The Impact of Computing Infrastructure Construction on Innovation in Manufacturing Enterprises: Evidence from a Quasi-Natural Experiment Based on the Establishment of China’s National Supercomputing Centers

Abstract

This study examines the establishment of China’s national supercomputing centers as an exogenous policy shock. Utilizing data from Chinese manufacturing enterprises listed between 2003 and 2023, it applies a multi-period difference-in-differences (DID) model to assess the impact of computing infrastructure on innovation within Chinese manufacturing enterprises. Results indicate that computing infrastructure significantly enhances manufacturing innovation, a finding that is robust across various tests. This effect is positively moderated by the internal R&D investment of enterprises and the external market share. Heterogeneity analysis reveals that the enhancement effect of computing infrastructure on innovation is more pronounced in non-state-owned enterprises, those located in the eastern region, and those with low ownership concentration. Furthermore, computing infrastructure not only boosts the quantity of innovation but also enhances its quality. This paper offers micro-level evidence for emerging countries to advance sustainable development, transformation, and upgrading of the manufacturing sector through computing infrastructure.

1. Introduction

The rapid advancement of artificial intelligence has triggered a surge in global demand for computing power. The escalating volumes of data and increasingly complex algorithms necessitate enhanced large-scale computing capabilities. As a General-Purpose Technology (GPT), computing power is integral to sustainable growth, emerging as a crucial element for nations seeking competitive advantages and industrial innovation. Consequently, computing power and its associated components have garnered significant attention from both the international community and academia [1,2].

The predictive reliability of Moore’s Law in forecasting computing power growth has diminished [3], while the role of computing power as a catalyst for economic output has underscored its importance as a strategic production asset. Computing power is now pivotal in fostering innovation-driven development and economic expansion. As the backbone of China’s economy, the manufacturing sector faces the challenge of overcoming the “low-end lock-in” [4], amid global value chain restructuring and escalating technological competition. It is imperative for China’s manufacturing industry to advance in innovation-driven and sustainable development and to enhance its standing in the global innovation network. However, a significant gap remains between China’s manufacturing innovation levels and those of leading international manufacturers, primarily due to deficiencies in core technologies and the need for improved research and development (R&D) quality [5,6]. Thus, evaluating the impact of computing infrastructure on manufacturing enterprise innovation is both an immediate practical concern and a theoretical necessity.

Computing power is pivotal for countries aiming to lead in development. To secure strategic advantages, developed nations and regions globally are intensifying efforts to enhance computing infrastructure, supporting sustainable growth and industrial transformation in areas like artificial intelligence, industrial software, and complex simulations [7,8,9]. Emerging economies are similarly accelerating their computing infrastructure development. China, for instance, has introduced the “Action Plan for High-quality Development of Computing Infrastructure,” which forms the policy backdrop for this study.

While major global regions are rapidly developing computing infrastructure, there is scant literature on its impact on manufacturing enterprise innovation. Existing research primarily addresses infrastructure’s influence on enterprise innovation through traditional and new infrastructure lenses [10]. Traditional infrastructure, such as railways [11] and highways [12], fosters innovation by reducing costs and enhancing factor flow and knowledge diffusion [13,14,15]. However, these mechanisms fall short in elucidating the R&D and innovation models propelled by data and algorithms. In the digital economy’s computing era, traditional infrastructure inadequately meets enterprises’ growing data and computing needs, prompting the need for accelerated new infrastructure development. China categorizes new infrastructure into information, integrated, and innovation infrastructure [16].

Information infrastructure encompasses communication networks, emerging digital technologies, and computing infrastructure. Studies indicate that communication infrastructures like 5G, industrial Internet, and broadband networks [17,18,19], along with technological infrastructures such as artificial intelligence, cloud computing, and blockchain [20,21,22], enhance enterprise innovation. These infrastructures primarily reduce R&D costs [23], ease financing constraints [24,25], and facilitate knowledge spillover [26]. Unlike communication and digital technology infrastructures, computing infrastructure directly supports complex model training and large-scale data processing via high-performance computing, fundamentally altering enterprise R&D and innovation boundaries. Nonetheless, the specific impact of computing infrastructure on enterprise innovation remains underexplored.

Converged infrastructure integrates technologies like the Internet, big data, and artificial intelligence to modernize traditional systems, such as intelligent transportation and logistics [27] and energy infrastructure [28]. Studies indicate that it enhances enterprise logistics efficiency, fosters industrial and supply chain collaboration, and indirectly supports innovation [29]. In contrast, computing infrastructure operates more directly. As a strategic infrastructure with general-purpose technology attributes, it is embedded in enterprises’ R&D and knowledge creation through extensive computing and algorithmic support, differing functionally from converged infrastructure.

Innovation infrastructure encompasses public-welfare systems that bolster scientific research and technological advancement, including major scientific and technological facilities [30] and science and education infrastructure [31]. Studies indicate that these facilities enhance enterprises’ technical support and innovation by strengthening industry-university-research collaborations and scientific research networks [32]. Conversely, computing infrastructure transcends merely providing external research conditions. It is directly integrated into enterprises’ R&D processes and knowledge creation stages via extensive computing and algorithm environments, serving as a core driver for enhancing innovation quality and efficiency. This mechanism contrasts with the “external support” function of traditional innovation infrastructure.

Computing infrastructure exhibits characteristics of general-purpose technology (GPT), policy exogeneity, and regional variation. Yet, its innovation impact has not been systematically explored within a unified framework. Current research reveals three main gaps: Theoretically, computing infrastructure is not integrated into the GPT framework, nor are theories like absorptive capacity and the resource-based view employed to elucidate how manufacturing enterprises convert external computing resources into innovation. Empirically, there is a lack of analysis using the exogenous shock of national supercomputing centers to assess the effects of computing infrastructure on innovation quantity, quality, and enterprise heterogeneity. Mechanistically, the influence of factors such as enterprises’ R&D investment and market share on the relationship between computing infrastructure and innovation remains insufficiently examined. This paper addresses these deficiencies by constructing a unified analytical framework.

