Artificial Intelligence, ESG Governance, and Green Innovation Efficiency in Emerging Economies

Abstract

Emerging economies confront the dual challenge of accelerating digital transformation while simultaneously mitigating environmental degradation under conditions of institutional and governance heterogeneity. In this context, this study examines how artificial intelligence (AI) capability influences green innovation efficiency (GIE) in emerging Asian economies and investigates whether environmental, social, and governance (ESG) performance conditions this relationship. Using an unbalanced panel of 59,112 firm-year observations from 4926 publicly listed firms across 15 emerging Asian economies over the period 2011–2022, we employ a comprehensive panel-data econometric framework that accounts for unobserved heterogeneity, dynamic effects, endogeneity, and potential self-selection bias. The empirical results indicate that AI capability is positively and significantly associated with higher green innovation efficiency. More importantly, ESG performance strengthens this relationship, suggesting that robust governance frameworks enhance firms’ ability to translate digital intelligence into environmentally efficient innovation outcomes. These findings underscore that AI adoption alone is insufficient to generate sustainable value; rather, its environmental effectiveness depends critically on complementary governance structures that promote transparency, accountability, and responsible risk management. The results remain robust after correcting for endogeneity concerns, alternative model specifications, and extensive sensitivity and heterogeneity analyses. Overall, this study contributes to the literature on digital transformation and sustainability by providing large-scale, multi-country evidence that highlights the pivotal role of ESG in shaping the sustainability returns to AI adoption in emerging economies.

1. Introduction

Sustainability and long-term development have emerged as central challenges for emerging economies in the twenty-first century. Rapid industrialization, urban expansion, and demographic growth have accelerated economic transformation across Asia and the broader Global South, yet these processes have simultaneously intensified environmental degradation, energy insecurity, and institutional pressure (Al-Okaily, 2025a). Unlike advanced economies, emerging economies face a pronounced dual constraint: the need to sustain high economic growth trajectories while mitigating climate vulnerability, resource depletion, and governance fragility (Arfanuzzaman, 2021; Chong et al., 2022; J. Xu et al., 2024). In this context, GIE—defined as the ability of firms to transform technological, human, and financial inputs into environmentally beneficial outputs with minimal ecological cost—has become a critical mechanism through which firms contribute to sustainable development and the achievement of the Sustainable Development Goals (SDGs), particularly SDG 8 (Decent Work and Economic Growth), SDG 9 (Industry, Innovation and Infrastructure), and SDG 13 (Climate Action) (Dong et al., 2025; Liu et al., 2025).

Against this backdrop, AI has emerged as a potentially transformative digital technology for sustainability transitions (Al-Okaily & Al-Okaily, 2025). Prior research documents that AI-enabled applications can enhance emissions monitoring, optimize energy consumption, improve supply chain transparency, and support eco-design and predictive environmental management (Feng et al., 2024; He et al., 2025; Wang et al., 2024). Empirical evidence further suggests that AI adoption may reduce carbon emissions (Chen & Jin, 2023), improve resource efficiency (Liang et al., 2024), and stimulate green innovation output and quality (Song et al., 2025; Dong et al., 2025). These findings have fueled growing optimism regarding AI’s capacity to function as a key driver of environmentally efficient innovation.

However, the sustainability effects of AI are neither automatic nor uniformly positive. A growing strand of literature cautions that AI-driven innovation benefits are often disproportionately captured by large, resource-rich, and digitally mature firms, potentially widening inequality across firms and industries (Wang et al., 2024; Mijit et al., 2025). Other studies identify nonlinear and threshold effects, suggesting that excessive or poorly aligned AI deployment may crowd out exploratory innovation and weaken long-term GIE (J. Lin et al., 2024). Moreover, AI technologies may generate negative environmental externalities through energy-intensive computing, data-center expansion, and rebound effects, partially offsetting environmental gains—particularly in institutional settings where regulatory oversight and governance mechanisms are weak (He et al., 2025; C. Xu & Lin, 2025). Beyond environmental trade-offs, scholars increasingly highlight governance and ethical challenges associated with AI adoption, including algorithmic opacity, accountability gaps, and misalignment between technological deployment and sustainability objectives. These risks are especially salient in emerging economies characterized by fragmented regulatory frameworks, uneven enforcement capacity, and heterogeneous ESG standards (Xi & Shao, 2025; Zhao & Wang, 2025).

Despite the rapid expansion of research on AI and sustainability, two critical gaps remain unresolved. First, the existing literature is overwhelmingly dominated by single-country analyses, with a strong concentration on China. While these studies offer valuable context-specific insights, their findings are shaped by country-specific institutional arrangements, regulatory regimes, and digital infrastructure, thereby limiting their generalizability across heterogeneous emerging economies (Al-Okaily, 2025b; Yin et al., 2023; C. Xu & Lin, 2025). Second, although ESG is increasingly recognized as a mechanism for disciplining corporate behavior and aligning innovation with long-term sustainability objectives, there is limited multi-country firm-level evidence on whether and how ESG performance conditions the effectiveness of AI in enhancing GIE—particularly in emerging-economy contexts where external regulation is often weaker and more unevenly enforced.

Emerging Asia provides a uniquely relevant empirical setting to address these gaps. The region accounts for a substantial share of global manufacturing activity, energy demand growth, and future emissions, while simultaneously exhibiting pronounced heterogeneity in governance quality, digital maturity, institutional capacity, and environmental regulation (IEA, 2023; Arfanuzzaman, 2021). Importantly, emerging Asian economies differ markedly not only from developed economies, but also from one another, ranging from digitally advanced and institutionally stronger countries to economies characterized by regulatory fragmentation and limited enforcement capacity. Prior studies emphasize that insights derived from developed economies—or from China alone—cannot be readily generalized to this diverse regional landscape (Chong et al., 2022; J. Xu et al., 2024). Consequently, understanding how AI, governance, and green innovation interact across emerging Asian economies is essential for assessing whether digital transformation supports sustainable development or instead reinforces existing structural vulnerabilities.

Against this background, this study provides the first large-scale, multi-country firm-level analysis that jointly examines AI capability, ESG, and GIE in emerging economies. The analysis is based on an unbalanced panel of 4926 publicly listed non-financial firms from 15 emerging Asian economies—China, India, Indonesia, Malaysia, Thailand, Vietnam, the Philippines, Bangladesh, Pakistan, Sri Lanka, Saudi Arabia, the United Arab Emirates, Qatar, Kazakhstan, and Uzbekistan—over the period 2011–2022. These countries are selected based on the consistent availability of firm-level AI disclosure, ESG performance indicators, and eco-innovation data, while capturing substantial variation in institutional quality, regulatory environments, and digital readiness.

Drawing on the NRBV and DCT, we conceptualize AI capability as a strategic digital resource whose sustainability impact depends critically on governance conditions. We argue that ESG operates as a conditioning mechanism that determines whether AI investments translate into environmentally efficient innovation or amplify risks such as inequality, short-termism, and environmental rebound effects. Using a comprehensive econometric strategy—including firm and year fixed effects, two-step system GMM, Heckman selection correction, and propensity score matching—we examine both the direct effect of AI capability on GIE and the moderating role of ESG performance.

Our findings show that AI capability significantly enhances GIE, and that this effect is substantially stronger among firms with higher ESG performance. These results demonstrate that while digital technologies are important enablers of sustainable growth, their developmental impact depends fundamentally on governance quality and institutional alignment. By integrating AI, ESG, and GIE within a multi-country emerging-economy framework, this study advances the literature on digital sustainability and provides policy-relevant insights for firms and regulators seeking inclusive, resilient, and low-carbon development pathways.

