive capability is measured by R&D intensity, calculated as the ratio of annual R&D expenditure to operating revenue. Adaptive capability is proxied by the coefficient of variation in three major expenditures: R&D, capital, and advertising. To ensure directional consistency (i.e., a higher value indicating stronger adaptability), the coefficient of variation is multiplied by −1. The final dynamic capabilities index is constructed by averaging the three standardized dimensions.

(4)

Moderating Variable

Following Zhang et al. (2025) [38], this study uses the KZ Index to evaluate a firm’s level of financial constraints. The KZ Index is constructed using variables such as the cash ratio, cash flow ratio, dividends per share, debt ratio, and Tobin’s Q. A higher KZ Index value indicates a greater degree of financial constraint.

(5)

Control Variables

Based on prior research, the following control variables are included: firm size (Size), return on equity (ROE), asset turnover ratio (ATO), cash flow ratio (Cashflow), CEO duality (Dual), firm age (FirmAge), leverage ratio (Lev), and the shareholding ratio of the largest shareholder (Top1). Both firm and year fixed effects are included in the regression analysis.

The measurement indicators and data sources for all variables are summarized in

Table 1. Variable definitions.

Variable Type	Variable Type	Symbol	Measurement Method	Data Source
Dependent Variable	Innovation Resilience	Resi	Constructed using the entropy weight method	CSMAR Database; Wind Database; the annual reports of listed companies
Independent Variable	AI Technology	AI	Logarithm of the number of AI-related keywords in annual reports plus one
Mediating Variable	Dynamic Capabilities	Dynamic	Average of absorptive, innovative, and adaptive capabilities
Moderating Variable	Financial Constraints	KZ	KZ Index
Control Variables	Firm Size	Size	Logarithm of total assets
	Return on Equity	ROE	Net profit/Net assets
	Asset Turnover Ratio	ATO	Net sales revenue/Average total assets
	Cash Flow Ratio	Cashflow	Net operating cash flow/Current liabilities
	CEO Duality	Dual	Equals 1 if the chairman and CEO are the same person, 0 otherwise
	Firm Age	FirmAge	Current year minus the year of establishment
	Leverage Ratio	Lev	Total liabilities/Total assets
	Ownership Concentration	Top1	Shareholding ratio of the largest shareholder

.

3.2. Data Sources and Descriptive Statistics

This study uses data from A-share listed companies in China from 2013 to 2023 as the research sample. The data were processed using the following steps: (1) excluding firms designated as ST, *ST, and PT; (2) removing firms in the financial and real estate sectors; (3) eliminating samples with severe data missingness; and (4) applying a two-sided 1% winsorization to the selected sample to mitigate the influence of outliers.

Table 2. Descriptive statistics.

Variable	N	Mean	p50	SD	Min	Max
AI	29,046	1.287	1.099	1.358	0	4.956
Resi	29,046	1.023	0.6050	2.166	0	68.74
Dynamic	29,046	−0.183	−0.194	0.176	−0.536	0.335
KZ	29,046	1.054	1.353	2.468	−7.276	6.349
Size	29,046	22.20	22	1.258	19.99	26.13
ROE	29,046	0.0590	0.0730	0.136	−0.639	0.353
ATO	29,046	0.6270	0.5480	0.386	0.0940	2.393
Cashflow	29,046	0.0500	0.0490	0.0660	−0.141	0.243
Dual	29,046	0.3370	0	0.473	0	1
FirmAge	29,046	2.982	2.996	0.307	1.386	4.190
Lev	29,046	0.394	0.384	0.196	0.0550	0.871
Top1	29,046	0.328	0.304	0.144	0.0820	0.721

presents the descriptive statistics of the main variables.

4. Results

4.1. Baseline Regression

Table 3. Baseline regression results.

Variable	(1)	(2)
	Resi	Resi
AI		0.052 ***
		(0.010)
Size	0.292 ***	0.281 ***
	(0.019)	(0.019)
ROE	−0.200 ***	−0.197 ***
	(0.060)	(0.060)
ATO	0.275 ***	0.273 ***
	(0.037)	(0.037)
Cashflow	−0.120	−0.116
	(0.112)	(0.112)
Dual	0.033	0.033
	(0.020)	(0.020)
FirmAge	0.740 ***	0.731 ***
	(0.162)	(0.161)
Lev	−0.380 ***	−0.376 ***
	(0.073)	(0.073)
Top1	0.251 *	0.259 **
	(0.131)	(0.131)
_cons	−7.009 ***	−6.805 ***
	(1.074)	(1.074)
N	29,046.000	29,046.000
R-squared	0.043	0.044
id	YES	YES
year	YES	YES

Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

presents the baseline regression results, which examine the main effect of AI technology on innovation resilience. Column (1) reports the results with control variables only, while Column (2) shows the results after adding the independent variable. The coefficient of AI technology is 0.052 and is statistically significant at the 1% level, indicating that AI adoption has a positive impact on enterprise innovation resilience. Therefore, Hypothesis H1 is supported.