This paper develops a theoretical framework for leveraging computing infrastructure to enhance innovation in manufacturing enterprises, drawing on General-Purpose Technology (GPT), Absorptive Capacity Theory, and the Resource-Based View (RBV). GPT theory underscores the influence of the ubiquity, continuous advancement, and broad complementarity of computing infrastructure on fostering enterprise innovation. Absorptive Capacity Theory emphasizes the importance of R&D investment in enabling enterprises to identify, assimilate, and utilize external computing resources. Meanwhile, RBV illustrates how market share, as an indicator of scarce resources and competitive advantage, influences enterprises’ capacity to harness computing resources for innovation. This study examines data from Chinese listed manufacturing enterprises between 2003 and 2023, using the establishment of national supercomputing centers as a quasi-natural experiment. Employing a multi-period difference-in-differences (DID) approach, it assesses the impact of computing infrastructure on manufacturing innovation. The findings are: First, the development of computing infrastructure, exemplified by national supercomputing centers, significantly enhances manufacturing innovation, a conclusion supported by robustness tests. Second, both internal R&D investment and external market share positively influence the innovation effects of computing infrastructure on these enterprises. Third, the impact of computing infrastructure on innovation is more pronounced in non-state-owned manufacturing enterprises, those in the eastern region, and those with low equity concentration. Lastly, computing infrastructure not only increases the scale of innovation but also enhances its quality.

This paper offers several key contributions: Firstly, it integrates computing infrastructure into General-Purpose Technology (GPT) theory, highlighting its innovation effects through general applicability, continuous improvement, and extensive complementarity. Additionally, it applies absorptive capacity theory and the resource-based view to examine how enterprises’ R&D investments and market share moderate the impact of computing infrastructure on manufacturing innovation. This approach broadens the theoretical scope of information infrastructure research and provides a systematic framework for analyzing the relationship between new production factors and enterprise innovation in the digital economy era.

This paper employs the exogenous policy shock of establishing China’s national supercomputing centers to construct a multi-period difference-in-differences (DID) model, examining the impact of computing infrastructure on manufacturing innovation. Unlike most studies that use high-speed railways, the Broadband China initiative, or 5G policies as quasi-natural experiments, this research focuses on computing infrastructure, exemplified by national supercomputing centers. Unlike traditional or other digital infrastructure, computing infrastructure is both strategic and systematic. As a general-purpose technology (GPT), it spans industries and complements data, algorithms, and organizational processes, reshaping enterprise R&D models and providing novel insights into infrastructure’s influence on innovation.

This study empirically examines the innovation effects of computing infrastructure by assessing both innovation quantity and quality. Unlike previous research that often relies on patent counts, potentially neglecting the issue of “high quantity but low quality,” this paper emphasizes a dual focus. By systematically evaluating the utility and impact of innovation outcomes, it offers a more comprehensive understanding of computing power’s effects. This approach addresses the limitation of equating innovation with mere scale expansion and meets the manufacturing industry’s need to shift from growth to high-quality development. It also provides new micro-evidence on how computing fosters long-term enterprise competitiveness through high-quality achievements.

The paper is structured as follows: Section 2 reviews the policy background of computing infrastructure, exemplified by National Supercomputing Centers, and theoretically examines its impact on innovation in Chinese manufacturing enterprises. Section 3 outlines the data, variables, and empirical models. Section 4 presents the empirical results, including analyses of the moderating mechanism, robustness, and heterogeneity tests; Section 5 provides a further examination of how computing infrastructure construction influences the quality of innovation in manufacturing enterprises. Section 6 presents the discussion, and Section 7 provides the conclusions.

2. Policy Background and Theoretical Analysis

2.1. Policy Background Related to the National Supercomputing Centers in China

Computing power serves society primarily through computing infrastructure, a modern information framework that combines computational capabilities, network bandwidth, and data storage. This infrastructure enables centralized computing, storage, transmission, and application of information, characterized by its pervasive, intelligent, agile, secure, reliable, and environmentally sustainable attributes. Recently, China has actively advanced the construction of computing infrastructure, notably through national supercomputing centers. Policy evolution in China’s supercomputing sector has shifted from merely establishing these centers to enhancing coordinated computing construction, culminating in a tiered infrastructure approach.

The National Supercomputing Centers in China, endorsed by the Ministry of Science and Technology, serve as vital national strategic information infrastructure and platforms for scientific and technological innovation. Initiated with the establishment of the inaugural center in Tianjin in 2009, these centers have significantly bolstered scientific and technological advancements, societal welfare, and the digital economy’s growth in China.

2.2. Theoretical Analysis and Research Hypotheses

2.2.1. Computing Infrastructure Construction Can Promote Innovation in Manufacturing Enterprises

To escape the constraints of low-end production and ascend both ends of the smile curve, China’s manufacturing sector must urgently pursue innovation-driven transformation and upgrading. In this context, computing infrastructure serves as a crucial general-purpose technology (GPT) in the digital economy era. As per GPT theory, such technologies permeate various industries, systematically advancing technological progress and innovation due to their broad applicability, continuous enhancement, and extensive complementarity [33]. Computing infrastructure exemplifies these attributes, fostering innovation within manufacturing enterprises.

The universal applicability of computing infrastructure renders it a ubiquitous innovation element for the manufacturing industry. Supercomputing platforms and digital R&D platforms find broad application in product design, process enhancement, and optimization, enabling enterprises to devise distributed design solutions and real-time R&D path optimization, thereby expediting research and development efforts [34].

Continuous improvement involves the ongoing enhancement of computing, facilitating enterprises to perform virtual experiments and model tests using high-performance simulation platforms. This diminishes reliance on physical samples and traditional experiments, consequently lowering R&D expenses [35].

Extensive complementarity plays a crucial role in enabling computing infrastructure to synergize with diverse components like data, algorithms, and organizational management, thereby generating a compounding effect. For instance, leveraging a digital twin environment empowers enterprises to continuously enhance computational models, pinpoint and mitigate risks proactively, and bolster innovation certainty.

This paper posits Hypothesis 1: the construction of computing infrastructure can boost innovation in manufacturing enterprises.

