How Does Participation in AI Standardisation Affect the Sustainable Development of Strategic Emerging Enterprises Under the Background of Uncertainty? Evidence from China

Abstract

Under the background of uncertainty brought by the rapid development of AI, participation in AI standardisation is becoming the key for strategic emerging enterprises (SEEs) to break through and achieve sustainable development. This paper selects listed SEEs from the China Strategic Emerging Industries Composite Index jointly issued by China Securities Index Co., Ltd. and the Shanghai Stock Exchange in 2017 as the initial sample. We collect 3430 observations from 380 companies spanning 2010 to 2023. This paper employs a two-way fixed effects model incorporating enterprise clustering. It thoroughly investigates and empirically tests how participation in AI standardisation affects the sustainable development of SEEs under uncertainty. It is found that participation in AI standardisation in the context of uncertainty has a significant positive effect on the sustainable development of SEEs, and this conclusion still holds after employing instrumental variables, difference-in-difference, and a series of robustness tests. Mechanism tests indicate that two transmission paths exist between participation in AI standardisation and the sustainable development of SEEs under uncertainty: digital technology innovation and the dynamic capabilities in the dimensions of learning and absorption as well as change and reconfiguration. However, the dynamic capabilities in the coordination and integration dimensions do not play a significant mediating role. Heterogeneity analyses indicate that participation in AI standardisation contributes more significantly to the sustainable development of SEEs that are not state-owned, face lower environmental and information uncertainty, and are under higher economic policy uncertainty. The findings enrich the research related to AI standardisation and firm sustainability and provide policy recommendations for the sustainable development of SEEs in the context of uncertainty.

1. Introduction

The rapid development of AI has brought great convenience to people’s daily life. However, it has also brought about problems such as changes in the employment structure, accelerated technological change, and impacts on social ethics and law. These issues have profoundly affected the sustainable development of enterprises, as well as the economy and society. As the main force of key technological innovation in emerging fields [1], strategic emerging enterprises (SEEs) are at the forefront of the AI technology competition. However, their AI endeavours are inherently characterized by technological uncertainty [2], market unpredictability [3], and regulatory ambiguity [4]. Consequently, SEEs face substantial risks to their sustainability, including disruptive technological shifts, unpredictable returns on investment [5], and escalating pressures related to data security and compliance [6]. The “2025 Accenture China Enterprise Digital Transformation Index” research released by Accenture pointed out that 46% of the surveyed enterprises are applying generative AI at scale, but only 9% of the enterprises have achieved significant value transformation through generative AI. Therefore, navigating AI development while ensuring their own sustainable development presents a critical challenge for SEEs. Participation in AI standardisation emerges as a pivotal strategy to address this challenge. Standards serve as a core governance mechanism, fostering industry consensus, guiding technology development pathways, and reducing market transaction costs and compatibility risks [7]. Crucially, standardisation directly mitigates the core uncertainties faced by SEEs: it reduces ecosystem- and enterprise-level uncertainty [8], moderates the pace of technological iteration [9], and strengthens enterprise AI governance capabilities [10]. Consequently, governments increasingly recognize and support enterprise participation in standardisation as a key strategic initiative. However, the rapid development and societal impact of AI technologies can lead to cognitive biases, and is there a need to revisit the scope and objectives of AI standardisation efforts [11]? Continuously advancing technological innovation or rapidly developing standards braking [12]? This all makes the development and implementation of AI standards face the same uncertainty [13], and the impacts of participation in AI standardisation highlight a high degree of heterogeneity. In this context, can participation in AI standardisation promote the sustainable development of SEEs? How can participation in AI standardisation contribute to the sustainable development of SEEs? And under what background can participation in AI standardisation more effectively promote the sustainable development of SEEs? This paper investigates these questions.

Reviewing existing studies, scholars have fully explored how AI affects corporate sustainable development. Research indicates that the adoption and implementation of AI technology can substantially promote corporate sustainable development [14,15,16] and lead to notable improvements in environmental performance. These enhancements are demonstrated through reduced carbon intensity [17,18], decreased pollutant emissions [19], and strengthened green innovation capability [20]. On the one hand, the mechanisms through which AI affects corporate sustainability are summarised as green technological innovation [21], operational optimization [19], risk management [22], and governance and decision-making [23]. On the other hand, studies generally reveal strong heterogeneity in AI’s impact on corporate sustainable development. Specifically, non-state-owned [14], growing [17], less financially constrained [14], and large or old [19] firms respond more significantly to AI. Additionally, heavily polluting industries [20], manufacturing industries [21], and capital-intensive and non-high-technology-intensive industries [14,19] can achieve more significant improvements in environmental performance through AI technology application. In addition, AI is more effective in supporting environmental SDGs than social and economic goals [24]. These differences deeply reflect the numerous uncertainties that companies face in applying AI [25]. However, although existing studies have provided insights into the impact of AI technology itself on corporate sustainability, they have not yet systematically explored the mechanism of engaging in AI standardisation, a forward-looking strategic action, for SEEs to achieve sustainable development in the context of uncertainty.

Therefore, this paper employs micro-enterprise data from China’s strategic emerging industries to explore and empirically test the impact of participation in AI standardisation on the sustainable development of SEEs, the mechanism of its effect, and the key situational factors based on the context of uncertainty. The possible contributions of this paper are as follows: at the theoretical level, this paper expands the research perspective of AI and enterprise sustainable development, reveals the unique mechanism by which participation in AI standardisation affects the sustainable development of SEEs, and deepens the understanding of the heterogeneity of the aftereffects of enterprises’ participation in standardisation; at the practical level, this paper provides a basis for SEEs to formulate AI strategies, a reference for policymakers to optimise the AI standardisation governance framework, and a path for SEEs’ sustainable development under uncertainty.

The rest of the paper is arranged as follows: Section 2 sorts out the relevant theories and puts forward the research hypotheses based on the theoretical analyses. Section 3 introduces the sample data and econometric models. Section 4 presents the results of the empirical analyses. Section 5 contains further analyses of the nature of property rights and the heterogeneity of the context of uncertainty. Section 6 is the discussion of the article. Section 7 presents the conclusions and policy implications.