4.2. Endogeneity Test

To address potential endogeneity concerns, this study employs an instrumental variable (IV) approach, using the one-period lag of the independent variable as an instrument. The regression results are presented in

Table 4. Regression results using the instrumental variable approach.

Variable	(1)	(2)
	First	Second
	AI	Resi
LAI	0.306 ***
	(45.26)
AI		0.151 ***
		(3.94)
Size	0.161 ***	0.281 ***
	(11.73)	(11.18)
ROE	0.089 **	−0.228 ***
	(2.26)	(−3.35)
ATO	0.048 *	0.337 ***
	(1.88)	(7.53)
Cashflow	−0.081	−0.135
	(−1.08)	(−1.04)
Dual	−0.002	0.038
	(−0.18)	(1.62)
FirmAge	−0.000	0.481 **
	(−0.00)	(2.16)
Lev	−0.070	−0.375 ***
	(−1.37)	(−4.23)
Top1	0.031	0.131
	(0.34)	(0.83)
Under-identification Test	2050.82
LM Statistic	[0.000]
Weak Instrument Test	2048.88
Wald Statistic	{16.38}
First-stage F Statistic	244.09
Observations	23,080	23,080
R-squared	0.188	0.032
Number of id	4089	4089

t-statistics in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1.

.

Column (1) reports the first-stage regression, where the coefficient of the instrumental variable is significantly positive, confirming a strong correlation between the instrument and the endogenous variable.

Column (2) presents the second-stage regression results, showing that the coefficient of the endogenous variable remains significantly positive. After accounting for endogeneity, the positive effect of AI technology on innovation resilience persists, providing further support for Hypothesis H1.

4.3. Robustness Tests

To ensure the robustness of the empirical results, this study conducts a series of robustness checks by (1) replacing the measurement of the core independent variable, (2) replacing the measurement of the dependent variable, and (3) introducing high-dimensional fixed effects. The regression results are presented in

Table 5. Robustness checks.

Variable	(1)	(2)	(3)
	Resi	LResi	Resi
MD&A	0.004 ***
	(0.001)
AI		0.044 ***	0.052 ***
		(0.012)	(0.010)
Size	0.288 ***	0.271 ***	0.281 ***
	(0.019)	(0.023)	(0.020)
ROE	−0.190 ***	0.052	−0.215 ***
	(0.060)	(0.072)	(0.060)
ATO	0.281 ***	0.250 ***	0.292 ***
	(0.037)	(0.043)	(0.037)
Cashflow	−0.110	0.137	−0.107
	(0.112)	(0.130)	(0.113)
Dual	0.034 *	0.021	0.032
	(0.020)	(0.024)	(0.020)
FirmAge	0.701 ***	0.436 **	0.771 ***
	(0.162)	(0.198)	(0.163)
Lev	−0.375 ***	−0.296 ***	−0.374 ***
	(0.073)	(0.087)	(0.074)
Top1	0.280 **	−0.029	0.238 *
	(0.131)	(0.157)	(0.133)
_cons	−6.860 ***	−5.291 ***	−6.760 ***
	(1.074)	(1.153)	(1.358)
N	29,042.000	23,561.000	29,046.000
R-squared	0.045	0.035	0.047
id	YES	YES	YES
year	YES	YES	YES
industry	NO	NO	YES

Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

. Column (1) replaces the measurement of the independent variable. Following the approach of Song et al. (2025) [39], we construct an alternative proxy for AI technology based on the content of the Management Discussion and Analysis (MD&A) section in annual reports. The results remain significant, indicating that AI technology continues to exert a positive effect on innovation resilience, confirming the robustness of the findings. Column (2) replaces the measurement of the dependent variable by using the lagged value of innovation resilience. The coefficient of the core explanatory variable remains significantly positive and is consistent with the baseline regression results. Column (3) presents the results after controlling for firm, year, and industry fixed effects using a high-dimensional fixed effects model. The results remain stable and consistent with the baseline regression, further confirming the robustness of the conclusions.