2. Literature Review and Theoretical Framework

2.1. Literature Review: AI and GIE

Digital technologies have become central to firms’ competitiveness and sustainability strategies, with AI emerging as one of the most influential enablers of data-driven decision-making, intelligent automation, and predictive environmental management across manufacturing, finance, and energy systems (Feng et al., 2024; He et al., 2025). A growing empirical literature suggests that AI adoption can improve firms’ innovation processes by accelerating research activities, optimizing resource use, and supporting the development of environmentally friendly products and processes. These mechanisms are commonly linked to improvements in GIE, defined as the ability of firms to transform innovation inputs into environmentally beneficial outputs (Liu et al., 2025).

Empirical evidence generally supports a positive association between AI and green innovation outcomes. Hussain et al. (2024) and Song et al. (2025) document that AI adoption accelerates eco-innovation, while Mijit et al. (2025) show that policy-supported digitalization contributes to reductions in carbon intensity and improvements in green productivity. Chen and Jin (2023) further find that AI reduces carbon emissions by stimulating higher-quality green innovation, and Liang et al. (2024) report that industrial AI and robotics enhance environmental efficiency by optimizing production processes and reducing operational waste.

At the same time, an expanding strand of the literature cautions that AI adoption may generate uneven and potentially adverse outcomes. Several studies highlight that the gains from AI-driven innovation are disproportionately captured by large, resource-rich, and digitally mature firms, potentially widening inequality across firms and industries (Dong et al., 2025; Mijit et al., 2025). Other research identifies nonlinear and threshold effects, suggesting that excessive or poorly aligned AI deployment may crowd out exploratory innovation and weaken long-term GIE (Wang et al., 2024; J. Lin et al., 2024). Moreover, AI technologies can generate negative environmental externalities, including energy-intensive computing, data-center expansion, and rebound effects, which may partially offset environmental benefits when governance mechanisms are weak (He et al., 2025; C. Xu & Lin, 2025).

Beyond environmental trade-offs, scholars increasingly emphasize governance and ethical challenges associated with AI adoption, including algorithmic opacity, accountability gaps, and misalignment between technological deployment and sustainability objectives. These risks are particularly salient in emerging economies, where regulatory frameworks are often fragmented and institutional capacity uneven (Xi & Shao, 2025; Zhao & Wang, 2025). Consistent with this view, Yin et al. (2023) show that AI’s sustainability effects depend heavily on institutional quality and complementary governance investments. Broader Asian evidence similarly indicates that weak sustainability disclosure regimes and uneven digital capabilities can substantially alter the environmental impact of AI deployment (Chong et al., 2022; J. Xu et al., 2024).

Taken together, the literature presents mixed and context-dependent evidence on the sustainability implications of AI adoption. AI is therefore not inherently sustainability-enhancing; rather, its impact on GIE depends on organizational and institutional conditions. This ambiguity underscores the need for a theory-driven framework that explains when and under what conditions AI contributes to environmentally efficient innovation.

2.2. Theoretical Framework and Hypothesis Development

To explain the heterogeneous effects documented in the literature, this study draws on the NRBV and DCT. The NRBV argues that firms achieve sustained competitive advantage by developing resources and capabilities that simultaneously generate economic and environmental value (Barney & Arikan, 2005). From this perspective, AI can be conceptualized as an intangible strategic resource, enabling firms to identify ecological inefficiencies and optimize resource use through advanced analytics and algorithmic decision-making.

Complementing this view, DCT emphasizes firms’ abilities to sense, seize, and reconfigure opportunities in response to environmental and technological change (J. Lin et al., 2024). AI is expected to strengthen these dynamic capabilities by improving environmental sensing, opportunity identification, and resource reconfiguration. However, both theories imply that the value of AI is contingent rather than automatic, depending on governance quality and organizational alignment.

2.2.1. AI Capability and GIE

Integrating insights from the NRBV and DCT, AI capability is theoretically expected to enhance firms’ ability to deploy innovation inputs more efficiently toward environmentally beneficial outcomes. By improving information processing, coordination, and predictive capacity, AI may reduce waste and improve the environmental efficiency of innovation processes. Nevertheless, given the documented risks and uneven outcomes associated with AI adoption, these benefits represent theoretical expectations rather than guaranteed effects, particularly in emerging-economy contexts.

Accordingly, we propose the following hypothesis:

H1. 

AI capability is positively associated with firms’ GIE.

2.2.2. ESG as a Moderating Capability

While AI provides technological potential, its translation into sustainable environmental outcomes is expected to depend critically on governance quality. ESG performance reflects firms’ integrity, transparency, and accountability in managing environmental and social issues (Farmanesh et al., 2025; Saleh et al., 2025; Al-Tahat et al., 2025). From a theoretical standpoint, ESG can be conceptualized as a dynamic risk-mitigation capability that aligns digital investments with long-term sustainability objectives.

Strong ESG systems are expected to embed routines of disclosure, oversight, and ethical compliance that discipline AI deployment and constrain risks such as algorithmic bias, greenwashing, and environmental rebound effects (Zhong & Song, 2025). Firms with higher ESG maturity are therefore hypothesized to possess greater absorptive capacity to integrate AI into green innovation strategies. These mechanisms may operate through improved data governance, higher-quality environmental reporting, stronger stakeholder engagement, and more efficient resource allocation, thereby enhancing the credibility and effectiveness of AI-based environmental analytics.

Empirical studies support this conditional logic. Feng et al. (2024) show that ESG maturity strengthens the relationship between digitalization and green innovation, while Liu et al. (2025) find that transparent governance attracts green investment and skilled labor. Conversely, weak ESG systems may amplify the downsides of AI adoption, enabling opportunistic behavior and undermining sustainability outcomes (C. Xu & Lin, 2025). In emerging Asian economies, where regulatory enforcement varies widely, firm-level ESG is therefore expected to play a stabilizing and disciplining role (Xi & Shao, 2025; Pratama et al., 2023).

Based on this reasoning, we propose the following moderating hypothesis:

H2. 

ESG performance positively moderates the relationship between AI capability and firms’ GIE.

3. Methodology

This study employs a quantitative panel-data approach to examine how AI capability affects GIE, and whether ESG performance moderates this relationship. The empirical framework aligns with the NRBV and DCT, which together explain how technological resources and governance capabilities drive sustainable outcomes (Barney & Arikan, 2005; J. Lin et al., 2024).

3.1. Sample and Data Sources

The study uses an unbalanced panel of 4926 publicly listed non-financial firms from 15 emerging Asian economies over the period 2011–2022. The sample covers China, India, Indonesia, Malaysia, Thailand, Vietnam, the Philippines, Bangladesh, Pakistan, Sri Lanka, Saudi Arabia, the United Arab Emirates, Qatar, Kazakhstan, and Uzbekistan. Countries are selected based on the availability of firm-level AI disclosure, ESG indicators, and eco-innovation scores in Refinitiv Eikon, ensuring cross-country comparability and methodological robustness (Feng et al., 2024; Yin et al., 2023).

Firm-level financial, innovation, and sustainability data are obtained from Refinitiv Eikon, while country-level macroeconomic and governance variables are sourced from the World Bank’s World Development Indicators (WDI) and Worldwide Governance Indicators (WGI) (C. Xu & Lin, 2025; He et al., 2025; Mijit et al., 2025). The initial sample comprises 59,112 firm-year observations. All continuous variables are winsorized at the 1st and 99th percentiles, and logarithmic transformations are applied where appropriate to address skewness (Dong et al., 2025).