2.2.2. Enterprise R&D Investment Positively Moderates the Relationship Between Computing Infrastructure Construction and Innovation in Manufacturing Enterprises

Computing infrastructure supports manufacturing enterprises’ innovation by providing external computational resources. An enterprise’s ability to leverage these resources is linked to its internal R&D investment. As per absorptive capacity theory [36], an enterprise’s capability to identify, assimilate, and apply external knowledge and resources hinges on its internal R&D accumulation and knowledge base. Greater R&D investment enhances an enterprise’s absorptive capacity, enabling more effective utilization of external computing and transformation into innovative outcomes.

Research and development (R&D) investment plays a crucial role in augmenting an enterprise’s knowledge reservoir and enhancing its learning capacity. This, in turn, facilitates the swift identification and comprehension of novel knowledge and tools made available through computing infrastructure. Enterprises that allocate substantial resources to R&D typically boast highly skilled R&D teams and comprehensive knowledge frameworks, thereby bolstering their proficiency in deciphering and assimilating new information [37]. As a result, they can adeptly leverage computing platforms to bolster product design, streamline processes, and enhance overall operational efficiency. Conversely, organizations with inadequate R&D funding struggle to amass knowledge effectively, impeding their ability to seamlessly integrate computing capabilities into their R&D workflows and consequently hampering innovation output.

Furthermore, R&D investment significantly improves an enterprise’s resource allocation capabilities [38]. Manufacturing enterprises with substantial R&D investment can develop an extensive system of data, algorithms, and technology, seamlessly integrating with computing power infrastructure to maximize resource mobilization for R&D activities. Conversely, enterprises with minimal R&D investment face resource shortages and lack essential hardware and organizational structures, hindering the effective use of computing infrastructure.

High levels of research and development (R&D) investment enhance enterprises’ capacity for external collaboration. Enterprises that allocate significant resources to R&D typically forge enduring partnerships with academia and research institutions, fostering robust industry-university-research networks. Leveraging computing infrastructure platforms, these enterprises establish close innovation alliances with a wide array of partners, spanning universities, research institutions, and both upstream and downstream collaborators. This interdisciplinary collaborative innovation approach markedly broadens and deepens the utilization of computing resources. Conversely, organizations with limited R&D investment struggle to establish effective cooperation mechanisms and collaborative networks, impeding their ability to efficiently integrate external computing resources.

Hypothesis 2 posits that R&D investment positively moderates the impact of computing infrastructure on manufacturing innovation.

2.2.3. Market Share Positively Moderates the Relationship Between Computing Infrastructure Construction and Innovation in Manufacturing Enterprises

During the evolution and modernization of the manufacturing sector, the computing infrastructure plays a crucial role in providing external computing resources for enterprises. The extent to which enterprises can effectively leverage these resources hinges not only on their internal research and development capabilities but also on their market positioning. Drawing from the resource-based view theory [39], a firm’s competitive edge is derived from possessing or accessing scarce, unique, and irreplaceable resources. Market share, a key metric for assessing a firm’s market stance and competitive advantage [40], dictates the firm’s ability to assimilate and enhance the efficacy of external computing resources. Consequently, the impact of computing infrastructure on driving innovation is likely subject to significant modulation by the firm’s market share.

Market share is indicative of a firm’s competitive edge in data resources. Enterprises with a substantial market share benefit from a broader customer base and a more intricate transaction system, allowing them to amass a wealth of high-quality market data. These data, as a unique corporate asset, can be integrated into computing platforms to boost the precision and predictive capacity of computing models, thus enhancing the efficacy of innovation [41]. Conversely, enterprises with a smaller market share have restricted data volume, leading to inadequate samples being fed into computing platforms, thereby diminishing the support provided by computing resources for innovation.

High market share signifies a firm’s prominent standing within the industrial ecosystem, positioning them at the center of the industrial chain with significant decision-making authority and technological prowess [42]. This pivotal industrial niche empowers them to seamlessly infuse computing capabilities into crucial R&D activities and prompt collaboration among upstream and downstream entities to exchange computing models and insights, thereby fostering resource spillover and expansion. In contrast, enterprises with low market share encounter challenges in leveraging the innovation potential of computing resources effectively within collaborative networks, given their limited industry influence.

Market share is intricately linked to the disparity in capital resources among enterprises. Enterprises with a substantial market share tend to enjoy elevated profitability and possess the necessary financial means to establish robust computing infrastructure and assemble expert teams. This accumulation of resources aligns with the concept of “scarce and challenging-to-replicate internal resources” highlighted by the Resource-Based View (RBV) theory, empowering enterprises to leverage computing capabilities more effectively in intricate innovation endeavors. Conversely, enterprises with a smaller market share face constraints due to limited financial resources, often resorting to leasing or sharing arrangements, thereby impeding the depth and sustainability of their computing applications.

Hypothesis 3 posits that market share positively moderates the impact of computing infrastructure on manufacturing innovation.

2.2.4. Heterogeneous Effects of Computing Infrastructure Construction on Innovation in Manufacturing Enterprises

The impact of computing infrastructure construction on innovation in manufacturing enterprises is clearly heterogeneous. The influence of computing infrastructure on innovation within manufacturing enterprises is notably varied. Enterprises with differing ownership structures exhibit distinct organizational dynamics, innovation incentives, and computing resource costs, all of which shape the impact on innovation. State-owned enterprises often face inefficiencies in utilizing computing resources due to complex organizational hierarchies, protracted approval processes, and burdensome decision-making in innovation [43]. In contrast, non-state-owned enterprises benefit from streamlined structures and decision-making processes, enabling swift deployment of computing resources to foster innovation. Furthermore, state-owned enterprises prioritize stability and compliance, which can make innovators less responsive to market shifts and diminish enthusiasm for exploring computing resources. Conversely, non-state-owned enterprises emphasize performance and market adaptability, encouraging innovators to actively engage with computing infrastructure to enhance innovation.