4.4. Moderating Effect Test

To further examine the moderating role of financial constraints in the relationship between artificial intelligence (AI) adoption and enterprise innovation resilience, this study introduces an interaction term “AI × KZ” into the baseline model to construct a moderation model. The aim is to assess how financial constraints influence the strength of this relationship. The regression results are presented in

Table 6. Results of the moderating effect test.

Variable	(1)
	Resi
AI	0.036 ***
	(0.011)
KZ	0.032 ***
	(0.006)
AI × KZ	0.009 ***
	(0.002)
Size	0.306 ***
	(0.019)
ROE	−0.137 **
	(0.060)
ATO	0.310 ***
	(0.037)
Cashflow	0.504 ***
	(0.136)
Dual	0.036 *
	(0.020)
FirmAge	0.523 ***
	(0.163)
Lev	−0.658 ***
	(0.081)
Top1	0.280 **
	(0.130)
_cons	−6.787 ***
	(1.072)
N	29,046.000
R-squared	0.048
id	YES
year	YES

Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

.

The results show that the coefficient of the interaction term between AI adoption and financial constraints is significantly positive at the 1% level, indicating that financial constraints positively moderate the effect of AI adoption on innovation resilience. In other words, the higher the level of financial constraint, the stronger the positive effect of AI adoption on a firm’s innovation resilience.

This finding suggests that under resource-constrained conditions, firms tend to allocate limited financial resources to high-potential, high-impact AI technologies in order to improve their adaptability and innovation continuity in uncertain environments. Consequently, such firms exhibit higher levels of innovation resilience. Hypothesis H3 is therefore supported.

4.5. Mediation Effect Test

To examine the mediating role of dynamic capabilities in the relationship between AI technology adoption and enterprise innovation resilience, this study adopts the three-step regression approach. Consistent with the prior theoretical framework, dynamic capabilities are regarded as the core organizational ability to integrate, reconfigure, and adapt resources in response to external environmental changes. The regression results are reported in

Table 7. Results of the mediation effect test.

Variable	(1)	(2)	(3)
	Resi	Dynamic	Resi
AI	0.052 ***	0.005 ***	0.049 ***
	(0.010)	(0.001)	(0.010)
Dynamic			0.690 ***
			(0.073)
Size	0.281 ***	−0.001	0.281 ***
	(0.019)	(0.002)	(0.019)
ROE	−0.197 ***	−0.067 ***	−0.151 **
	(0.060)	(0.005)	(0.060)
ATO	0.273 ***	−0.018 ***	0.285 ***
	(0.037)	(0.003)	(0.037)
Cashflow	−0.116	−0.036 ***	−0.092
	(0.112)	(0.010)	(0.112)
Dual	0.033	−0.000	0.033
	(0.020)	(0.002)	(0.020)
FirmAge	0.731 ***	0.008	0.725 ***
	(0.161)	(0.014)	(0.161)
Lev	−0.376 ***	−0.084 ***	−0.318 ***
	(0.073)	(0.006)	(0.073)
Top1	0.259 **	0.027 **	0.240 *
	(0.131)	(0.012)	(0.130)
_cons	−6.805 ***	−0.265 ***	−6.622 ***
	(1.074)	(0.095)	(1.072)
N	29,046.000	29,046.000	29,046.000
R-squared	0.044	0.037	0.048
id	YES	YES	YES
year	YES	YES	YES

Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

.

As shown in Column (2), AI technology has a significantly positive effect on dynamic capabilities. In Column (3), both AI technology and dynamic capabilities exhibit significantly positive coefficients in predicting innovation resilience. These results indicate a significant mediation pathway, where AI adoption enhances innovation resilience by improving firms’ dynamic capabilities, thereby supporting Hypothesis H2.

From a theoretical perspective, this finding is aligned with the dynamic capability theory proposed by Teece et al. (1997) [24]. The theory emphasizes that in highly uncertain and rapidly changing environments, firms must develop capabilities to sense opportunities, reconfigure resources, and respond quickly—foundations that are critical for sustaining innovation. As a general-purpose technology, AI not only improves operational efficiency but also empowers decision-making through big data analytics, algorithmic modeling, and intelligent forecasting. AI adoption facilitates the acquisition and processing of external information, the integration and recombination of internal knowledge, and the dynamic adjustment of strategic responses, thereby comprehensively strengthening a firm’s dynamic capabilities.