2. Theory and Hypotheses

2.1. Participation in AI Standardisation and Sustainable Development of SEEs in the Context of Uncertainty

It has been shown that standards play a significant role in reducing technological uncertainty, standardising market access, and reducing transaction costs [26,27]. In the field of AI, there are various technology paths and standards that are not yet unified, and SEEs are facing a highly uncertain technology and market environment. Participating in AI standardisation enables enterprises to anticipate trends in technological evolution and mitigate the risk of technological disruption [28]. It also allows them to influence the content of standards, thereby ensuring the compatibility of their technologies and compliance of their products, which enhances market adaptability and predictability. In addition, the knowledge sharing and network construction in the process of standardisation can also help enterprises to access external resources and enhance their strategic resilience in an uncertain environment [29]. Therefore, participation in AI standardisation can significantly enhance the sustainability of SEEs in the context of uncertainty.

Specifically, in the context of uncertainty, the participation of SEEs in AI standardisation can significantly promote the sustainable development of enterprises through the triple mechanism of reducing the ambiguity of technology paths, mitigating the risk of market fluctuations, and strengthening the adaptability of compliance. First, SEEs can transform dispersed technological resources into a structured capability system by participating in standardisation. Standardisation is essentially a technological governance tool that can coordinate the technological routes of diversified innovation subjects and reduce the lock-in risk caused by the rapid pace of technological change. In addition, by participating in the development of AI standards, SEEs can know the industrial technology route in advance, align their own technology development with the industry consensus, and reduce the resource mismatch caused by the misjudgement of technology direction. Second, the participation of SEEs in AI standardisation can ease market acceptance barriers by reducing transaction costs and enhancing mutual trust in products. Especially in emerging areas, promoting unified technology standards can significantly reduce consumer concerns about AI product compatibility and accelerate enterprise market expansion. At the same time, the signalling mechanism of participating in AI standardisation can reduce information asymmetry [30] and make it easier for enterprises’ AI products to gain market acceptance, thus alleviating the “lemon market” dilemma faced by enterprises [31]. Finally, in terms of regulatory adaptability, participation in AI standardisation can give SEEs the ability to look ahead for compliance: AI ethics and algorithmic governance have gradually become the focus of global regulation, and participation in the formulation of relevant standards in advance enables enterprises to anticipate compliance requirements and embed ethical design principles into the R&D process, which can, to a certain extent, avoid the losses that may be caused by compliance rectification at a later stage [32]. This ability to proactively shape rules transforms enterprises from “passive compliers” to “standard-setters”, significantly enhancing their strategic initiative in the context of regulatory dynamics. In conclusion, participation in AI standardisation is a strategic choice to build order in an uncertain environment, which can effectively promote the sustainable development of SEEs.

Accordingly, the following hypotheses are proposed:

H1. 

Participation in AI standardisation has a significant positive impact on the sustainable development of SEEs in the context of uncertainty.

2.2. The Mediating Role of Digital Technological Innovation

Technology standards are not only a governance tool, but also a catalyst for digital innovation [33]. Digital technology innovation refers to the process and results of enterprises or organisations developing new products, new processes, and new business models based on digital technology [34]. Based on Resource Orchestration Theory (ROT) [35,36], participation in AI standardisation provides SEEs with critical external resource inputs by embedding them within technological networks to access cutting-edge knowledge and technical feedback. Through resource structuring, SEEs effectively integrate fragmented technical knowledge, market intelligence, and collaborative relationships gained from standardisation, establishing a digital infrastructure that supports innovation. Subsequently, via resource bundling, enterprises deeply converge externally acquired AI knowledge and technical feedback with their internal green technology expertise and existing capabilities, catalysing integrated innovations such as AI algorithm optimization, green computing power deployment, and low-carbon product design [37,38]. Ultimately, through resource leveraging, enterprises transform these orchestrated resources into concrete digital technology innovations. Such standardisation-triggered digital technological innovations, rooted in resource orchestration processes, not only directly enhance resource utilization efficiency and carbon emission performance [39] but also optimize the flow and analysis of environmental, production, and energy consumption data by enabling open data exchange environments and AI-powered decision-making capabilities [40]. This significantly strengthens organizational adaptability and competitive advantage in uncertain environments. In summary, participation in AI standardisation can promote the sustainable development of SEEs through digital technology innovation.

Accordingly, the following hypotheses are proposed:

H2. 

Digital technology innovation mediates between participation in AI standardisation and sustainable development of SEEs under uncertainty.

2.3. The Mediating Role of Dynamic Capabilities

In the context of uncertainty, the contribution of participation in AI standardisation to the sustainable development of SEEs also needs to be channelled through environmental adaptation, knowledge transformation, and increased levels of organisational change, which are the dynamic capabilities of firms. On the one hand, Dynamic Capabilities Theory (DCT) emphasizes how enterprises achieve continuous adaptation in uncertain environments through coordination and integration, learning and absorption, and change and reconfiguration [41]. On the other hand, participation in AI standardisation provides SEEs with a platform for cross-organizational learning and resource integration, inherently encompassing these three capability dimensions. Consequently, we conceptualize dynamic capabilities through this tripartite framework. First, coordination and integration capability refer to an organization’s capacity to leverage external opportunities by integrating internal and external resources and reconfiguring organisational structures. This capability enables SEEs to effectively integrate resources and networks in the standard-setting process—where multiple parties have conflicting interests—and enhances the influence and strategic consistency of the standard [28]. By facilitating in-depth dialogues with regulators and competitors, and by forming a consensus on the expectations of ethical and compliance requirements for the technology, the policy uncertainty faced by the enterprises can be significantly reduced. Secondly, learning and absorption capability refers to identifying, assimilating, and transforming knowledge; this capability enables SEEs to continuously absorb external knowledge and internalise it into technical capabilities during the process of AI standardisation. The cutting-edge knowledge gathered in the process of compiling standards can significantly reduce the cognitive barriers of enterprises to the complex technology of AI, thus enhancing the innovation efficiency of enterprises and the ability of green and intelligent transformation [42]. Finally, change and reconfiguration capability refers to an organization’s capacity to proactively restructure resources in response to environmental shifts; this capability enables SEEs to adjust their technology routes and business models according to the evolution of standards and policy changes in a timely manner, and achieve the transformation from passive adaptation to active leadership [43], thus further promoting enterprises to take the lead in completing the organisational structure adapted to AI and win the competitive advantage in the wave of AI development. In summary, participation in AI standardisation can promote the sustainable development of SEEs through the dynamic capabilities of coordination and integration, learning and absorption, and change and reconfiguration.