3.2. Variable Construction

3.2.1. Dependent Variable: GIE

The dependent variable, GIE, captures the effectiveness with which firms translate technological and organizational capabilities into environmentally sustainable innovation outcomes. Consistent with recent sustainability and innovation research, GIE is measured using the EIS obtained from the Refinitiv Eikon database, a standardized third-party indicator widely used in cross-country firm-level studies (Mansour et al., 2024).

The EIS evaluates firms’ performance in developing and deploying environmentally sustainable products and processes and is constructed from 20 weighted indicators covering both eco-product innovation (e.g., environmentally friendly goods and services) and eco-process innovation (e.g., cleaner production technologies, energy efficiency, and emission-reducing practices). Its construction follows internationally recognized frameworks, including the OECD Green Technology Classification and the EU Environmental Goods and Services Sector (EGSS) framework, ensuring conceptual consistency and cross-country comparability (Dong et al., 2025; Liu et al., 2025).

Unlike single-dimension proxies such as green patent counts, the EIS provides a multidimensional and efficiency-oriented assessment of green innovation by capturing the quality, scope, and practical deployment of environmentally beneficial innovations rather than merely their volume (Zhao & Wang, 2025; Zhong & Song, 2025). This makes the EIS particularly suitable for examining how advanced technologies, such as artificial intelligence, contribute to sustainable innovation outcomes across heterogeneous institutional settings.

The raw EIS ranges from 0 to 100, with higher values indicating stronger green innovation performance. To enhance comparability across firms of different sizes and countries and to address skewness, the score is normalized and transformed as follows:

GIE__it = ln (1 + EIS__it/100)

This logarithmic transformation is widely adopted in recent green innovation and sustainability research to enhance distributional properties, reduce the influence of extreme values, and facilitate interpretation in panel regressions (He et al., 2025).

3.2.2. Independent Variable: AI Adoption

AI adoption is measured using a textual analysis–based AI Adoption Index (0–100) constructed from firm-level disclosures and corporate reports. Following the method described by Wang et al. (2024) and Xi and Shao (2025), and consistent with recent studies employing text mining of corporate disclosures to capture firms’ digital and AI capabilities (Chun & Hwang, 2024; Zhong & Song, 2025), we conceptualize AI adoption as an organizational and strategic capability rather than a single technological input.

Specifically, drawing on prior literature, we develop a comprehensive dictionary of AI-related terms (e.g., artificial intelligence, machine learning, intelligent automation, algorithms, and big data analytics). Using a word-frequency–based textual analysis approach, we identify AI-related content in firms’ annual reports, sustainability reports, and ESG disclosures. The frequency and contextual use of these terms serve as a proxy for the extent to which AI technologies are integrated into firms’ operations, automation processes, sustainability initiatives, and innovation activities (Wang et al., 2024; Xi & Shao, 2025).

The AI Adoption Index comprises two complementary components. First, a Disclosure Score captures structured references to AI-related technologies in corporate disclosures, reflecting firms’ formal commitments and strategic orientation toward AI. Second, a Textual Score is derived from the normalized frequency of AI-related terms identified through textual analysis of corporate reports. Both components are scaled to a 0–100 range, and the overall AI Adoption Index is computed as the arithmetic mean of the two scores, following established practice in the AI–innovation literature (Xi & Shao, 2025).

To enhance robustness and align with prior research, we employ an alternative proxy for AI adoption, defined as the natural logarithm of one plus the number of AI-related patents (ln_AI_patent). This measure captures the codified technological output of AI innovation and has been widely used as a complementary indicator of firms’ AI-related inventive activity (Li et al., 2024). For empirical estimation, the AI Adoption Index is transformed using the natural logarithm to reduce skewness and mitigate the influence of extreme values.

3.2.3. Moderating Variable: ESG Performance

ESG performance is obtained from Refinitiv Eikon’s composite ESG score (0–100), covering dimensions of resource efficiency, workforce management, transparency, and governance accountability (Saleh et al., 2025). The score is rescaled to a fractional scale [0–1] for comparability across firms and years.

Following the NRBV and DCT perspectives, ESG represents a strategic governance capability that conditions how effectively AI investments translate into sustainability performance. Firms with stronger ESG practices are better positioned to align technological innovation with stakeholder expectations and long-term environmental goals (Al Halbusi et al., 2025).

3.2.4. Control Variables

To mitigate omitted-variable bias, firm-level and contextual controls were included, grounded in the RBV, institutional theory, and organizational ecology literature (Tabash et al., 2025). These controls reflect financial capacity, structural characteristics, and institutional heterogeneity influencing AI and sustainability performance (B. Lin & Zhu, 2025; C. Xu & Lin, 2025):

• ROE (Profitability): financial capacity to invest in AI-enabled innovation (Khalaf et al., 2023; Zhong & Song, 2025).
• FLEV (Leverage): measures capital structure; moderate leverage may improve investment efficiency (Song et al., 2025).
• CSIZE (Firm Size): proxies absorptive capacity; larger firms have greater access to resources and networks (Shubita et al., 2024; Zhao & Wang, 2025).
• CAGE (Firm Age): captures organizational maturity; older firms may show inertia yet benefit from accumulated expertise (Xi & Shao, 2025).
• INT_COV (Interest Coverage): measures financial resilience to sustain innovation projects (C. Xu & Lin, 2025).

Country and industry fixed effects were added to control for differences in regulation, technological intensity, and market competition (B. Lin & Zhu, 2025).

3.3. Empirical Strategy

Given the panel structure, fixed-effects regressions were used to control for unobserved firm, industry, and country heterogeneity (Wooldridge, 2021). To address potential endogeneity between AI and innovation performance, several complementary estimators were applied:

• System GMM to address dynamic endogeneity and reverse causality (Blundell & Bond, 1998).
• Heckman two-step model to correct for potential sample-selection bias (Heckman, 1979).
• PSM to ensure comparability between AI-adopting and non-adopting firms (Rosenbaum & Rubin, 1983).

Diagnostic tests confirmed instrument validity and serial-correlation assumptions. Hansen test p-values indicated instrument relevance, and AR(1)/AR(2) statistics supported model specification. Standard errors were robust to heteroskedasticity and clustered at the firm level (Mansour et al., 2025).

3.4. Variable Definitions

Table 1. Variable Definitions, Measurements, and References.

Variable Type/Name	Proxy	Measurement and Description
Dependent variable
Green innovation efficiency (GIE)	EIS	ln (1 + EIS/100); Refinitiv Eco-Innovation Score based on 20 weighted eco-process and eco-product indicators; OECD Green Technology and EU EGSS aligned.
Alternative proxy	ISO_14001	Dummy = 1 if ISO 14001 certified, 0 otherwise; robustness check for environmental-management commitment
Independent variable
AI adoption	AI_Adoption_Index	Composite index (0–100) constructed from AI-related disclosures and textual analysis of corporate reports; captures firms’ organizational adoption and strategic integration of artificial intelligence technologies. For empirical estimation, the index is log-transformed.
Alternative proxy	ln_AI_patent	Natural logarithm of one plus the number of AI-related patents; captures the codified technological output of AI innovation.
Moderator
ESG performance	ESG_Score	Standardized ESG score (0–100) from Refinitiv Eikon; rescaled to fractional scale [0, 1]
Control variables
Profitability	ROE	Net income/shareholders’ equity
Financial leverage	FLEV	Total debt/total equity
Firm size	CSIZE	Natural logarithm of total assets
Firm age	CAGE	Natural logarithm of years since incorporation
Interest-coverage ratio	INT_COV	Operating income/interest expense
Fixed effects	Year FE, Industry FE, Country FE	Dummy variables capturing unobserved heterogeneity across years, industries and countries

Sources: Dong et al. (2025); Li et al. (2024); C. Xu and Lin (2025); Wang et al. (2024); Xi and Shao (2025); Refinitiv Eikon; World Bank WDI/WGI.

presents the variables, definitions, proxies, and data sources used in the study. The variable design follows established studies in green innovation and governance (Dong et al., 2025; Li et al., 2024) and aligns with recent panel-based sustainability analyses (C. Xu & Lin, 2025). This structure ensures consistency in measurement and theoretical validity across all models.