4.6. Heterogeneity Analysis

This study further conducts subgroup regression analyses based on three typical firm heterogeneity characteristics: ownership structure, firm size, and industry competition intensity, to examine whether the effect of AI technology on innovation resilience varies under different conditions.

4.6.1. Ownership Heterogeneity

Ownership structure fundamentally shapes a firm’s property rights, governance mechanisms, and strategic orientation, thereby influencing how emerging technologies such as AI are adopted and embedded within organizational systems. As a complex and systemic digital technology, AI requires deep integration into firms’ operational routines and managerial processes to fully realize its potential in enhancing innovation resilience.

Compared to state-owned enterprises (SOEs), non-state-owned enterprises (NSOEs) operate under stricter market discipline and face greater environmental uncertainty. Lacking access to implicit policy protection or financial safety nets, NSOEs must rely more heavily on the substantive transformation of technological capabilities to sustain innovation and adaptability. Consequently, they are more likely to treat AI not merely as a symbolic adoption, but as a strategic tool for building resilience—leveraging dynamic capabilities such as absorptive capacity, organizational learning, and adaptive reconfiguration to mitigate external shocks.

Empirical evidence presented in

Table 8. Regression results for ownership heterogeneity.

Variable	(1)	(2)
	Resi (Non-State-Owned Enterprise)	Resi (State-Owned Enterprise)
AI	0.056 ***	0.043 *
	(0.011)	(0.024)
Size	0.263 ***	0.384 ***
	(0.020)	(0.045)
ROE	−0.289 ***	−0.070
	(0.064)	(0.132)
ATO	0.282 ***	0.220 ***
	(0.039)	(0.081)
Cashflow	−0.062	−0.258
	(0.118)	(0.259)
Dual	0.033	0.042
	(0.021)	(0.051)
FirmAge	0.772 ***	1.421 ***
	(0.170)	(0.390)
Lev	−0.205 ***	−0.730 ***
	(0.076)	(0.176)
Top1	0.486 ***	−0.479 *
	(0.153)	(0.252)
_cons	−6.671 ***	−11.565 ***
	(0.995)	(1.454)
N	20,827.000	8219.000
R-squared	0.039	0.065
id	YES	YES
year	YES	YES

Standard errors in parentheses. * p < 0.1, *** p < 0.01.

supports this view: the coefficient of AI adoption on innovation resilience is significantly larger for NSOEs than for SOEs. This indicates that market-driven firms—with greater operational flexibility, stronger performance-based incentives, and fewer bureaucratic rigidities—are better positioned to convert AI investments into tangible improvements in dynamic capabilities and systemic innovation response. In contrast, SOEs may suffer from organizational inertia, misaligned incentive structures, and overreliance on government support, all of which weaken the effective internalization of AI as a resilience-enhancing mechanism.

4.6.2. Firm Size Heterogeneity

Firm size plays a crucial role in shaping the conditions under which AI adoption translates into enhanced innovation resilience. Large enterprises generally possess superior resource endowments, well-established organizational systems, and advanced digital infrastructure. These advantages allow them to build comprehensive AI application platforms, sustain long-term technological investment, and develop complementary capabilities such as knowledge management and strategic coordination.

In contrast, small firms often operate under tighter financial constraints, face higher technological adoption thresholds, and lack mature managerial structures. These limitations constrain their ability to absorb digital technologies and implement organization-wide transformations. Given that the capability enhancement enabled by AI is inherently a systemic and infrastructure-intensive process, the ability to link AI adoption with dynamic capabilities—such as sensing, learning, and reconfiguration—is more likely to be realized in larger firms.

Specifically, large enterprises are better positioned to embed AI into core innovation routines through mechanisms such as process reengineering, digital platform construction, and intelligent system integration. These efforts collectively enhance firms’ dynamic capabilities and enable a more structured and resilient innovation response to external shocks. In contrast, while small firms may adopt AI tools, they frequently lack the internal absorptive and adaptive capacity necessary to fully leverage them, resulting in limited or unsustainable impact on innovation resilience.

As reported in

Table 9. Regression results for firm size heterogeneity.