Accordingly, the following hypotheses are proposed:

H3. 

Dynamic capabilities mediate between participation in AI standardisation and sustainable development of SEEs under uncertainty.

H3a. 

Dynamic capabilities in the coordination and integration dimension mediate between participation in AI standardisation and the sustainable development of SEEs in the context of uncertainty.

H3b. 

Dynamic capabilities in the learning and absorption dimension mediate between participation in AI standardisation and the sustainable development of SEEs in the context of uncertainty.

H3c. 

Dynamic capabilities in the change and reconfiguration dimension mediate between participation in AI standardisation and sustainable development of SEEs in the context of uncertainty.

2.4. Theoretical Framework

The mechanism through which participation in AI standardisation affects the sustainable development of SEEs can be more comprehensively explained through the synergistic lens of ROT and DCT. ROT reveals how standardisation provides enterprises with access to critical resources. Through structuring, bundling, and leveraging processes, these resources are transformed into capabilities that directly drive both digital technology innovation and sustainable development. DCT further explains how the standardisation process itself builds and strengthens higher-order adaptive capacities within highly uncertain AI environments. Thus, ROT manages critical input resources and drives digital technological capabilities. Meanwhile, DCT ensures enterprises can effectively deploy these resources and capabilities amidst turbulence while continuously adapting to change. Together, they form the core transmission chain from participation in standardisation to sustainable development.

Specifically, ROT serves as the foundational starting point, focusing on how enterprises identify, acquire, structure, bundle, and leverage diverse resources from standardisation to provide the essential foundation for value creation. DCT acts as the higher-order enabler and catalyst. It addresses how enterprises manage ROT complexity, accelerate resource-to-capability conversion, and flexibly adjust resource orchestration strategies amid uncertainty. This is achieved through coordination and integration, learning and absorption, and change and reconfiguration capabilities. Without effective resource orchestration, dynamic capabilities lack operational grounding; without strong dynamic capabilities, ROT efficiency and sustainability become significantly compromised in highly volatile AI domains. This synergism creates a virtuous cycle: ROT provides the resource base and context for dynamic capability development, while DCT safeguards and optimizes ROT effectiveness in dynamic environments. Collectively, they propel SEEs toward sustainable development, as illustrated in our theoretical framework (Figure 1).

3. Data and Methodology

3.1. Data Sources

Based on data availability, this paper selects listed enterprises in the China Strategic Emerging Industries Composite Index issued by CSI Index Co. and Shanghai Stock Exchange in 2017 as the initial sample, excludes ST and *ST enterprises, as well as enterprises with missing data and outliers, and ultimately obtains 3430 samples from 380 enterprises in the period of 2010–2023.

Among them, the data related to the standardisation of participating AIs come from China National Knowledge Infrastructure (CNKI)’s General Database of Standard Data, the data on corporate sustainability come from the ESG rating data issued by Shanghai Huazheng Index Information Service Co. Ltd., and the data on corporate finance and other data come from China Stock Market & Accounting Research Database (CSMAR). In addition, in order to avoid the interference of extreme values, this paper shrinks all continuous variables by 1% up and down.

3.2. Variable Measurement

3.2.1. Dependent Variable

Sustainable development of SEEs (Esg). ESG is a common evaluation system to measure the prospect and potential of sustainable development of enterprises [44]; drawing on established research practices [45], the annual average of Huazheng ESG rating scores is used to measure the sustainable development of SEEs.

3.2.2. Independent Variable

Participation in AI standardisation (AI_Standard). Referring to the practice of existing research [28], the number of participations in the development of AI standards was used to measure participation in AI standardisation. AI standards were searched as follows: Referring to the AI lexicon provided by Yao et al. [46], we searched for national and industry standards covering 73 topics (including AI, computer vision, and image recognition). This was carried out by inputting relevant expressions in the professional search box of CNKI’s General Database of Standard Data. The search yielded 413 national standards and 57 industry standards released between 1990 and 2025.

3.2.3. Control Variables

This study employs a set of firm-level control variables: firm size (Size), firm age (Age), return on assets (Roa), management shareholding ratio (Share), management expense ratio (Manfee), cash flow ratio (Cash), proportion of independent directors (Indep), growth (Growth), high and new technology enterprise recognition (High), and nature of property rights (Soe).

3.2.4. Mediating Variables

This paper employs digital technology innovation and dynamic capabilities as mediating variables. Digital technology innovation (DTI) is measured by the number of granted digital patents held by enterprises [47]. Dynamic capabilities are categorised into coordination and integration capability (CIC), learning and absorption capability (LAC), and change and reconfiguration capability (CRC), which are respectively measured using total asset turnover, the proportion of employees with a bachelor’s degree or higher, and the ratio of R&D expenditure [48,49,50].

The variables and their measurements are summarised in

Table 1. Variables and their measurement methods.