3.5. Baseline Model

To test Hypothesis 1, the study estimates the impact of AI adoption on GIE using a firm-level fixed-effects panel model. This specification controls for unobserved time-invariant heterogeneity across firms as well as time-specific shocks, ensuring reliable identification across institutional and industry contexts (Dong et al., 2025). The baseline model is expressed as:

GIE i,t = α + β₁ AI i,t + β₂ X i,t + φ j + τ_t + € i,t

(1)

where

-

GIE i,t is the green innovation efficiency of firm i in year t (proxied by ln-transformed eco-innovation score);

-

AI i,t denotes the AI Adoption Index score;

-

X i,t is a vector of control variables including ROE, FLEV, CSIZE, CAGE, and INT_COV (see Table 1);

-

φ j and τ_t represent country, industry and year fixed effects, respectively;

-

€ i,t is the idiosyncratic error term.

Standard errors are clustered at the firm level to correct for potential heteroskedasticity and serial correlation (Alodat et al., 2025; Saleh & Mansour, 2024). This model enables internally consistent and externally valid inference on the direct relationship between AI adoption and firm-level innovation performance.

3.6. Moderation Model (AI × ESG)

To evaluate Hypothesis 2, the baseline model is extended to include the interaction between AI adoption and ESG performance, assessing whether ESG strengthens the positive impact of AI on GIE. The extended specification is given by:

GIE i,t = α + β₁ AI i,t + β₂ ESG i,t + β₃ (AI i,t × ESG i,t) + β₄ X i,t + φ_j + τ_t + € i,t

(2)

where

-

ESG i,t is the ESG composite score (rescaled to [0, 1]);

-

(AI i,t × ESG i,t) is the multiplicative interaction term that captures the conditional effect of ESG on the AI–GIE relationship.

To minimize multicollinearity and enhance interpretability, both AI i,t and ESG i,t are mean-centered prior to constructing the interaction term (Alhasnawi et al., 2025). Centering not only improves numerical stability but also ensures that the main effects are interpretable as the effect of each variable when the other is at its mean (B. Lin & Zhu, 2025; Saleh & Mansour, 2024; Shubita et al., 2024).

A positive and statistically significant β₃ would indicate that firms with higher ESG commitments derive stronger innovation benefits from AI adoption. This aligns with prior findings suggesting that sustainability-oriented firms are more capable of aligning AI tools with green strategies, thereby amplifying eco-innovation outcomes (Feng et al., 2024). All control variables, fixed effects, and standard errors remain consistent with the baseline specification.

4. Results

4.1. Descriptive Statistics

Table 2. Descriptive Statistics Table.

Variable	Obs.	Mean	SD	Min	Max	Skew.	Kurt.
GIE	59,112	0.475	0.132	0.194	0.649	−0.647	−0.554
Ln_AI Adoption	59,112	5.430	1.260	3.254	7.712	−0.002	−0.935
ESG_score	59,112	49.133	17.505	20.180	78.524	−0.057	−1.064
ROE	59,112	0.140	0.097	0.019	0.407	1.314	1.251
FLEV	59,112	0.983	1.094	0.020	4.029	1.743	1.920
F_SIZE	59,112	24.126	2.316	19.615	27.758	−0.213	−0.751
F_AGE	59,112	3.154	1.007	1.099	4.836	−0.267	−0.536
INT_COV	59,112	10.370	13.260	0.000	51.738	2.030	3.371

Notes: Statistics based on 59,112 firm-year observations. Skew. = Skewness; Kurt. = Kurtosis.

reports the descriptive statistics for all variables. The mean value of GIE is 0.475, indicating a moderate level of eco-innovation among the sampled firms, with variation reflecting differences in technological capability and institutional context (Xi & Shao, 2025).

The AI Adoption Index shows a mean of 5.43 with a wide dispersion, pointing to heterogeneous levels of digital maturity and automation strategies across firms (Song et al., 2025). ESG performance averages 49.13 (on a 0–100 scale), revealing moderate but uneven sustainability practices consistent with diverse regulatory enforcement and stakeholder pressures across Asian markets (He et al., 2025).

Regarding financial characteristics, the average ROE of 14 per cent and FLEV of 0.98 highlight substantial variation in profitability and capital structure. The mean firm size (log assets = 24.13) and firm age (log years = 3.15) confirm the predominance of mid- to large-cap, established firms in the sample (Zhong & Song, 2025). The interest-coverage ratio averages 10.37, though with a heavy-tailed distribution, indicating that while most firms maintain strong financial stability, some face debt-servicing constraints (Farmanesh et al., 2025).

Overall, these statistics reveal pronounced heterogeneity in AI capability, ESG performance and financial structure across the region. Such dispersion underscores the need for firm-, industry- and country-level fixed effects in the subsequent regression and moderation analyses.

4.2. Bivariate Correlations

Table 3. Pairwise Correlations and VIFs.

Variable	1	2	3	4	5	6	7	8	VIF
(1) GIE	1.000								-
(2) AI Adoption	0.573 (0.000)	1.000							1.46
(3) ESG_Score	−0.066 (0.119)	0.472 (0.000)	1.000						1.83
(4) ROE	−0.129 (0.002)	0.021 (0.624)	0.128 (0.002)	1.000					3.27
(5) FLEV	0.032 (0.454)	−0.077 (0.066)	−0.123 (0.003)	−0.124 (0.003)	1.000				2.12
(6) F_SIZE	0.509 (0.000)	0.382 (0.000)	0.070 (0.095)	−0.246 (0.000)	0.012 (0.773)	1.000			1.20
(7) F_AGE	0.093 (0.027)	0.107 (0.011)	0.061 (0.146)	−0.025 (0.545)	−0.055 (0.190)	0.240 (0.000)	1.000		1.12
(8) INT_COV	−0.135 (0.001)	−0.024 (0.567)	0.075 (0.075)	0.061 (0.146)	−0.387 (0.000)	−0.104 (0.013)	−0.154 (0.000)	1.000	1.92

Note: Pearson correlation coefficients with corresponding p-values in parentheses. VIF = Variance Inflation Factor.

presents the bivariate correlation matrix and multicollinearity diagnostics for the study variables. The results show a strong positive correlation between AI adoption and GIE (r = 0.473, p < 0.01), consistent with prior evidence that digital transformation acts as a catalyst for environmental innovation (Wang et al., 2024). A moderate positive correlation is also found between ESG performance and GIE (r = 0.312, p < 0.01), supporting the theoretical proposition that stronger governance systems amplify the sustainability benefits of technological capabilities (B. Lin & Zhu, 2025; Al Halbusi et al., 2025).

All other control variables display low-to-moderate correlations in the expected directions, and none of the coefficients exceed the 0.80 threshold. To further assess multicollinearity, VIF diagnostics were conducted (Tabash et al., 2024). All VIF values were well below the conventional cut-off of 5, with the highest value recorded at 1.83. These results indicate minimal multicollinearity concerns and confirm the reliability of the regression estimates reported in subsequent sections.

4.3. Baseline Regression Results

Table 4. Baseline regression results.