Regional differences in financial resources, innovation capability, and industrial structures lead to different effects of computing infrastructure on manufacturing innovation. China’s eastern region has more developed financial markets and different financing mechanisms, giving manufacturing enterprises access to ample funding for computing infrastructure and technological upgrades, which is helpful for ongoing optimization. The underdeveloped financial markets in central and western regions hinder enterprises from continuously optimizing computing resource deployment, thereby reducing the efficiency of resource utilization. Moreover, the eastern region has more advanced digital infrastructure with many research institutions and innovation support facilities [44]. This has led to a strong industry-academia-research collaborative network. Manufacturing enterprises have easier access to computing platforms, optimizing these resources for innovation. Central and Western regions have a weaker technology base and a lack of high-end talent. Limited access and restrictions in computing resources inhibit the computing infrastructure’s role in supporting enterprise innovation. In addition, advanced manufacturing clusters are mostly located in the eastern region, where geographical proximity reduces the costs of sharing computing resources and developing collaborative platforms.

Ownership concentration influences the internal governance, innovation decisions, and talent development of manufacturing enterprises, affecting the impact of computing infrastructure construction on innovation. In companies with concentrated ownership, decision-making power typically rests with major shareholders or core management, leading to hierarchical governance. This structure can hinder computing resources due to multi-level reviews and restricted cross-departmental collaboration. In contrast, in enterprises with low ownership concentration, governance can be decentralized, which allows rapid deployment and efficient use of computing resources. High ownership concentration can also result in innovation behavior biased towards the risk preferences of major shareholders, leading to conservative innovation decisions and preventing internal innovation openness [45]. Manufacturing enterprises with lower ownership concentration tend to be more innovative and inclusive, leading to risk-sharing mechanisms and encouraging the use of computing resources for innovation. Even more importantly, enterprises with high ownership concentration struggle to attract external computing experts and technical talent. Smaller-concentration enterprises can more readily engage independent directors and senior industry experts in governance and attract talent through equity incentives and options, supporting innovation activities leveraging computing infrastructure.

Therefore, hypothesis 4 posits that the influence of computing infrastructure on manufacturing innovation varies based on ownership types, regional differences, and ownership concentration. This effect is more pronounced in non-state-owned enterprises, those located in the eastern region, and enterprises with low ownership concentration.

3. Data and Methods

3.1. Model Specification

This study treats the construction of computing infrastructure in China as a quasi-natural experiment. Utilizing the methodology of Zhao and Liu [46], a multi-period difference-in-differences (DID) approach is employed to assess the impact of this infrastructure on manufacturing enterprise innovation. The model is constructed as follows:

$${Innovation}_{i,t+1}={\beta }_0+{\beta }_1\times {DID}_{i, t}+{\beta }_2{Controls}_{i,t}+{\delta }_i+{\psi }_t+{\epsilon }_{it}$$

(1)

In this study, ${DID}_{i, t}$ is a dummy variable. The variable is assigned a value of 1 if enterprise i is located in a city that initiated computing infrastructure construction in year t or thereafter; otherwise, it is set to 0. The coefficient ${\beta }_1$ of ${DID}_{i, t}$ is central to our analysis, indicating the effect of computing infrastructure on manufacturing enterprise innovation. A positive ${\beta }_1$ suggests that such infrastructure promotes innovation, whereas a negative ${\beta }_1$ implies inhibition.

${Innovation}_{i,t+1}$ denotes the innovation of manufacturing enterprises in period t + 1. Given that innovation requires time, the construction of computing infrastructure is unlikely to influence the innovation of manufacturing enterprises in the current period. Following Cui and Wang [47], we lag the innovation variable by one period. The model includes a set of time-varying control variables (${Controls}_{i,t}$) at both enterprise and city levels, individual fixed effects (${\delta }_i$), year fixed effects (${\psi }_t$), and a random disturbance term (${\epsilon }_{it}$).

3.2. Variable Definitions

3.2.1. Explanatory Variables

Computing infrastructure serves society by providing essential computational capabilities. Recently, China has actively advanced the construction of such infrastructure, exemplified by its National Supercomputing Centers. These centers, sanctioned by the Ministry of Science and Technology, are pivotal in gauging a nation’s or region’s scientific and technological prowess. Since the inaugural National Supercomputing Center in Tianjin in 2009, these facilities have significantly driven industrial transformation and innovation. This study employs the establishment of National Supercomputing Centers as a quasi-natural experiment to proxy the construction of computing infrastructure. The dummy variable ${DID}_{i, t}$ indicates whether a National Supercomputing Center has been established in the city where enterprise i is located from year t onwards.

3.2.2. Explained Variables

Previous research commonly employs the number of patent applications [48] or granted patents [49] as indicators of enterprise innovation. Given that it typically takes 3 to 5 years for a patent to be granted from the application date, relying solely on the number of granted patents may not provide a current assessment of innovation. Following [50], this study utilizes the natural logarithm of patent applications by manufacturing enterprises as a measure of their innovation capacity.

3.2.3. Control Variables

Adopting the methodology of Li et al. [51], the enterprise-level control variables include firm size (Size), capital structure (Lev), profitability (Roe), fixed asset proportion (Fix), intangible asset proportion (Intangible), and operating revenue (Revenue). At the macro-level, the control variables are foreign direct investment (FDI) and real GDP per capita (GDP) (

Table 1. Definition of variables.

Variable Type	Variable Symbol	Variable Definition
Explanatory variable	DID	The treatment group is assigned a value of 1 and the control group is assigned a value of 0
Explained variable	Innovation	ln(1 + corporate patent applications)
Control variable	Size	ln(1 + total corporate assets)
	Lev	Total liabilities/Total assets
	Roe	Net profit/Shareholders’ equity
	Fix	Fixed assets/Total assets
	Intangible	Intangible assets/Total assets
	Revenue	Operating revenue
	FDI	Foreign direct investment
	GDP	Real GDP per capita

).