Variable Type	Variable Name	Code	Measurement Method
Dependent variable	Sustainable development of SEEs	Esg	Annual average of Huazheng ESG rating score
Independent variable	Participation in AI standardisation	AI_Standard	The number of participations in the development of AI standards
Control variables	Firm size	Size	Ln (total assets)
	Age of the enterprise	Age	Ln (current year − year of establishment + 1)
	Return on assets	Roa	Net profit/average total assets
	Management shareholding ratio	Share	Management shareholding/total share capital
	Management expense ratio	Manfee	Administrative Expenses/Operating Income
	Cash flow ratio	Cash	Net cash flow from operating activities/total assets
	Ratio of independent directors	Indep	Number of Independent Directors/Number of Directors
	Growth	Growth	Current year’s operating income/previous year’s operating income −1
	High and new technology enterprise recognition	High	High and new technology enterprises take the value of 1, otherwise 0
	Nature of property rights	Soe	State-owned enterprises take the value of 1, otherwise 0
Mediating variables	Digital technology innovation	DTI	The number of digital patents
	Coordination and integration capabilities	CIC	Total asset turnover
	Learning and absorption capabilities	LAC	Percentage of employees with bachelor’s degree or above
	Change and reconfiguration capabilities	CRC	Percentage of R&D expenditures

.

3.3. Model Design

Based on the preceding theoretical analysis, this paper constructs a two-way fixed effects model with firm-level clustering to examine how participation in AI standardisation influences the sustainable development of SEEs:

$${\mathrm{E}\mathrm{s}\mathrm{g}}_{it}={\beta }_0+{\beta }_1{\mathrm{A}\mathrm{I}\_\mathrm{S}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{d}\mathrm{a}\mathrm{r}\mathrm{d}}_{it}+\gamma \Sigma {\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}\mathrm{s}}_{it}+\Sigma \mathrm{I}\mathrm{n}\mathrm{d}+\Sigma \mathrm{Y}\mathrm{e}\mathrm{a}\mathrm{r}+{\epsilon }_{it}$$

(1)

where the subscripts i and t denote firms and years, respectively, Esg denotes the sustainable development of SEEs, AI_Standard denotes the number of firms participating in AI standardisation in that year, Controls represents a series of firm-level control variables, Ind denotes industry fixed effects, Year denotes year fixed effects, and ε is a random perturbation term. β₁, the core estimated coefficient of interest in this paper, represents the net effect of SEEs’ participation in AI standardisation on their sustainable development; if β₁ is significantly positive, Hypothesis H1 is verified.

4. Results and Discussion

4.1. Descriptive Statistics Analysis

The results of descriptive statistics of the main variables in this paper are shown in

Table 2. Descriptive statistics.

VARNAME	Obs	Mean	SD	Min	Median	Max
Esg	3430	75.126	4.849	60.980	75.280	85.480
AI_Standard	3430	0.666	3.108	0.000	0.000	23.000
Size	3430	22.348	1.360	19.274	22.284	25.949
Share	3338	8.892	15.487	0.000	0.259	61.874
Roa	3430	0.045	0.058	−0.182	0.040	0.217
Manfee	3430	0.093	0.072	0.009	0.074	0.425
Age	3429	19.968	6.003	7.000	20.000	38.000
Cash	3430	0.052	0.063	−0.115	0.045	0.238
Indep	3430	37.790	5.605	33.330	36.360	57.140
Soe	3215	0.493	0.500	0.000	0.000	1.000
High	3430	0.228	0.420	0.000	0.000	1.000
Growth	3430	0.216	0.493	−0.696	0.132	3.216
DTI	3430	0.104	0.384	0.000	0.000	2.303
CIC	3430	0.554	0.342	0.064	0.505	2.425
LAC	3308	0.389	0.231	0.043	0.345	0.910
CRC	3086	6.382	6.387	0.030	4.620	38.920

: First, the mean Esg of the sample enterprises is 75.126, which is close to the median of 75.280, indicating that the data is symmetrically distributed, and the level of sustainable development of the SEEs is stable as a whole, while the standard deviation of 4.849 indicates that the difference in the ESG scores among enterprises is limited, and the industry homogeneity is strong. Second, the median AI_Standard of the sample enterprises is 0, which is significantly lower than the mean value of 0.666; the standard deviation of 3.108 is 4.67 times higher than the mean value, indicating that the data has an extreme rightward bias; more than half of the enterprises are not involved in standardisation; and the maximum value is 23, indicating that a few SEEs are deeply involved in AI standardisation. Overall, SEEs show the characteristics of ESG performance convergence and AI standardisation stratification.

4.2. Benchmark Regression Results

The baseline regression results examining how participation in AI standardisation affects the sustainability of SEEs are presented in

Table 3. Benchmark regression results.

	(1)	(2)	(3)	(4)
VARIABLES	Esg	Esg	Esg	Esg
AI_Standard	0.248 ***		0.181 ***	0.166 ***
	(0.026)		(0.027)	(0.045)
Size		0.644 ***	0.529 ***	0.740 ***
		(0.076)	(0.077)	(0.154)
Share		0.026 ***	0.022 ***	0.018
		(0.006)	(0.006)	(0.012)
Roa		12.303 ***	11.027 ***	10.391 ***
		(1.725)	(1.724)	(2.732)
Manfee		−0.587	−1.653	0.662
		(1.439)	(1.438)	(2.489)
Age		0.032 **	0.035 **	0.008
		(0.015)	(0.015)	(0.029)
Cash		−1.320	−0.703	−0.557
		(1.526)	(1.518)	(2.071)
Indep		0.101 ***	0.099 ***	0.096 ***
		(0.015)	(0.015)	(0.027)
Soe		0.986 ***	0.881 ***	0.835 **
		(0.195)	(0.195)	(0.382)
High		0.294	0.213	0.537 **
		(0.200)	(0.199)	(0.231)
Growth		−0.619 ***	−0.620 ***	−0.520 ***
		(0.171)	(0.170)	(0.152)
Ind	NO	NO	NO	YES
Year	NO	NO	NO	YES
Observations	3430	3124	3124	3123
R-squared	0.025	0.070	0.083	0.208

Note: (1) **, p < 0.05; ***, p < 0.01. (2) Standard errors in parentheses.

. Column (1) is the model with only the core explanatory variables, column (2) is the model with only the control variables, column (3) is the model with the inclusion of the control variables on top of column (1), and column (4) is the model with the inclusion of two-way fixed effects on top of column (3). The regression results show that the coefficients on the core explanatory variable AI_Standard remain positive and significant at the 1% level, regardless of the inclusion of control variables and fixed effects. This shows that participation in AI standardisation has a significant positive effect on the sustainable development of SEEs, and hypothesis H1 is supported.