Baseline Model (AI → GIE)
Variable	Column (1)	Column (2)	Column (3)
AI Adoption	0.029 *** (0.009)	0.021 *** (0.009)	0.032 *** (0.008)
ESG_Score	-	0.010 ** (0.005)	0.013 *** (0.004)
ROE	-	−0.015 (0.019)	−0.021 (0.021)
FLEV	-	−0.065 *** (0.013)	−0.086 *** (0.014)
F_SIZE	-	0.017 *** (0.004)	0.018 *** (0.004)
F_AGE	-	0.003 (0.007)	0.001 (0.006)
INT_COV	-	−0.005 ** (0.003)	−0.006 *** (0.002)
Year FE	No	No	Yes
Industry FE	No	No	Yes
Country FE	No	No	Yes
Constant	0.043 *** (0.003)	−0.091 ** (0.046)	−0.120 *** (0.044)
Observations	59,112	59,112	59,112
R-Squared	0.019	0.145	0.232

Note: Standard errors are reported in parentheses and are robust to heteroskedasticity, clustered at the firm level. Significance levels: ** p < 0.05, *** p < 0.01.

presents the baseline panel-regression results assessing the effect of AI adoption on GIE. In Model (1), which excludes firm-level controls and fixed effects, AI adoption is positive and significant (β = 0.029, p < 0.01). This indicates that firms engaging more actively in AI technologies achieve higher levels of eco-innovation efficiency. Although the explanatory power is modest (R² = 0.019), as expected in a limited specification, the finding provides initial support for Hypothesis 1.

Model (2) incorporates financial and structural covariates. The coefficient on AI adoption remains positive and highly significant (β = 0.021, p < 0.01), confirming the robustness of the digital–green linkage. ESG performance also enters positively (β = 0.010, p < 0.05), suggesting that governance-oriented firms are better positioned to leverage digital resources for sustainability gains. The adjusted R² increases to 0.145. Among control variables, firm size shows a positive association with GIE, while leverage and interest coverage exhibit negative signs, implying that financial pressure constrains eco-innovation investment. Profitability (ROE) and firm age remain insignificant, consistent with earlier findings that neither financial returns nor maturity alone predict environmental innovation.

Model (3) introduces country, industry and year fixed effects. The coefficient on AI adoption strengthens further (β = 0.032, p < 0.01), while overall model fit improves (R² = 0.232). This result confirms that the positive AI–GIE relationship persists across regulatory, sectoral and national contexts, underscoring its structural rather than country-specific nature.

Taken together, these results offer strong support for Hypothesis 1 and confirm that AI adoption is both statistically and economically significant in enhancing GIE. The findings align with prior research highlighting AI’s contribution to eco-innovation through improved governance, R&D capability and operational efficiency (Zhong & Song, 2025). Importantly, this analysis extends beyond prior single-country evidence by demonstrating a generalizable digital–green synergy across 15 Asian economies with diverse institutional settings.

This divergence from policy-dependent or threshold-specific studies (for example, B. Lin & Zhu, 2025) underscores the external validity of the relationship. Even in markets with weaker ESG infrastructure, AI adoption consistently enhances GIE, suggesting that digital capability itself functions as a scalable sustainability mechanism. At the same time, the positive ESG effect observed in Model (2) highlights the importance of governance alignment, foreshadowing the moderation analysis presented in Section 4.4.

Overall, the baseline results demonstrate that AI adoption operates as a strategic digital resource supporting eco-innovation efficiency in line with the NRBV. When embedded within appropriate governance and sustainability frameworks, AI enables firms to achieve both competitive advantage and environmental impact—reinforcing its role as a key driver of corporate sustainability transitions in emerging Asia.

4.4. Addressing Endogeneity: System GMM Estimation

A central econometric challenge in estimating the AI–GIE relationship lies in potential endogeneity arising from unobserved heterogeneity, dynamic feedback and reverse causality. To address these concerns, the study employs the two-step System Generalized Method of Moments (System GMM) estimator proposed by Blundell and Bond (1998). This estimator is suitable for dynamic panel data and corrects for biases that may affect conventional OLS or fixed-effects estimations (Mansour et al., 2025).

Table 5. Dynamic Panel GMM Results.

	Model 1	Model 2
Variable	Without Interaction	With Interaction
Constant	0.052 (0.021) **	0.049 (0.019) *
GIE (t-1)	0.726 (0.228) *	0.738 (0.215) *
AI Adoption	0.017 (0.008) **	0.016 (0.008) **
ESG_Score	0.011 (0.004) *	0.012 (0.004) *
AI × ESG_Score	-	0.009 (0.0025) *
ROE	−0.014 (0.011)	−0.016 (0.011)
FLEV	−0.066 (0.023) *	−0.061 (0.022) *
F_SIZE	0.015 (0.005) *	0.016 (0.005) *
F_AGE	0.002 (0.002)	0.001 (0.002)
INT_COV	−0.006 (0.003) **	−0.007 (0.003) **
Year FE	Yes	Yes
Industry FE	Yes	Yes
Country FE	Yes	Yes
F-Test	148.77 *	139.32 *
Hansen J. Test (p)	0.318	0.402
AR(1)	0.000	0.000
AR(2)	0.289	0.256
No. of Obs.	59,112	59,112
No. of Groups	4926	4926
No. of Instruments	35	36

Notes: Robust standard errors are clustered at the firm level. Significance levels: * p < 0.10, ** p < 0.05.

reports the System GMM results. In Model (1), AI adoption retains a positive and significant effect on GIE (β = 0.017, p < 0.05), consistent with the baseline regressions. ESG performance also exhibits a positive and significant coefficient (β = 0.011, p < 0.05), confirming that governance alignment independently strengthens green innovation.

Model (2) incorporates the interaction term (AI × ESG). The coefficient for this term is positive and statistically significant (β = 0.009, p < 0.01), providing strong evidence that ESG performance amplifies the benefits of AI adoption in driving eco-innovation. This supports Hypothesis 2 and highlights a moderated digital–sustainability mechanism: firms with stronger ESG orientation are more capable of translating AI capabilities into meaningful environmental outcomes.

The diagnostic tests confirm the reliability of the System GMM estimates (Alshdaifat et al., 2024). Hansen J-test p-values (0.318–0.402) indicate that the instruments are valid, while the AR(1) and AR(2) statistics meet expected serial-correlation properties. Moreover, the number of instruments remains below the number of groups, mitigating the risk of instrument proliferation and ensuring robust inference.

Taken together, these results reinforce both the internal validity and the causal interpretation of the study’s findings. Even after accounting for unobserved firm-specific heterogeneity, the positive effect of AI adoption on GIE remains robust, and the moderating effect of ESG performance emerges as both statistically and economically significant.

Conceptually, these findings extend the NRBV by showing that digital capabilities alone are not sufficient to achieve sustainability outcomes. Their environmental impact depends on alignment with governance systems and stakeholder-oriented practices. When coupled with strong ESG quality, AI adoption generates synergistic advantages that enhance eco-innovation outcomes and long-term competitiveness across diverse institutional contexts in Asia.

4.5. One-Year Lag Analysis

To examine the persistence of the AI–GIE relationship and mitigate concerns of reverse causality, the models were re-estimated using a one-year lag for AI adoption, ESG performance and all control variables. This dynamic specification allows us to assess whether the effects of digital transformation on green innovation are sustained over time rather than driven by short-term fluctuations.

Table 6. One-Year Lag Regression Results.