3.3. Data Sources and Descriptive Statistics

This study utilizes manufacturing enterprise data from the CSMAR database, urban data from the China Statistical Yearbook, and information on National Supercomputing Centers compiled by the authors from public sources. With the first National Supercomputing Center established in Tianjin in 2009, the research period spans from 2003 to 2023 to ensure an adequate number of observations before these centers were established. In addition, to enhance the representativeness of the sample, this study processes the data by excluding ST and *ST enterprises, as well as eliminating abnormal observations with asset-liability ratios outside the [0, 1] range. Missing data are addressed using linear and average interpolation, and continuous variables undergo 1% winsorization. The final dataset comprises 26,046 observations from 2585 listed manufacturing enterprises.

Table 2. Descriptive statistics.

Variable	N	Mean	SD	Min	Max
Innovation	26,046	0.813	1.068	0	4.585
DID	26,046	0.142	0.349	0	1
Size	26,046	22.01	1.184	19.70	25.48
Lev	26,046	0.421	0.202	0.0520	0.965
Roe	26,046	0.973	43.50	−76.76	3209
Fix	26,046	0.232	0.140	0.0120	0.634
Intangible	26,046	0.0430	0.0360	0	0.206
Revenue	26,046	7.202 × 109	2.614 × 1010	0	8.876 × 1011
FDI	26,046	797,476	1.447 × 106	0	7.143 × 106
GDP	26,046	97,315	51,746	9330	203,489

presents the basic statistical characteristics of the main variables, indicating that all variables fall within a reasonable range and exhibit notable differences.

4. Analysis of Empirical Results

4.1. Benchmark Regression Analysis

Table 3. Impact of computing infrastructure construction on the innovation of manufacturing enterprises.

	(1)	(2)	(3)
	FE1	FE2	FE3
DID	0.111 ***	0.102 ***	0.105 ***
	(0.038)	(0.037)	(0.037)
Size		0.115 ***	0.115 ***
		(0.010)	(0.010)
Lev		0.210 ***	0.210 ***
		(0.035)	(0.035)
Roe		0.000	0.000
		(0.000)	(0.000)
Fix		0.408 ***	0.408 ***
		(0.054)	(0.054)
Intangible		0.933 ***	0.932 ***
		(0.168)	(0.168)
Revenue		0.000 ***	0.000 ***
		(0.000)	(0.000)
FDI			−0.000 **
			(0.000)
GDP			0.000
			(0.000)
_cons	0.797 ***	−1.991***	−2.037 ***
	(0.007)	(0.216)	(0.218)
Individual-fixed	Yes	Yes	Yes
Time-fixed	Yes	Yes	Yes
N	26,046	26,046	26,046
R2	0.646	0.656	0.657

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

displays the benchmark regression results assessing the effect of computing infrastructure construction on manufacturing enterprise innovation. Column (1) presents the regression without control variables. Column (2) includes enterprise-level controls, while Column (3) incorporates both enterprise and city-level controls.

All results are significantly positive, supporting the conclusion that the construction of computing infrastructure, exemplified by the National Supercomputing Centers, enhances manufacturing enterprise innovation, thereby confirming Hypothesis 1.

4.2. Parallel Trend Test

Before performing multi-period DID regression, it is crucial to verify that the experimental and control groups satisfy the parallel trend assumption, ensuring that any observed innovation effect in manufacturing enterprises is attributable to the construction of computing infrastructure. We conducted a parallel trend test, using the year prior to the computing infrastructure establishment as the baseline. Figure 1 presents the results, showing no significant effects before the computing infrastructure’s construction, which indicates no significant time trend differences between the control and treatment groups during and prior to the policy implementation year. Thus, the parallel trend test is satisfied. Additionally, the impact of computing infrastructure on manufacturing innovation is delayed.

4.3. Robustness Tests

4.3.1. Placebo Test

To mitigate the influence of random factors and confirm that the empirical results are not due to chance, we employed a placebo test following the methodology of Bertrand et al. [52]. We conducted 500 independent, random experiments. Figure 2 shows that the estimated treatment effect (indicated by the vertical solid line) is positioned on the right side of the placebo effect distribution, marking it as an unusually extreme value. Furthermore, both the two-sided and right-sided p-values are below 0.05, allowing us to strongly reject the null hypothesis that the treatment effect is zero.

4.3.2. Excluding the Impacts of Other Policies

Throughout the study period, alongside computing infrastructure policies, the government implemented additional similar policies. These could potentially influence the research outcomes, thereby impacting the reliability of the conclusions. This paper accounts for these policies in the model to mitigate their interference during the same period.

Intelligent manufacturing and computing infrastructure are crucial for advancing digital and intelligent transformation, with a notable correlation between them. Intelligent manufacturing focuses on the smart evolution of the manufacturing sector, whereas computing infrastructure represents a new information infrastructure. The findings in Column (1) of

Table 4. Excluding the interference of other policies.

	(1)	(2)
	Intelligent Manufacturing	Innovation City
DID	0.097 ***	0.243 ***
	(0.036)	(0.054)
Intelligent DID	0.744 ***
	(0.057)
Innovation DID		−0.051 *
		(0.026)
Lev	0.213 ***	0.297 ***
	(0.034)	(0.045)
Roe	0.000	−0.000
	(0.000)	(0.000)
Fix	0.404 ***	0.429 ***
	(0.053)	(0.070)
Intangible	0.898 ***	1.088 ***
	(0.166)	(0.207)
Revenue	0.000 ***	0.000 ***
	(0.000)	(0.000)
FDI	−0.000	−0.000 **
	(0.000)	(0.000)
GDP	0.000 *	0.000
	(0.000)	(0.000)
_cons	−1.953 ***	−1.852 ***
	(0.214)	(0.291)
Individual-fixed	Yes	Yes
Time-fixed	Yes	Yes
N	26,046	15,390
R2	0.661	0.679

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

confirm that the paper’s conclusions remain robust even when the impact of intelligent manufacturing is excluded.

The pilot policy for innovative cities is closely linked to enterprise innovation. To account for this, an interaction term was introduced to isolate the influence of innovative cities. As shown in Column (2) of Table 4, even after excluding the effects of this pilot policy, the study’s conclusions remain unchanged.

4.3.3. PSM-DID

To address potential sample selection bias in the benchmark regression, this study utilizes the Propensity Score Matching (PSM) method to address endogeneity.