4.3. Discussion of Endogeneity Issues

4.3.1. Instrumental Variables Approach

In order to alleviate the endogeneity problem caused by reverse causality, omitted variables, and measurement errors in the regression model, this paper uses instrumental variables for two-stage least squares regression [51].

No convincing instrumental variables have been identified in existing research on participation in AI standardisation. Thus, we refer to prior studies [52] and use the annual mean values of AI_Standard for enterprises in the same industry and province as the instrumental variable (Iv_AI_Standard).

Columns (1) and (2) of

Table 4. 2SLS regression results.

	(1)	(2)
VARIABLES	AI_Standard	Esg
Iv_AI_Standard	0.949 ***
	(0.238)
AI_Standard		0.356 ***
		(0.130)
Kleibergen-Paap rk LM statistic		8.234 ***
		[0.004]
Cragg-Donald Wald F statistic		1875.623 ***
		[16.380]
Controls	YES	YES
Ind	YES	YES
Year	YES	YES
Observations	3123	3123
R-squared		0.079

Note: (1) ***, p < 0.01. (2) Standard errors in parentheses.The figures in square brackets correspond to the p–value and the critical value for the Stock–Yogo weak instrument test at the 10% significance level, respectively.

report the first- and second-stage instrumental variable (IV) regression results, respectively. The estimates in Column (1) indicate that the coefficient of the instrumental variable, Iv_AI_Standard, is positive and significant at the 1% level, supporting the relevance condition. Furthermore, the Kleibergen–Paap rk LM statistic is significant at the 1% level, rejecting the null hypothesis of underidentification. The Cragg–Donald Wald F statistic exceeds the Stock–Yogo weak IV test critical value at the 10% level, suggesting that the instrument is not weak. Together, these diagnostic tests confirm the validity of the instrumental variable and support the assumption of exogeneity. After correcting for endogeneity bias using the IV approach, the results in Column (2) show that the coefficient of AI_Standard remains significantly positive at the 1% level, consistent with the baseline estimates in Table 3.

4.3.2. Difference in Difference Estimation

This paper further mitigates the potential endogeneity problem through the DID method [53]. On 8 July 2017, the State Council issued the “New Generation Artificial Intelligence Development Plan” (hereinafter referred to as the “Plan”), which takes the participation of SEEs in the standardisation of AI as an important support, and proposes that we should speed up the promotion of industry associations and alliances in the segmented application fields of unmanned vehicles, service robots, and other areas to formulate the relevant standards, accelerate the cultivation of a number of leading enterprises in emerging fields such as intelligent robots, intelligent cars, wearable devices, virtual reality, etc., and support AI enterprises to lead or participate in the development of international standards. This is a relatively exogenous shock to the participation of SEEs in AI standardisation, therefore, this paper takes the issuance of the Plan (2017) as the quasi-natural experiment to construct a DID model to further identify the relationship between participation in AI standardisation and the sustainable development of SEEs.

Specifically, firms are categorised into treatment and control groups based on their participation in AI standardisation. The variable Treat is set to 1 for the treatment group and 0 for the control group. For the time dimension, the variable Post is assigned a value of 0 for the year preceding the announcement of the Plan, and 1 for the year of the announcement and subsequent years. The impact of the issuance of the Plan on the participation of SEEs in AI standardisation is reported in Column (1) of

Table 5. DID regression results.

	(1)	(2)
VARIABLES	AI_Standard	Esg
Treat×Post	5.890 ***	2.254 ***
	(1.040)	(0.617)
Controls	YES	YES
Ind	YES	YES
Year	YES	YES
Observations	3123	3123
R-squared	0.310	0.208

Note: (1) ***, p < 0.01. (2) Standard errors in parentheses.

, with an estimated coefficient of 5.890 and is significant at the 1% level. Column (2) of Table 5 reports the impact of the issuance of the Plan on the sustainability of SEEs, with an estimated coefficient of 2.254, which is significant at the 1% level. The above results indicate that the implementation of the Plan does promote the participation of SEEs in AI standardisation and further positively affects the sustainability of SEEs.

To evaluate the parallel trends assumption of the difference-in-differences approach, interaction terms between the treatment indicator Treat and year dummies are included in the regression model. This specification tests whether pre-existing differences exist in Esg between the two groups, which is essential for unbiased DID estimation [54]. As shown in Figure 2, the coefficients on the interaction terms for the pre-treatment periods are statistically indistinguishable from zero, indicating no significant divergence in sustainable development between the treatment and control groups prior to the Plan’s introduction. These findings support the validity of the parallel trends assumption.

4.4. Robustness Tests

4.4.1. Independent Variables Lagged One Period

According to the Measures for the Administration of National Standards, national standards have to go through a series of work such as drafting, publicity, review, etc., from project to release, which has a certain lag; therefore, on the basis of adopting ESG ratings in the current period, this paper further adopts the core explanatory variables (L_AI_Standard) lagged by one period for regression. Column (1) of

Table 6. Robustness test results.

	(1)	(2)	(3)	(4)
VARIABLES	Esg	Bloombergesg	Esg_r	Esg
AI_Standard		0.103 ***	0.075 ***	0.163 ***
		(0.037)	(0.012)	(0.048)
L_AI_Standard	0.186 ***
	(0.047)
Controls	YES	YES	YES	YES
Ind	YES	YES	YES	YES
Year	YES	YES	YES	YES
Observations	2540	3123	3124	2536
R-squared	0.216	0.633		0.204
Pseudo R-squared			0.029

Note: (1) ***, p < 0.01. (2) Standard errors in parentheses.

reports the estimation results after lagging the independent variables by one period. The coefficient on L_AI_Standard remains positive and statistically significant at the 1% level, confirming the robustness of the findings.