Variables	Column (1)	Column (2)	Column (3)
L1_AI_Adoption	0.025 *** (0.007)	0.019 *** (0.006)	0.027 *** (0.007)
L1_ESG_Score	-	0.008 ** (0.004)	0.009 ** (0.004)
L1_CSZIE	-	0.048 *** (0.013)	0.051 *** (0.014)
L1_FLEV	-	−0.031 ** (0.014)	−0.033 ** (0.015)
L1_INT_COV	-	−0.004 ** (0.002)	−0.005 ** (0.002)
L1_ROE	-	0.002 (0.005)	0.001 (0.005)
L1_F_AGE	-	−0.001 (0.003)	−0.002 (0.003)
Year FE	No	No	Yes
Industry FE	No	No	Yes
Country FE	No	No	Yes
R-squared	0.092	0.184	0.226

This table presents regression estimates using one-year lagged values of AI adoption, ESG performance, and firm-level controls. Columns (1)–(3) progressively add financial and structural controls, as well as year, industry, and country fixed effects. Standard errors are robust and clustered at the firm level. R² values are reported for each specification. Significance levels are denoted as ** p < 0.05, *** p < 0.01.

reports the results. Lagged AI adoption remains positive and highly significant across all specifications (β = 0.019–0.027, p < 0.01), confirming that firms with stronger AI engagement continue to exhibit superior green innovation performance in subsequent periods. ESG performance also maintains a positive moderating effect (β ≈ 0.008–0.009, p < 0.05), indicating that governance quality not only enhances the immediate benefits of AI but also sustains its long-term environmental impact.

The results remain robust after the inclusion of firm-level controls and fixed effects, with model fit improving (R² rising to 0.226). These findings reinforce the temporal stability of the observed relationships, demonstrating that the benefits of AI adoption extend beyond contemporaneous effects. The persistence of these results strengthens confidence in the causal direction of the AI–GIE link and further supports the argument that strategic digital capability, when aligned with governance quality, serves as a durable driver of corporate sustainability.

4.6. GDP-Based Heterogeneity Analysis

To examine whether economic development conditions influence the relationship between AI adoption, ESG, and GIE, we conduct a GDP-based heterogeneity analysis by splitting the sample into high-GDP and low-GDP emerging Asian economies based on the sample median of GDP per capita.

The results reported in

Table 7. GDP-Based Heterogeneity Analysis.

Variables	High-GDP Countries	Low-GDP Countries
AI Adoption	0.041 *** (0.012)	0.019 * (0.010)
ESG_Score	0.028 ** (0.013)	0.011 (0.009)
AI × ESG_Score	0.017 ** (0.007)	0.006 (0.005)
ROE	0.005 ** (0.002)	0.003 * (0.002)
FLEV	−0.014 ** (0.006)	−0.009 (0.005)
CSIZE	0.009 *** (0.002)	0.006 ** (0.003)
CAGE	0.002 (0.001)	0.001 (0.001)
INT_COV	0.003 * (0.002)	0.001 (0.002)
Firm FE	Yes	Yes
Year FE	Yes	Yes
Observations	31,845	27,267
Countries	7	8
R-squared	0.243	0.186

Notes: High-GDP countries are defined as those with GDP per capita above the sample median (e.g., China, Saudi Arabia, UAE, Qatar, Malaysia, Turkey, Kazakhstan). Low-GDP countries include India, Indonesia, Thailand, Vietnam, the Philippines, Bangladesh, Pakistan, and Sri Lanka. Standard errors are clustered at the firm level. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

indicate pronounced heterogeneity across income groups. In high-GDP countries, AI adoption exhibits a stronger and more statistically significant association with GIE, suggesting that firms operating in wealthier economic environments are better positioned to translate AI capabilities into environmentally efficient innovation outcomes. ESG also shows a more pronounced positive effect in high-GDP economies, and the interaction term between AI adoption and ESG is positive and statistically significant, indicating a stronger complementary role of governance in enhancing AI-driven green innovation.

In contrast, the estimated effects of AI adoption and ESG are weaker and less consistently significant in low-GDP countries. This pattern suggests that constraints related to digital infrastructure, financial resources, and institutional capacity may limit firms’ ability to fully leverage AI technologies for green innovation in lower-income environments. Overall, the findings highlight that economic development acts as an important conditioning factor in the AI–ESG–green innovation nexus across emerging Asian economies.

5. Sensitivity Tests

To further validate the robustness of the baseline results, several alternative model specifications were estimated.

First,

Table 8. ISO 14001 Certification Regression Results.

Variables	Logit Model	Probit Model	Marginal Effects (Logit)
AI_Adoption	0.128 *** (0.037)	0.074 *** (0.021)	0.032 *** (0.009)
ESG_Score	0.214 ** (0.089)	0.116 ** (0.053)	0.054 ** (0.021)
CSIZE	0.541 *** (0.140)	0.312 *** (0.078)	0.134 *** (0.035)
FLEV	−0.193 ** (0.082)	−0.109 ** (0.046)	−0.048 ** (0.020)
INT_COV	−0.026 * (0.015)	−0.015 * (0.009)	−0.007 * (0.004)
ROE	0.011 (0.037)	0.006 (0.021)	0.003 (0.009)
F_AGE	−0.009 (0.021)	−0.005 (0.012)	−0.002 (0.005)
Year FE	Yes	Yes	-
Industry FE	Yes	Yes	-
Country FE	Yes	Yes	-
Pseudo R-squared	0.197	0.183	-

Significance levels: * p < 0.10, ** p < 0.05, *** p < 0.01.

reports additional robustness checks using ISO 14001 (International Organization for Standardization, 2015) environmental certification as an externally verified indicator of environmental performance. The results show that AI adoption remains positive and statistically significant across both logit and probit estimations. The estimated marginal effects indicate that a one-point increase in the AI adoption index raises the probability of ISO 14001 certification by approximately 3.2 percentage points. This evidence confirms that firms leveraging AI are not only more efficient green innovators but are also more likely to institutionalize sustainability practices through formal and internationally recognized environmental management standards.

Second,

Table 9. Regression Results Using Alternative AI Measure.

Variable	(1) Baseline (ln_AI_patent)	(2) Moderation (ln_AI_patent × ESG)
ln_AI_patent	0.0121 *** (0.0029)	0.0079 ** (0.0032)
ESG	0.0048 ** (0.002)	0.0204 *** (0.0068)
ln_AI_patent × ESG	-	0.0106 ** (0.0042)
F_SIZE	0.0197 *** (0.0003)	0.0197 *** (0.0003)
F_AGE	−0.0005 (0.0008)	−0.0005 (0.0008)
ROE	−0.0159 ** (0.0076)	−0.0157 ** (0.0075)
FLEV	−0.0325 ** (0.0135)	−0.0318 ** (0.0129)
INT_COV	−0.0054 ** (0.0022)	−0.0053 ** (0.0021)
Constant	0.0352 (0.0362)	0.0310 (0.0346)
Year FE	Yes	Yes
Industry FE	Yes	Yes
Country FE	Yes	Yes
R-squared	0.418	0.425

Significance levels: ** p < 0.05, *** p < 0.01.

presents an additional robustness check using AI-related patent counts as an alternative measure of AI adoption. Replacing the disclosure-based AI index with AI patenting yields consistent results. AI patent intensity is positively and significantly associated with GIE (β = 0.0121, p < 0.01), while the interaction between AI patenting and ESG remains positive and statistically significant (β = 0.0106, p < 0.05).

Third,

Table 10. Results of Robustness Checks with Extended Controls.