Table 5. PSM-DID.

	(1)	(2)
	Neighbor	Kernel
DID	0.264 ***	0.114 ***
	(0.073)	(0.042)
Size	0.183 ***	0.168 ***
	(0.037)	(0.025)
Lev	0.081	0.020
	(0.137)	(0.093)
Roe	−0.000	0.000
	(0.000)	(0.000)
Fix	0.271	0.484 ***
	(0.191)	(0.134)
Intangible	2.701 ***	1.612 ***
	(0.593)	(0.347)
Income	0.000 ***	0.000 ***
	(0.000)	(0.000)
FDI	0.000	−0.000
	(0.000)	(0.000)
GDP	0.000	0.000
	(0.000)	(0.000)
_cons	−3.480 ***	−3.081 ***
	(0.825)	(0.524)
Individual-fixed	Yes	Yes
Time-fixed	Yes	Yes
N	9116	25,640
R2	0.734	0.735

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

presents the results: Column (1) displays the regression outcomes from 1:2 nearest-neighbor matching without replacement, while Column (2) shows results following kernel matching. Both columns reveal significantly positive outcomes, confirming that the paper’s conclusions remain robust after applying propensity score matching.

4.4. Analysis of the Moderating Mechanism

Empirical analysis reveals that computing infrastructure enhances innovation in manufacturing enterprises. However, which factors influence this innovation effect? To address this, we develop a moderating-effect model to assess the impact of enterprise R&D investment and market share on the innovation effect of computing infrastructure. The model is defined as follows:

$${Innovation}_{i,t+1}={\beta }_0+{\beta }_1{DID}_{i,t}+{\beta }_2{Moderator}_{i,t}+{\beta }_3{DID}_{i,t}\times {Moderator}_{i,t}+\gamma {Controls}_{i,t}+{\delta }_i+{\psi }_t+{\epsilon }_{it}$$

(2)

${Moderator}_{i,t}$ serves as the moderating variable, while ${DID}_{i,t}\times {Moderator}_{i,t}$ represents the interaction term between this moderating variable and the construction of computing infrastructure. The coefficient ${\beta }_3$ of this interaction term is the primary focus of interest.

4.4.1. Enterprise R&D Investment

Incorporating the natural logarithm of enterprise R&D investment into Model (2), the regression results are presented in Column (1) of

Table 6. Moderating effect.

	(1)	(2)
	R&D	Competition
DID	0.076 **	0.106 ***
	(0.036)	(0.037)
R&D	0.003 *
	(0.002)
DID × R&D	0.023 ***
	(0.006)
Competition		0.226 *
		(0.132)
DID × Competition		0.582 *
		(0.351)
Income	0.000 ***	0.000 ***
	(0.000)	(0.000)
FDI	−0.000 **	−0.000
	(0.000)	(0.000)
GDP	0.000	0.000
	(0.000)	(0.000)
_cons	−1.851 ***	−2.027 ***
	(0.227)	(0.218)
Individual-fixed	Yes	Yes
Time-fixed	Yes	Yes
N	25,673	26,046
R2	0.655	0.657

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

. The coefficient for the core explanatory variable DID is significantly positive. Additionally, the interaction term between enterprise R&D investment and computing infrastructure also shows a significantly positive coefficient. This suggests that the innovation effect induced by the construction of computing infrastructure is positively moderated by enterprise R&D investment.

Enterprises with substantial R&D investment exhibit enhanced capacity to absorb external resources, thereby improving their utilization of computing resources. These enterprises demonstrate superior resource allocation capabilities, enabling them to mobilize computing infrastructure more effectively to bolster innovation. Moreover, enterprises with significant R&D investments possess a systematic approach to developing collaborative resources. By establishing cooperative networks, they connect with computing infrastructure platforms, ultimately enhancing their innovation efficiency.

Consequently, R&D investment positively moderates the innovation impact of computing infrastructure construction, supporting Hypothesis 2.

4.4.2. Market Share

In Table 6, Column (2) shows a significantly positive coefficient for the core explanatory variable. Additionally, the coefficient for the interaction term is significantly positive at the 1% level, suggesting that market share positively moderates the influence of computing infrastructure construction on innovation in manufacturing enterprises.

Enterprises with a significant market share benefit from superior data resources. This advantage enhances the precision of computing model predictions, optimizes the utilization of computing resources, and accelerates innovation within the organization. Such enterprises typically occupy a core ecological niche within the industry ecosystem, enabling them to fully leverage computing resources both internally and across the supply chain, thereby enhancing their innovation capacity. High-market-share enterprises tend to allocate substantial financial resources to computing infrastructure, enabling large-scale deployment and effectively amplifying their innovation-enhancing effects.

Consequently, market share plays a positive role in moderating the innovation outcomes resulting from the construction of computational infrastructure, thus confirming Hypothesis 3.

4.5. Heterogeneity Analysis

4.5.1. Ownership Type

This study examines the varying effects of computing infrastructure construction on manufacturing enterprises with different ownership structures by categorizing listed manufacturing enterprises into non-state-owned and state-owned enterprises. Columns (1) and (2) of

Table 7. Heterogeneous impact of computing infrastructure construction on innovation in manufacturing enterprises.