4.4.2. Replacing the Dependent Variable Measures

In addition to Huazheng ESG rating, some scholars choose Pengbo ESG Composite Score Index to measure corporate sustainability [55]. The Pengbo ESG rating system can provide a comprehensive and objective assessment of corporate sustainability performance through algorithms and models by leveraging multiple sources of information such as financial reports, environmental disclosures, news, and government documents. Therefore, this paper further adopts the Pengbo ESG composite score index to measure corporate sustainability (Bloombergesg). Column (2) of Table 6 reports the estimation results after replacing the dependent variables. The coefficient on AI_Standard remains positive and statistically significant at the 1% level, confirming the robustness of the findings.

4.4.3. Replacing the Regression Model

In this paper, the ESG rating system of Huazheng is further coded, and the nine grades of AAA, AA…C are assigned as 1–9 [56] from low to high as the proxy variable of corporate sustainability (Esg_r), and the values of ESG ratings are discrete at this point, so this paper further replaces the model with the ordered Logit model to re-regress. Column (3) of Table 6 reports the estimation results after replacing the regression model. The coefficient on AI_Standard remains positive and statistically significant at the 1% level, confirming the robustness of the findings.

4.4.4. Shortening the Time Window

During the COVID-19 pandemic, the business activities and cooperation exchanges of SEEs were affected, and the sustainable development level of enterprises was greatly impacted. Therefore, this paper excludes the data of 2020–2022, which is affected by the COVID-19 pandemic. Column (4) of Table 6 reports the estimation results after shortening the time window. The coefficient on AI_Standard remains positive and statistically significant at the 1% level, confirming the robustness of the findings.

4.5. Analysis of Intermediary Mechanism

According to the previous theoretical analysis, participation in AI standardisation in the context of uncertainty can positively affect the sustainable development of SEEs through digital technology innovation and dynamic capabilities. Referring to the existing research [57], it has been shown that the mechanism variables digital technology innovation and dynamic capabilities can significantly and positively affect the sustainable development of enterprises [58,59], so this paper focuses on the causal effect of participation in AI standardisation on the mechanism variables under the background of uncertainty, and the empirical model is set as follows:

$${\mathrm{Z}}_{it}={\beta }_0+{\beta }_2{\mathrm{A}\mathrm{I}\_\mathrm{S}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{d}\mathrm{a}\mathrm{r}\mathrm{d}}_{it}+\gamma \Sigma {\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}\mathrm{s}}_{it}+\Sigma \mathrm{I}\mathrm{n}\mathrm{d}+\Sigma \mathrm{Y}\mathrm{e}\mathrm{a}\mathrm{r}+{\epsilon }_{it}$$

(2)

where Z represents the mediating variables, respectively, while all other variables retain their definitions from the benchmark regression.

4.5.1. Digital Technology Innovation Mechanism

Column (1) of

Table 7. Mechanism test results.

	(1)	(2)	(3)	(4)
VARIABLES	DTI	CIC	LAC	CRC
AI_Standard	0.048 ***	0.001	0.013 ***	0.127 **
	(0.005)	(0.005)	(0.002)	(0.059)
Controls	YES	YES	YES	YES
Ind	YES	YES	YES	YES
Year	YES	YES	YES	YES
Observations	3123	3123	3001	2806
R-squared	0.282	0.533	0.539	0.573

Note: (1) **, p < 0.05; ***, p < 0.01. (2) Standard errors in parentheses.

presents the regression results examining the impact of AI standardisation participation on the digital technology innovation of SEEs under conditions of uncertainty. The coefficient on the core explanatory variable, AI_Standard, is 0.048 and significant at the 1% level, indicating a strong positive effect. As theorised previously, digital technology innovation is a key determinant of sustainable development in SEEs. These results therefore suggest that digital innovation serves as a mediating mechanism between participation in AI standardisation and sustainable development in SEEs, providing support for Hypothesis H2.

4.5.2. Dynamic Capabilities Mechanism

Columns (2) to (4) of Table 7 present the estimated effects of AI standardisation participation on the dynamic capabilities of SEEs under uncertainty. The coefficients on AI_Standard are 0.001, 0.013, and 0.127, respectively. Statistical significance is observed only for learning and absorptive capacity (at the 1% level) and change reconfiguration capacity (at the 5% level). Thus, participation exerts a significant positive influence on these two capabilities, but no significant effect on coordination and integration capacity. A possible explanation is that participation in AI standardisation primarily acts at the technical level, providing clear pathways and efficiency gains for technology-related learning and change in enterprises. In contrast, factors influencing coordination and integration capabilities—such as leadership, communication, and organisational culture—are less affected by standardisation. Additionally, the unique organisational structures and resource constraints of SEEs limit the extent to which participation in AI standardisation enhances their coordination capabilities. As outlined in the theoretical framework, dynamic capabilities facilitate the sustainable development of SEEs by enabling the ongoing perception, integration, and reconfiguration of resources in response to environmental shifts. Accordingly, this study concludes that dynamic capabilities—specifically learning absorption and change reconfiguration—mediate the relationship between participation in AI standardisation and sustainable development under uncertain conditions, thus providing support for hypotheses H3b and H3c.

5. Heterogeneity Analysis

5.1. Nature of Enterprise Property Rights

Table 8. Results of the analysis of heterogeneity in the nature of firms’ property rights.

	(1)	(2)
	State-Owned Enterprises	Non-State-Owned Enterprises
VARIABLES	Esg	Esg
AI_Standard	0.103 *	0.246 ***
	(0.060)	(0.057)
Controls	YES	YES
Ind	YES	YES
Year	YES	YES
Observations	1505	1616
R-squared	0.282	0.226

Note: (1) *, p < 0.1; ***, p < 0.01. (2) Standard errors in parentheses.

presents the results of the heterogeneity analysis based on firms’ ownership type. The coefficient on AI_Standard is 0.103 (significant at the 10% level) for state-owned SEEs, and 0.246 (significant at the 1% level) for non-state-owned SEEs. These results imply that the positive effect of AI standardisation participation on sustainable development is more pronounced among non-state-owned SEEs than in state-owned counterparts. The possible reason is that state-owned enterprises usually have inherent advantages in resource acquisition, market position, and risk resistance, and these inherent advantages support their sustainable development to a certain extent, which makes the marginal effect of participation in AI standardisation relatively limited; while non-state-owned enterprises usually face more severe market competition and resource constraints, their decision-making mechanism is more flexible, market response is more agile, and their demand for efficiency improvement and competitive advantage construction is more urgent; therefore, active participation in AI standardisation can bring more direct benefits to them, thus driving their sustainable development more effectively under the background of uncertainty.