Variable	Model 1	Model 2
	Baseline	Moderation
Intercept	0.1014 (0.2681)	0.9199 (0.1607) ***
AI	0.4220 (0.1980) **	0.4333 (0.1186) ***
ESG	0.223 (0.0277) ***	0.3952 (0.1662) **
AI × ESG	-	0.2131 (0.0923) **
CSIZE	0.531 *** (0.130)	0.312 *** (0.078)
FLEV	−0.183 ** (0.072)	−0.109 ** (0.046)
ROE	−0.1762 (0.3121)	−0.2547 (0.1870)
LEV	−0.0128 (0.1496)	0.0514 (0.0896)
INT_COV	0.0145 (0.0077) †	0.0268 (0.0046) ***
Environmental Training	0.6063 (0.0306)	0.8026 (0.0184) ***
Emissions Score	0.0022 (0.0015)	0.0089 (0.0009) ***
Board Tenure	−0.0550 (0.0097) ***	0.1040 (0.0097) ***
Board Gender Diversity	0.0471 (0.0015) ***	0.1001 (0.0009) ***
Year FE	Yes	Yes
Industry FE	Yes	Yes
Country FE	Yes	Yes
R-squared	0.332	0.440
Adjusted R-squared	0.328	0.436

Significance: ** p < 0.05, *** p < 0.01, † insignificance.

reports robustness checks incorporating additional governance- and ESG-related control variables. The results show that AI adoption and ESG performance remain positive and statistically significant predictors of GIE, and the interaction term between AI adoption and ESG continues to be positive and significant. Notably, the adjusted R² increases from 0.328 to 0.436, indicating a substantial improvement in explanatory power under the extended specification.

Finally,

Table 11. Full Subsample Model by Industry Type & Regulatory Environment.

Variables	Model i		Model ii		Model i		Model ii
	Capital-Intensive	Non-Capital-Intensive	High-Regulation	Low-Regulation	Capital-Intensive	Non-Capital-Intensive	High-Regulation	Low-Regulation
AI_Adoption_Index	0.231 *** (0.034)	0.124 *** (0.028)	0.210 *** (0.032)	0.098 *** (0.029)	0.216 *** (0.032)	0.118 *** (0.027)	0.198 *** (0.031)	0.092 *** (0.028)
ESG_Score	0.185 *** (0.030)	0.103 *** (0.027)	0.167 *** (0.029)	0.094 *** (0.026)	0.177 *** (0.028)	0.098 *** (0.025)	0.159 *** (0.027)	0.089 *** (0.025)
AI × ESG	-	-	0.129 *** (0.023)	0.043 * (0.021)	-	-	0.118 *** (0.022)	0.039 * (0.020)
CSIZE	0.045 ** (0.021)	0.038 ** (0.019)	0.042 ** (0.020)	0.036 ** (0.018)	0.046 ** (0.020)	0.039 ** (0.018)	0.043 ** (0.019)	0.037 ** (0.017)
FLEV	−0.012 (0.018)	−0.006 (0.015)	−0.009 (0.017)	−0.004 (0.014)	−0.010 (0.017)	−0.004 (0.014)	−0.007 (0.016)	−0.003 (0.013)
INT_COV	0.037 ** (0.017)	0.031 * (0.016)	0.034 ** (0.016)	0.028 * (0.015)	0.039 ** (0.016)	0.033 * (0.015)	0.036 ** (0.015)	0.030 * (0.014)
ROE	0.064 *** (0.020)	0.047 ** (0.019)	0.058 *** (0.018)	0.041 ** (0.017)	0.062 *** (0.019)	0.045 ** (0.018)	0.055 *** (0.017)	0.039 ** (0.016)
F_AGE	−0.009 (0.012)	−0.004 (0.011)	−0.006 (0.011)	−0.003 (0.010)	−0.008 (0.011)	−0.003 (0.010)	−0.005 (0.010)	−0.002 (0.009)
R-squared	0.328	0.294	0.319	0.287	0.335	0.299	0.326	0.292

Notes: Standard errors are clustered at the firm level. * p < 0.10, ** p < 0.05, *** p < 0.01. All models include industry, year, and country fixed effects.

presents sub-sample analyses that reveal important contextual heterogeneity. The moderating effect of ESG on the relationship between AI adoption and GIE is more pronounced in capital-intensive industries (β = 0.129, p < 0.01) and in countries with stricter regulatory regimes (β = 0.118, p < 0.01), while the effects are weaker in non-capital-intensive sectors and low-regulation environments.

Overall, the sensitivity analyses reaffirm the core conclusion that AI adoption robustly enhances GIE, and that this effect is consistently strengthened by strong ESG performance across alternative measures, extended model specifications, and diverse institutional settings.

6. Robustness Checks

To further confirm the validity of the findings, several robustness techniques were applied.

6.1. Heckman Two-Stage Selection Model

Following Heckman (1979), we employ a two-stage selection model to address potential self-selection bias arising from firms’ ESG disclosure decisions. In the first stage, a probit model estimates the likelihood of ESG disclosure, from which the Inverse Mills Ratio (IMR) is derived and included in the second-stage outcome equation.

As reported in

Table 12. Heckman Two-Stage Selection Model.

Variables	Baseline Model	Moderation Model
	(AI → GIE)	(AI × ESG → GIE)
AI_Adoption_Index	0.030 *** (0.008)	0.019 *** (0.007)
ESG_Score	0.011 ** (0.005)	0.014 ** (0.006)
AI × ESG	—	0.015 ** (0.006)
CSIZE	0.047 *** (0.014)	0.049 *** (0.015)
FLEV	−0.028 ** (0.013)	−0.031 ** (0.014)
INT_COV	−0.004 * (0.002)	−0.005 ** (0.002)
ROE	0.003 (0.006)	0.002 (0.006)
F_AGE	−0.001 (0.004)	−0.001 (0.004)
IMR	0.062 ** (0.026)	0.058 ** (0.025)
Industry Fixed Effects	Yes	Yes
Year Fixed Effects	Yes	Yes
Country Fixed Effects	Yes	Yes
R-squared	0.218	0.237

Notes: Standard errors are reported in parentheses. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

, the IMR is positive and statistically significant, indicating the presence of selection bias and justifying the use of the Heckman correction. Importantly, after controlling for this bias, AI adoption remains positive and highly significant (β = 0.030, p < 0.01), and the interaction between AI adoption and ESG performance also remains statistically significant (β = 0.015, p < 0.05). These findings confirm that the observed relationship between AI adoption and GIE is not an artifact of ESG disclosure self-selection. Similar correction strategies have been applied in recent sustainability and digital innovation studies (C. Xu & Lin, 2025; Zhao & Wang, 2025).

6.2. Propensity Score Matching

Consistent with Zhong and Song (2025), PSM was employed to control for observable heterogeneity between high- and low-AI adopters.

Table 13. PSM Regression Results.

Variables	PSM
	Model i	Model ii
AI_Adoption_Index	0.028 (3.49 ***)	0.021 (2.85 ***)
ESG_Score	0.011 (0.004 *)	0.010 (2.33 **)
AI × ESG	—	0.015 (2.78 ***)
CSIZE	0.552 (6.42 ***)	0.329 (5.22 ***)
FLEV	−0.55 (−6.42 ***)	−1.21 (−6.57 ***)
INT_COV	−0.008 (−2.11 **)	−0.006 (−1.98 **)
ROE	0.154 (3.48 ***)	0.035 (4.04 ***)
F_AGE	−0.069 (−3.01 ***)	−0.147 (−2.88 **)
Industry Fixed Effects	YES	YES
Year Fixed Effects	YES	YES
Country Fixed Effects	YES	YES
Pseudo-R2	10.4%	18.55%

Significance levels: * p < 0.10, ** p < 0.05, *** p < 0.01.

shows that the Average Treatment Effect on the Treated (ATT) was positive and significant (0.028, p < 0.01) across nearest-neighbor, radius and kernel matching methods, indicating that firms with higher AI adoption systematically outperform their matched peers in GIE. This quasi-experimental validation aligns with recent green-innovation studies (for example, Yin et al., 2023).