Variable	(1)	(2)
	Resi (Small Size)	Resi (Large Size)
AI	0.024 ***	0.098 ***
	(0.004)	(0.020)
Size	0.065 ***	0.618 ***
	(0.010)	(0.047)
ROE	−0.221 ***	−0.287 **
	(0.023)	(0.121)
ATO	0.232 ***	0.418 ***
	(0.016)	(0.076)
Cashflow	−0.089 **	−0.264
	(0.041)	(0.236)
Dual	−0.001	0.067
	(0.008)	(0.043)
FirmAge	0.475 ***	1.467 ***
	(0.069)	(0.334)
Lev	−0.038	−0.693 ***
	(0.029)	(0.170)
Top1	0.410 ***	0.004
	(0.060)	(0.259)
_cons	−2.463 ***	−17.345 ***
	(0.282)	(1.361)
N	14,523.000	14,523.000
R-squared	0.091	0.060
id	YES	YES
year	YES	YES

Standard errors in parentheses. ** p < 0.05, *** p < 0.01.

, the effect of AI adoption on innovation resilience is statistically significant and stronger in large enterprises, while it is weaker or non-significant in smaller firms. This finding suggests that the effectiveness of AI is contingent upon organizational scale and maturity, reinforcing the view that the integration of AI and dynamic capabilities requires a supportive structural and resource environment.

4.6.3. Industry Competition Heterogeneity

Industry competition intensity plays a critical role in shaping firms’ innovation strategies, resource allocation priorities, and organizational responsiveness. In highly competitive industries, firms are exposed to greater market volatility, rapid technological substitution, and intensified performance pressure. To survive and sustain competitive advantage, they are compelled to enhance internal capabilities, emphasize rapid decision-making, and build flexible innovation systems. Under such conditions, the adoption of AI is more likely to be integrated into core business processes to support dynamic sensing, learning, and recovery—thus reinforcing innovation resilience.

In contrast, firms operating in low-competition industries—often protected by monopolistic positions or stable market structures—face weaker innovation incentives and limited external threats. These firms may adopt AI technologies for symbolic or compliance-oriented purposes, rather than as a strategic mechanism for capability transformation. As a result, the actual integration of AI into their innovation and resilience systems tends to be shallow and fragmented.

The effectiveness of AI adoption in fostering innovation resilience thus depends significantly on the intensity of competitive pressure, which acts as a catalyst for firms’ dynamic adjustment and organizational learning behaviors. This study adopts the Herfindahl–Hirschman Index (HHI) to measure industry competition intensity, where lower HHI values represent higher levels of competition. As shown in

Table 10. Regression results for industry competition heterogeneity.

Variable	(1)	(2)
	Resi (High-Competitive Industries)	Resi (Low-Competition Industries)
AI	0.088 ***	0.027 *
	(0.013)	(0.016)
Size	0.279 ***	0.243 ***
	(0.025)	(0.031)
ROE	−0.489 ***	−0.012
	(0.079)	(0.090)
ATO	0.485 ***	0.128 **
	(0.052)	(0.054)
Cashflow	−0.367 ***	−0.009
	(0.141)	(0.174)
Dual	0.036	0.011
	(0.025)	(0.033)
FirmAge	0.899 ***	0.602 **
	(0.206)	(0.268)
Lev	−0.295 ***	−0.519 ***
	(0.091)	(0.119)
Top1	0.109	0.369 *
	(0.178)	(0.203)
_cons	−7.190 ***	−6.436 ***
	(1.028)	(0.980)
N	14,540.000	14,506.000
R-squared	0.061	0.038
id	YES	YES
year	YES	YES

Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

, the positive effect of AI adoption on innovation resilience is significantly stronger in highly competitive industries compared to less competitive ones. This finding highlights that in environments characterized by strong external pressure, firms are more motivated to transform AI into a core driver of innovation resilience through the development of agile, adaptive, and responsive capabilities.

5. Conclusions, Policy Recommendations, and Research Limitations

5.1. Conclusions

Grounded in dynamic capability theory, this study investigates how the adoption of AI enhances firms’ innovation resilience. The main findings are summarized as follows:

First, AI adoption significantly strengthens enterprise innovation resilience. By improving resource allocation, information processing, and organizational responsiveness, AI enables firms to sustain and recover innovation activities in the face of external shocks. This effect is particularly salient in the Chinese context, where traditional manufacturing firms often operate under volatile conditions, face tighter financial constraints, and undergo rapid institutional transitions. As such, the application of AI in promoting innovation resilience is not only urgent but also structurally complex, requiring coordinated digital integration and capability upgrading. This study builds on and extends prior research—for example, Hu et al. (2024) showed that AI enhances the resilience of regional innovation ecosystems by boosting evolutionary potential and systemic fluidity [40]. Our firm-level findings reinforce and refine this insight, demonstrating AI’s efficacy in improving innovation resilience under uncertainty.