5.2. Uncertainty Context

In this paper, we categorise the uncertainty context into micro-level firm environment and information uncertainty and macro-level economic policy uncertainty. Firm environment uncertainty is measured by the standard deviation of a firm’s sales revenue over the past five years, excluding the normal growth component and adjusted for the industry [60]; firm information uncertainty is measured by the five-year standard deviation of a firm’s cash flow volatility, i.e., net cash flow from operating activities divided by total assets [61]; and economic policy uncertainty is measured by the Economic Policy Uncertainty Index (EPU) [62].

Table 9. Results of uncertainty context heterogeneity analysis.

	Enterprise Environment Uncertainty		Enterprise Information Uncertainty		Economic Policy Uncertainty
VARIABLES	Esg	Esg	Esg	Esg	Esg	Esg
AI_Standard	0.212 *	0.194 ***	0.104	0.189 ***	0.253 ***	0.075
	(0.107)	(0.040)	(0.122)	(0.036)	(0.041)	(0.059)
Controls	YES	YES	YES	YES	YES	YES
Ind	YES	YES	YES	YES	YES	YES
Year	YES	YES	YES	YES	YES	YES
Observations	166	1285	657	2202	1620	1502
R-squared	0.384	0.266	0.256	0.256	0.251	0.191

Note: (1) *, p < 0.1; ***, p < 0.01. (2) Standard errors in parentheses.

presents the results of the heterogeneity analysis across different uncertainty contexts. Under high firm environmental uncertainty, the estimated coefficient for AI_Standard is 0.212, significant at the 10% level, compared to 0.194 (significant at 1%) in low environmental uncertainty settings. For firm information uncertainty, the coefficient is insignificant in the high uncertainty group, but 0.189 (significant at 1%) in the low uncertainty group. Similarly, under high economic policy uncertainty, a significant coefficient of 0.253 (1% level) is observed; whereas, the relationship is not statistically significant under low economic policy uncertainty. It indicates that the positive effect of participation in AI standardisation on the sustainability of SEEs is more pronounced under low firm environmental uncertainty, low firm information uncertainty, and high economic policy uncertainty. The possible reasons are as follows. First, under low corporate environmental uncertainty, a stable market environment provides SEEs with a predictable basis for implementing AI standardisation strategies and sufficient resources. This enables them to systematically integrate the technological dividends from participation in AI standardisation, thereby stably enhancing their sustainable development performance. Second, under low corporate information uncertainty, clear and stable internal operational information is a prerequisite for SEEs to effectively plan, assimilate, and apply complex AI standards. In contrast, high information uncertainty often reflects chaotic internal management or high risks, which seriously hinders the transformation of external standardisation knowledge into internal operational effectiveness. Third, under high economic policy uncertainty, traditional development paths fail, compliance risks rise, and future rules are uncertain. At this point, active participation in AI standardisation becomes a key strategic tool for SEEs to proactively shape the future technological ecology, seize rule-making power, and reduce compliance risks. However, in periods of low economic policy uncertainty, the stable policy environment makes enterprises more likely to form path dependence, reducing the strategic urgency and marginal value of participating in AI standardisation, thus weakening its impact on sustainable development.

6. Discussion

The impact of AI technology on corporate sustainable development has been confirmed by many scholars. However, against a backdrop of uncertainty, there is limited research that systematically investigates the effect of participation in AI standardisation on the sustainable development of strategic emerging enterprises (SEEs). Studies on its mechanisms and heterogeneity are particularly scarce. This paper systematically analyses how participation in AI standardisation affects the sustainable development of SEEs, considering different property rights and uncertainty contexts. It further introduces two mediating variables—digital technology innovation and dynamic capabilities—and constructs a two-way fixed effects model with firm clustering. This model is used to explore and empirically test the effect and mechanisms of AI standardisation participation on SEEs’ sustainable development.

First, the study finds that the positive impact of participating in AI standardisation on sustainable development is more significant for non-state-owned SEEs. From a contextual perspective, non-state-owned enterprises face greater pressure in market competition. Their survival and development rely more heavily on their own market adaptability and technological innovation capabilities. State-owned enterprises often receive more policy support and resource advantages, making them less dependent on standardisation. Participation in AI standardisation can provide non-state-owned enterprises with clearer technical direction and industry rules. This helps them to develop in a standardised manner within the competitive landscape, reducing the costs and risks associated with technological disunity. Consequently, it more significantly promotes their sustainable development. Therefore, the government should increase support for non-state-owned SEEs to participate in AI standardisation, for instance, by providing funding subsidies and establishing communication platforms. Non-state-owned SEEs should proactively engage in the development of AI standards, leveraging standardisation to enhance their competitiveness.

Second, the research finds that the positive impact of participating in AI standardisation on the sustainable development of SEEs is more pronounced under conditions of low firm environmental uncertainty, low firm information uncertainty, and high economic policy uncertainty. In contexts of low environmental and information uncertainty, firms operate in a relatively stable market environment with reliable information access. Under such conditions, participation in AI standardisation enables firms to plan their development within a stable framework. This allows them to fully leverage the advantages of standardisation while reducing interference from environmental and informational chaos. Under high economic policy uncertainty, frequent policy changes introduce numerous unknown risks for business development. AI standardisation, as a unified industry norm, can provide firms with a relatively stable basis for development, mitigating the impact of policy fluctuations. Thus, the government should strive to maintain a stable environment and information context for business development. Simultaneously, during economic policy adjustments, it should enhance communication with enterprises to reduce the negative impacts of policy uncertainty. Enterprises, facing different uncertainties, should rationally assess their own situation and actively participate in AI standardisation to address challenges arising from various uncertainties.