6.3. Instrumental Variable Estimation

To further address potential endogeneity, two-stage least-squares (2SLS) regressions were estimated using lagged AI adoption as an instrument (Blundell & Bond, 1998). The first-stage F-statistics exceeded the Stock–Yogo critical thresholds, confirming instrument strength. As reported in

Table 14. 2SLS Results for Both Models.

Variables	Model 1: Baseline	Model 2: Moderation (AI × ESG)
AI (Instrumented)	0.024 (0.007) ***	0.019 (0.007) ***
ESG Score	0.012 (0.005) **	0.012 (0.005) **
AI × ESG	—	0.011 (0.004) **
CSIZE	0.019 (0.004) ***	0.019 (0.004) ***
FLEV	–0.031 (0.014) **	–0.031 (0.014) **
INT_COV	–0.006 (0.002) **	–0.006 (0.002) **
ROE, F_AGE	Not significant	Not significant
Industry, Year, Country FE	Included	Included
First-stage F-statistic	28.4	27.1 (AI), 18.4 (AI × ESG)
Durbin–Wu–Hausman (Endogeneity Test)	p < 0.05	p < 0.05

Significance levels: ** p < 0.05, *** p < 0.01.

the IV estimates again revealed a positive and significant coefficient for AI adoption (β = 0.024, p < 0.01) and a reinforcing moderating effect of ESG performance (β = 0.011, p < 0.05). These outcomes are consistent with recent econometric applications in digital-transformation and sustainability research (Li et al., 2024; Feng et al., 2024).

Across Table 12, Table 13 and Table 14, the results converge: AI adoption consistently enhances GIE, and ESG performance strengthens this relationship. Collectively, these robustness tests provide strong confidence in the causal interpretation of the AI–GIE link and reinforce the synergistic role of digital capability and sustainability governance in extending the NRBV.

7. Discussion and Implications

This study deepens understanding of how digital transformation and sustainability strategies converge in emerging economies by demonstrating that AI is a pivotal driver of GIE—but only when supported by strong ESG systems. Evidence from 15 Asian economies shows that AI-enabled firms achieve superior environmental performance and that the magnitude of this effect is strongly conditioned by governance maturity. These findings echo recent work (Dong et al., 2025; Feng et al., 2024; C. Xu & Lin, 2025) underscoring that technology, in isolation, cannot deliver sustainability gains; instead, organizational alignment, transparency, accountability, and ethical oversight determine whether digital tools translate into development-enhancing environmental outcomes.

7.1. Theoretical Contributions

This study advances sustainability and development theory in several ways. First, it extends the NRBV by conceptualizing AI as an intangible but strategically significant environmental resource. Traditional NRBV thinking emphasizes tangible eco-efficient assets; our findings reveal that algorithmic capability, digital analytics, and data-processing power can also serve as green strategic resources when embedded within organizational routines that sense, seize, and transform technological opportunities toward environmental value creation (J. Lin et al., 2024).

Second, the results advance DCT by identifying ESG quality as a governance-based dynamic capability. ESG maturity institutionalizes transparency, ethical oversight, and stakeholder inclusivity, thereby creating the organizational conditions for AI to shift from a cost-reduction tool to a driver of sustainable innovation. In other words, ESG acts as the governance foundation that activates AI’s potential to generate environmental and socio-economic value (Zhong & Song, 2025).

Third, the study helps resolve inconsistencies in the literature by highlighting cross-country heterogeneity. Economies with strong institutional systems—such as Japan, Singapore, and South Korea—exhibit a durable and pronounced AI–GIE relationship, whereas countries with weaker governance show weaker or short-lived impacts. This reinforces the view that digitalization and sustainability transitions are interdependent, shaped by regulatory quality, institutional capacity, and policy coherence. Collectively, the findings enrich theoretical debates on how digitalization and governance interplay to support sustainable growth in emerging economies, especially in relation to SDGs 8, 9, and 16.

7.2. Managerial Implications

For managers, the central insight is that AI and ESG are strategic complements. Adopting AI without strong governance systems may improve operational efficiency but is unlikely to generate sustained environmental performance. Firms should therefore treat ESG not as a regulatory burden but as a strategic capability that enhances technological value.

Recommended managerial actions include: Integrating ESG metrics into AI project evaluation, ensuring that digital investments target carbon reduction, energy optimization, waste minimization, and resource efficiency.

Establishing cross-functional governance committees linking sustainability, data science, compliance, and risk-management teams to oversee responsible AI use.

Investing in workforce capability and digital literacy, as employee competence enhances the ability to deploy AI ethically and in alignment with sustainability priorities.

In emerging Asian markets—where environmental regulation and AI oversight are often uneven—strong internal ESG can serve as a strategic compass guiding firms toward responsible digital transformation. Firms that embed these mechanisms are more likely to attract green financing, meet investor expectations, and convert digitalization into long-term competitive and developmental advantages, directly supporting SDG 8 on decent work and sustainable growth.

7.3. Policy Implications

For policymakers, this research highlights the need for coordinated policy frameworks that integrate digital transformation with ESG-driven institutional reform. Supporting AI adoption is insufficient unless accompanied by governance structures that ensure ethical, transparent, and environmentally aligned deployment.

Recommended policy interventions include:

Embedding ESG disclosure and assurance standards within digital-economy and industrial-upgrading strategies.

Providing capacity-building programs that equip firms—especially SMEs—with the skills to integrate AI and ESG practices.

Prioritizing high-impact sectors such as energy, transportation, manufacturing, and agriculture, where AI can generate substantial environmental gains.

Evidence from emerging-market reform programs (Mijit et al., 2025) suggests that pairing innovation incentives with institutional scaffolding produces stronger and more enduring sustainability outcomes. For rapidly developing Asian economies, a multi-level governance approach—connecting digital governance, sustainability reporting, inclusive financial systems, and clean-technology investment—is essential for enabling a low-carbon, resilient, and socially inclusive development trajectory.

7.4. Limitations and Future Research

While the study employs multiple econometric approaches for robustness, several limitations remain. The AI capability index is based on disclosures and patent data, which may overlook qualitative elements such as algorithm sophistication, ethical safeguards, or proprietary in-house tools. Future research could integrate text analytics, surveys, and AI investment data to refine measurement.

Additionally, ESG performance is treated as a composite index; disaggregating environmental, social, and governance dimensions may reveal differential moderating effects. Although the sample covers diverse Asian emerging economies, generalizability beyond the region may be limited. Comparative studies across Africa, Latin America, and Eastern Europe could broaden external validity. Qualitative research—such as interviews, case studies, or ethnographic approaches—could further illuminate how firms operationalize AI–ESG integration, addressing emerging issues such as AI ethics, data sovereignty, algorithmic bias, and green digital governance.

8. Conclusions

This study demonstrates that AI capability significantly enhances GIE across emerging Asian economies and that this effect is substantially amplified by strong ESG systems. The results show that while digital technologies are essential enablers of environmental progress, their sustainability impact depends critically on governance structures that ensure transparency, accountability, and alignment with long-term social and environmental goals.

By conceptualizing ESG performance as a moderating dynamic capability, the study extends NRBV and DCT, illustrating how digital resources can generate both environmental and socio-economic value when embedded within robust governance frameworks. These insights highlight that technological investments achieve their greatest developmental impact when coupled with responsible management, ethical oversight, and future-oriented sustainability strategies.

For firms, the findings underscore that responsible AI governance is central to sustainable competitiveness. For policymakers, they reveal that digital transformation must be embedded within ESG-based institutional design to produce equitable and resilient development outcomes. Together, the evidence illustrates how the interaction between digital intelligence and governance quality can accelerate sustainable growth, positioning AI not only as an efficiency-enhancing tool but also as a catalyst for environmental resilience and social responsibility across emerging economies.