Second, dynamic capabilities mediate the relationship between AI adoption and innovation resilience. Rather than serving merely as a technical enhancer, AI functions as an enabler of internal capability transformation, strengthening sensing, learning, and adaptive capacity. This enriches dynamic capability theory by revealing how AI drives organizational renewal in digitally evolving environments. The findings align with Carayannis et al. (2025), who emphasize AI’s strategic value in enhancing firms’ adaptability and resilience in turbulent settings [41]. They also echo Al Dhaheri et al. (2024), who identify AI and dynamic capabilities as parallel drivers of firm performance under complexity [6]. However, this study advances the literature by conceptualizing AI not just as a parallel enabler but as a catalyst—one that triggers and amplifies dynamic capabilities, ultimately fostering long-term competitive advantage through improved innovation resilience.

Third, financial constraints positively moderate the effect of AI on innovation resilience. When firms face higher financial constraints, they tend to adopt AI more cautiously and strategically, leading to greater marginal benefits and a tighter coupling between technology deployment and capability enhancement. This finding underscores a critical boundary condition: under resource-scarce conditions, firms may achieve more efficient and impactful AI utilization. It further highlights the necessity of aligning digital investment decisions with organizational learning and strategic intent.

Fourth, the impact of AI adoption on innovation resilience exhibits significant heterogeneity across firm types. The positive effect is stronger among non-state-owned enterprises (NSOEs), large-scale firms, and those in highly competitive industries. These results suggest that organizational flexibility, resource endowments, and external market pressure are key contextual enablers of effective AI implementation. Without supportive absorptive structures and adaptive mechanisms, the potential of AI to generate innovation resilience may remain underutilized.

In sum, this study constructs a comprehensive theoretical framework and provides empirical evidence on how AI adoption contributes to innovation resilience through dynamic capability development. It offers novel insights into the organizational mechanisms that translate digital technology adoption into sustainable innovation outcomes. By situating the analysis in the context of Chinese listed firms, the study also illuminates the roles of ownership structure, resource constraints, and institutional environment in shaping digital transformation outcomes. While the findings are rooted in the Chinese context, they hold broader implications for emerging markets navigating similar digital transitions. Nevertheless, future research is needed to further test the generalizability of these conclusions across diverse institutional and industrial settings.

5.2. Policy and Managerial Recommendations

Grounded in dynamic capability theory and supported by empirical analysis of Chinese A-share listed firms (2013–2023), this study identifies the mediating effect of dynamic capabilities and the moderating role of financial constraints in the relationship between AI adoption and innovation resilience. Based on these findings, the following context-sensitive recommendations are proposed to guide government strategies and enterprise-level transformation, particularly in emerging economies facing institutional and resource constraints:

(a)

Strengthen digital infrastructure and innovation ecosystems. Empirical findings suggest that AI enhances innovation resilience by improving firms’ information processing and adaptive capacity. To enable such benefits at scale, governments—especially in developing or transitioning economies—should treat AI as a national strategic priority by investing in foundational digital infrastructure, including 5G networks, cloud computing, and open-access data platforms. Public–private partnerships can facilitate shared access to AI development and analytics tools, lowering barriers for SMEs. At the firm level, upgrading ICT systems, embedding IoT and cloud platforms, and digitizing core operations are necessary to support data-driven decision-making and organizational agility. These steps directly correspond to the mechanisms through which AI boosts dynamic capabilities and innovation recovery under external shocks.

(b)

Adapt internal governance structures to facilitate AI integration. Our analysis highlights the importance of internal capability restructuring in translating AI adoption into resilient innovation. Firms should align governance with AI-driven transformation by forming cross-functional AI steering committees, flattening decision hierarchies, and embedding innovation-related KPIs and real-time feedback systems. Designating roles such as Chief Data Officer and implementing strong data governance frameworks can mitigate AI-related risks—such as bias, privacy violations, and accountability gaps. Additionally, open innovation strategies and external collaboration expand absorptive capacity and accelerate the internalization of AI’s strategic value.