Third, the study finds that participating in AI standardisation can positively impact the digital technology innovation of SEEs, as well as their dynamic capabilities in the dimensions of learning and absorption, and change and reconfiguration. The AI standardisation process aggregates advanced technology and experience within the industry, providing a platform for learning and exchange for enterprises. By participating in standardisation, firms gain exposure to cutting-edge industry technologies and concepts, which fosters their own digital technology innovation. Furthermore, to better participate in the formulation and implementation of standards, firms must continuously acquire new knowledge, absorb new skills, and undergo change and reconfiguration according to standardisation requirements. This inherently enhances their learning absorption and change reconfiguration capabilities. Therefore, the government should promote the establishment of a robust AI standardisation system and encourage technological exchange and collaboration between enterprises. Strategic emerging enterprises should seize the opportunity to participate in AI standardisation, actively learn advanced technologies and experiences, and continually improve their digital technology innovation capabilities and dynamic capabilities to achieve sustainable development.

This paper also has some limitations. First, there are some limitations in sample selection. This paper takes listed enterprises in China’s strategic emerging industries as the research sample, although it can reflect the mainstream characteristics of the industry, but there are differences in resource endowment, governance structure, and other aspects between listed and non-listed enterprises, which may lead to the limitations of the universality of the research conclusions, and future research could expand the sample scope to enhance the universality of the conclusions. Second, the measurement dimension of uncertainty can be further expanded. This paper mainly focuses on enterprise environmental uncertainty, information uncertainty, and economic policy uncertainty, which may not be able to comprehensively capture the multifaceted characteristics of the uncertainty context. In the future, more types of uncertainty indicators can be introduced, or combined with longitudinal case studies, to track the dynamic process of enterprises’ participation in AI standardisation under different uncertainty situations, so as to provide more precise theoretical guidance for SEEs in coping with uncertainty and achieving sustainable development.

Building upon this research, future efforts could further explore the differential impact of AI standardisation on SEEs’ sustainable development across various national, regional, and industrial contexts. It would be valuable to delve deeper into the underlying reasons for the insignificant role of dynamic capabilities in the coordination and integration dimension. Researchers could also examine the synergistic effects between participating in AI standardisation and other industrial policies, such as digital transformation and green development initiatives. Simultaneously, sustained attention should be given to its ongoing influence on SEEs’ long-term technological accumulation and market positioning. These research endeavours may provide more specific references for SEEs to optimise their standardisation participation strategies in the artificial intelligence era. Such insights would help policymakers introduce more targeted support measures, promote closer alignment between AI standardisation and SEEs’ development needs, and consequently facilitate the overall upgrading of strategic emerging industries.

7. Conclusions and Policy Implications

7.1. Conclusions

This study examines the impact of participation in AI standardisation on the sustainable development of SEEs within uncertain contexts, using a sample of listed firms in China’s strategic emerging industries. The main findings are as follows:

(1)

Benchmark regression results suggest that AI standardisation participation exerts a significant positive effect on the sustainable development of SEEs under uncertainty. This finding remains robust after addressing endogeneity through instrumental variable and difference-in-differences approaches, along with a series of supplementary robustness checks.

(2)

Mechanism analysis reveals that participation in AI standardisation enhances sustainable development through two primary channels: digital technology innovation and dynamic capabilities—specifically, learning absorption and change reconfiguration. In contrast, the coordination and integration dimension of dynamic capabilities does not exhibit a significant mediating effect.

(3)

Heterogeneity analysis indicates that the positive effect of AI standardisation involvement is more pronounced among non-state-owned SEEs compared to state-owned enterprises. Furthermore, when disaggregating uncertainty contexts, the beneficial impact is stronger under conditions of low firm environmental uncertainty, low information uncertainty, and high economic policy uncertainty.

7.2. Policy Implications

Based on the above conclusions, this paper provides the following policy recommendations for the sustainable development of SEEs under the background of uncertainty:

(1)

Strengthen the top-level design of the AI standardisation system to create a favourable institutional environment for SEEs to participate in standardisation activities. Promote the development of AI standard systems by field. For key technical fields (e.g., artificial intelligence chips and algorithm frameworks), standard-setting working groups should be established, led by industry associations with enterprise participation. Quarterly technical seminars should be held to ensure that standards are updated in line with industrial practices. Meanwhile, set up a special fund for standardisation participation. Provide subsidies as a percentage of their standardisation R&D investment to non-state-owned enterprises with relatively low annual revenue. Rely on university artificial intelligence laboratories to build national-level standardisation practice platforms, offering free testing and certification services to enterprises to lower technical thresholds.

(2)

SEEs should take the initiative to combine participation in AI standardisation with their own digital technology innovation and dynamic capability cultivation to build an endogenous driving mechanism for sustainable development. In terms of digital technology innovation, they can set up special R&D projects around AI standard technical parameters. For example, in the field of intelligent manufacturing, address key problems in data transmission encryption technology in accordance with the requirements of equipment interconnection protocols. At the same time, ensure the proportion of relevant R&D funds in the total annual R&D investment. In terms of dynamic capability cultivation, focus on the dimensions of learning and absorption, and change and reconfiguration: establish a standardisation learning mechanism, organise technical personnel to participate in relevant seminars and training, set up a transformation team led by senior management, and optimize business processes and organizational structures every six months in light of problems in standard implementation.

(3)

Implement differentiated and precise policies to help SEEs cope with the uncertain environment. For non-state-owned enterprises, it is suggested that their participation in AI standardisation be included as a bonus item in the listing review of the Science and Technology Innovation Board. Regions with high economic policy uncertainty—such as economically active areas like the Beijing–Tianjin–Hebei region and the Yangtze River Delta—require timely support. After policies are issued, targeted interpretations should be provided to local enterprises. Among traditional manufacturing transformation enterprises with low environmental uncertainty, implement “standard-technology-product” transformation pilots and provide special funding support to pilot enterprises. For industries with low information uncertainty such as medical AI, regulatory authorities should take the lead in establishing industry databases to help enterprises accurately grasp market demand.