	(1)	(2)	(3)	(4)	(5)	(6)
	Non-State	State-Owned	Midwest	East	Concentration	Dispersed
DID	0.193 ***	0.010	0.076	0.132 ***	0.024	0.190 ***
	(0.054)	(0.058)	(0.062)	(0.051)	(0.060)	(0.049)
Size	0.164 ***	0.114 ***	0.066 ***	0.168 ***	0.096 ***	0.132 ***
	(0.014)	(0.018)	(0.015)	(0.014)	(0.020)	(0.015)
Lev	−0.009	0.622 ***	0.347 ***	0.105 **	0.242 ***	0.172 ***
	(0.045)	(0.061)	(0.053)	(0.048)	(0.064)	(0.048)
Roe	−0.000	0.000	−0.003 **	−0.000	0.000	0.000
	(0.000)	(0.000)	(0.001)	(0.000)	(0.000)	(0.000)
Fix	0.379 ***	0.496 ***	0.137 *	0.575 ***	0.486 ***	0.423 ***
	(0.072)	(0.088)	(0.081)	(0.075)	(0.092)	(0.073)
Intangible	0.974 ***	0.773 **	0.342	1.159 ***	1.267 ***	0.412 *
	(0.207)	(0.317)	(0.271)	(0.227)	(0.304)	(0.222)
Revenue	0.000 ***	0.000 ***	0.000 ***	0.000 ***	0.000 ***	0.000 ***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
FDI	0.000 *	−0.000 ***	−0.000 ***	−0.000 **	−0.000*	−0.000
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
GDP	0.000 **	−0.000	0.000 ***	−0.000 *	−0.000	0.000
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
_cons	−3.053 ***	−2.215 ***	−1.315 ***	−2.980 ***	−1.608 ***	−2.400 ***
	(0.299)	(0.392)	(0.340)	(0.307)	(0.433)	(0.317)
Individual-fixed	Yes	Yes	Yes	Yes	Yes	Yes
Time-fixed	Yes	Yes	Yes	Yes	Yes	Yes
N	16,361	9636	8110	17,463	11,564	14,370
R2	0.656	0.686	0.649	0.657	0.686	0.696

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

show the regression results for these subsamples, revealing that computing infrastructure more significantly enhances innovation in non-state-owned enterprises. This may be because non-state-owned enterprises, compared to their state-owned counterparts, exhibit a higher degree of market orientation, allowing them to respond more swiftly and accurately to changes in market demand. Their flexible organizational structures, pronounced innovation incentives, and shorter decision-making chains facilitate rapid resource integration for precise innovation, thereby enhancing innovation speed and efficiency. Additionally, small- and medium-sized private enterprises often face limitations in technical infrastructure. By leveraging the computing platform, they can lower fixed innovation and R&D costs, utilizing computing resources to conduct innovation activities at reduced marginal costs, thus boosting their innovation efficiency.

4.5.2. Regional Heterogeneity

This study categorizes enterprises based on their geographical location into those in the central and western regions and those in the eastern region. The regression findings, presented in Columns (3) and (4) of Table 7, indicate that computing infrastructure more significantly influences manufacturing innovation in the eastern region. This may be due to the more advanced financial markets in Eastern China, which facilitate greater access to financial resources and support the large-scale deployment of computing infrastructure. The majority of China’s data centers are situated in this region, characterized by robust digital infrastructure, a concentration of scientific research institutes, and a significant presence of leading manufacturing enterprises. Most of China’s data centers are located in this region, with robust digital infrastructure, a concentration of scientific research institutes, and a significant number of leading manufacturing enterprises. This region supports a strong innovation ecosystem with a high demand for computing resources and efficient utilization. Most advanced manufacturing clusters are located in the east, where upstream and downstream systems of the industrial chain are well developed. This facilitates collaboration among enterprises in sharing computing resources and constructing computing platforms.

4.5.3. Ownership Concentration

This paper investigated how ownership concentration affects innovation decisions in manufacturing enterprises. We divided the sample into two groups based on the mean number of shares: relatively concentrated and relatively dispersed. Different regressions were performed for each group of shares. Results are presented in Columns (5) and (6) of Table 7. Results showed that computing infrastructure affects innovation more strongly in manufacturing companies with a lower concentration of shares. The result may be attributed to several factors. In manufacturing enterprises with a low concentration of shares, decision-making power is more decentralized, and internal governance is more market-focused. This organizational structure allows for faster access to computing resources and facilitates the integration of external computing platforms into innovation activities. Additionally, such enterprises typically exhibit more dispersed ownership structures and diverse innovation objectives, fostering internal mechanisms for sharing innovation risks and encouraging openness in innovation-related decision-making. As a result, they are more inclined to pursue diversified and disruptive innovation initiatives across design, production, and sales processes by leveraging advanced computing resources. Furthermore, low-concentration enterprises tend to be more attractive to high-end computing talent, which improves the utilization efficiency of computing infrastructure and further enhances innovation performance.

In summary, the positive impact of computing infrastructure construction on manufacturing enterprise innovation is heterogeneous, with the effect being more significant in non-state-owned enterprises, those located in eastern regions, and firms with lower ownership concentration, thereby validating Hypothesis 4.

5. Further Analysis: The Impact of Computing Infrastructure Construction on the Quality of Innovation in Manufacturing Enterprises

We have examined the influence of computing infrastructure construction on innovation within manufacturing enterprises. Yet, the economic value derived from innovation activities is contingent not solely on their quantity, but also on their quality. Hence, we will now delve into the effects of computing infrastructure construction on the quality of innovation in manufacturing enterprises.

Computing infrastructure promotes the quality of innovation in manufacturing enterprises. High-quality innovation often involves technology-intensive tasks that require large-scale computing and high-performance modeling. The computing infrastructure provides technical support such as high-performance computing and digital twin modeling platforms, which provide a foundation for complex innovation activities. Computing infrastructure also accelerates data processing and improves information feedback efficiency for improved innovation quality. Data is a vital ingredient for innovation; manufacturing enterprises need valuable insights from large datasets generated by intelligent manufacturing, digital twins, and industrial simulations. Large-scale data processing facilitates data-driven innovation and refines innovation quality. Also, computing infrastructure extends the boundaries of knowledge acquisition and accumulation for manufacturing enterprises [53], thereby enhancing innovation quality. High-quality innovation necessitates long-term knowledge accumulation. By offering algorithm tool libraries and computing models, this infrastructure expands the knowledge network of manufacturing enterprises. It enables the structural integration of implicit knowledge, such as historical R&D data and experimental parameters, forming an internal innovation knowledge system.

To test this conjecture, we utilized patent knowledge breadth to develop an innovation quality indicator [54]. Typically, a patent that incorporates more knowledge is considered of higher quality. We employed the Herfindahl index at the subgroup level of patent classification numbers to quantify patent knowledge breadth, accounting for variations at this level. We integrated the constructed innovation quality indicators into the regression model, with results shown in

Table 8. Impact of computing infrastructure construction on innovation quality.