(c)

Design differentiated policy support based on ownership, firm size, and industry characteristics. Heterogeneity tests in this study reveal that AI has stronger resilience-enhancing effects in non-state-owned enterprises, larger firms, and competitive industries. Policymakers should respond with differentiated strategies. SOEs and SMEs often face capability and funding gaps, requiring tailored subsidies, workforce training, and technical support. In less competitive sectors, flexible regulatory environments and fiscal incentives may promote adoption, while innovation clusters and ecosystem-based policies can amplify AI’s value in dynamic industries. Enterprises should benchmark against digital leaders and actively participate in cross-firm knowledge sharing to reduce structural innovation disparities.

(d)

Build dynamic capabilities to sustain AI-driven resilience. Since dynamic capabilities serve as the core mediating mechanism, investment in organizational learning is critical. Governments should support capacity-building initiatives such as industry–university partnerships, regional innovation hubs, and AI-focused professional education. Firms, in turn, need to strengthen absorptive, adaptive, and innovative capacities by engaging in continuous training, iterative experimentation with AI technologies, and internal knowledge exchange. These efforts ensure AI is embedded not only in operations but also in the firm’s evolution logic, helping transform technical investment into systemic innovation resilience.

(e)

Address financial constraints and enable inclusive AI investment. This study finds that financial constraints intensify the marginal effect of AI, underscoring the need for targeted financial instruments. Governments can deploy AI-specific development loans, tax incentives, innovation bonds, and risk-sharing mechanisms (e.g., loan guarantees or insurance subsidies) to lower the adoption threshold. Firms facing budget limitations should focus on cost-effective AI use cases—such as predictive maintenance, automated quality control, and customer analytics—that offer high returns with low deployment complexity. Participating in digital platforms or consortia can reduce fixed costs and enhance access to technical expertise. Modular, scalable AI architectures further supports phased deployment aligned with cash flow. Crucially, linking AI investments to quantifiable resilience outcomes—such as recovery speed, decision agility, or innovation continuity—can help secure internal approval and external financing, reinforcing the adaptive transformation process outlined in this study.

5.3. Limitations and Future Research Directions

Despite its contributions, this study has several limitations that also present promising directions for future research.

First, the measurement of AI adoption, though aligned with current empirical practice, remains subject to limitations. In this study, the primary measure relies on the frequency of AI-related keywords in firms’ annual reports, which captures the intensity of AI discourse at the firm level. To ensure robustness, an alternative measure based on the content of the Management Discussion and Analysis (MD&A) section was employed. While both approaches are useful for large-sample panel data analysis, they do not distinguish between different types of AI technologies or their strategic roles within the firm. Given that AI encompasses a wide range of applications—from routine automation to strategic decision-making support—future research should seek to refine measurement by categorizing AI use cases and linking them to specific functional domains. This would enable a more nuanced understanding of how different types of AI adoption affect innovation resilience through distinct organizational mechanisms.

Second, although this study employs panel data spanning from 2013 to 2023 and incorporates a one-period lag of AI adoption to mitigate endogeneity concerns, the empirical focus remains on relatively short-term effects. However, the impact of AI adoption is inherently dynamic and may manifest more fully over extended periods. Future research should build on this by exploring longer-term trajectories of AI capability development and its sustained influence on innovation systems. Employing advanced panel techniques—such as dynamic panel models, distributed lag models, or time-varying coefficient models—could help capture the cumulative and evolving nature of AI’s impact on firms’ innovation resilience.

Third, although this study explores financial constraints as a contextual moderator, other important organizational and environmental contingencies remain underexplored. For instance, internal factors such as corporate culture, digital leadership, and top management team (TMT) characteristics may shape how AI is interpreted, embedded, and operationalized. Externally, factors like regulatory uncertainty, industry digitalization maturity, or cross-national institutional environments could significantly condition the effectiveness of AI in fostering innovation resilience. Expanding the scope of contextual variables would enrich both the explanatory power and generalizability of the findings.

Fourth, the empirical evidence is drawn from Chinese A-share listed firms, which operate within a unique institutional, economic, and regulatory setting. While this offers valuable insights into how AI enhances innovation under conditions of resource constraint and state-market hybridity, it also limits the external validity of the results. Future research should replicate and extend the analysis in different national contexts, industry settings, and firm types to validate the robustness of the AI-resilience link and improve cross-contextual theoretical transferability.

By addressing these limitations, future research can contribute to building a more comprehensive, context-sensitive, and temporally dynamic understanding of how AI adoption shapes organizational resilience in the digital era.