	(1)
	Quality
DID	0.050 ***
	(0.014)
Size	0.008
	(0.005)
Lev	−0.065 ***
	(0.018)
Roe	−0.000 **
	(0.000)
Fix	−0.026
	(0.028)
Intangible	0.301 ***
	(0.091)
Revenue	0.000 ***
	(0.000)
FDI	0.000 ***
	(0.000)
GDP	0.000 *
	(0.000)
_cons	0.248 **
	(0.110)
Individual-fixed	Yes
Time-fixed	Yes
N	24,064
R2	0.498

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

. These findings demonstrate that building computing infrastructure enhances both the quantity and quality of innovation in manufacturing enterprises.

6. Discussion

While existing research has explored the effects of traditional, communication, and digital infrastructures on enterprise innovation, micro-level evidence on how computing infrastructure influences manufacturing innovation remains scarce. This paper addresses this gap by using the establishment of national supercomputing centers as a quasi-natural experiment to systematically investigate the impact of computing infrastructure on manufacturing enterprise innovation. Research shows that computing infrastructure significantly enhances both the quantity and quality of innovation in the manufacturing sector, aligning with existing literature that underscores the role of digital infrastructure in boosting corporate innovation efficiency [55]. However, some studies suggest that while upgrading information infrastructure increases innovation input and output, it may suppress innovation quality [56]. This discrepancy highlights distinct mechanisms of different infrastructure types: information infrastructure primarily boosts quantity by lowering information acquisition costs and expanding R&D scale, whereas computing infrastructure, through high-performance computing and complex modeling, supports high-value R&D and complex process optimization, thereby substantially improving innovation quality. This study revises the notion that infrastructure mainly promotes innovation in quantity, emphasizing the unique role of computing power infrastructure in enhancing quality.

The test of the moderating mechanism shows that R&D investment and market share positively influence the impact of computing infrastructure on manufacturing innovation. This finding confirms the relevance of absorptive capacity theory and the resource-based view in the digital economy context. It underscores the critical roles of various stakeholders: policymakers must foster a supportive institutional and competitive environment, enterprise managers should strengthen R&D capabilities and market positions, and infrastructure providers need to offer efficient, demand-aligned computing. The effective transformation of computing into manufacturing innovation relies on the synergy of these elements.

The impact of computing infrastructure on innovation in manufacturing is more pronounced in non-state-owned enterprises, those located in the eastern region, and enterprises with lower equity concentration. Existing research consistently shows that digital infrastructure significantly enhances innovation, particularly in non-state-owned enterprises and those in the eastern region [57]. However, existing research presents conflicting views on the relationship between equity concentration and corporate innovation. High equity concentration can enhance major shareholder oversight and mitigate agency problems, thereby fostering innovation [58]. Conversely, it can also curtail R&D investment, inhibiting innovation [59]. Some studies propose an inverted U-shaped relationship, where moderate concentration is most conducive to innovation [60]. This paper aligns with the perspective that “high concentration inhibits innovation” and supports the notion within the inverted U-shaped framework that “excessive concentration is unfavorable to innovation,” but it does not corroborate the idea that “high concentration promotes innovation.” This suggests that the impact of equity structure on innovation is context-dependent: in environments with strong externalities and high complementarity, such as computing infrastructure, firms with low equity concentration can more flexibly leverage external resources, thereby enhancing their innovation potential. This study not only deepens the understanding of the equity structure-innovation relationship but also offers new empirical insights into the innovation effects of computing infrastructure across different governance environments.

7. Conclusions

7.1. Main Conclusions

Computing power has emerged as a critical productive force driving the transformation, upgrading, and sustainable development of the manufacturing sector. Prior research has predominantly examined the effects of traditional or general digital infrastructure on innovation, neglecting a systematic analysis of computing infrastructure, a novel form of information infrastructure. This study develops a comprehensive analytical framework grounded in General-Purpose Technology (GPT) theory, integrating absorptive capacity theory and the resource-based view. Utilizing the establishment of China’s National Supercomputing Centers as an exogenous policy shock, and analyzing data from Chinese listed manufacturing enterprises between 2003 and 2023, this paper applies the multi-period Difference-in-Differences (DID) approach to systematically assess the impact of computing infrastructure on manufacturing enterprise innovation.

The findings reveal that the establishment of computing infrastructure, exemplified by the National Supercomputing Centers, fosters innovation within manufacturing enterprises, a conclusion confirmed by robustness tests. Additionally, this innovation effect is positively influenced by enterprises’ internal R&D investment and external market share. The impact of computing infrastructure on innovation is more pronounced in non-state-owned manufacturing enterprises, those located in the eastern region, and those with low equity concentration. Furthermore, such infrastructure development enhances both the quantity and quality of innovation in manufacturing enterprises.

7.2. Theoretical and Practical Implications

This paper’s theoretical contribution lies in integrating computing infrastructure into the General-Purpose Technology (GPT) framework. By combining absorptive capacity theory with the Resource-Based View, it elucidates how computing infrastructure affects manufacturing enterprises’ innovation through the moderating roles of internal R&D investment and market share. This research extends the understanding of infrastructure’s impact on innovation and deepens insights into enterprise innovation mechanisms in the digital economy era. Empirical findings indicate that computing infrastructure enhances both the quantity and quality of innovation, with its effects shaped by factors such as ownership types, regional differences, and ownership concentration. This suggests that future models should avoid simplistic linear assumptions and incorporate the dual dimensions of quantity and quality, alongside market environment and enterprise characteristics, to improve the explanatory power and applicability of existing theories.

The practical significance of this study lies in showing that the construction of computing infrastructure can significantly promote innovation in the manufacturing sector, with particularly pronounced effects among non-state-owned enterprises, those in the eastern region, and those with lower ownership concentration. These findings highlight the importance of considering enterprise characteristics and regional disparities when the government strategizes regional computing deployment. Additionally, the results offer guidance for infrastructure developers regarding resource allocation and application scenario expansion. Furthermore, the research reveals that R&D investment and market share positively influence this dynamic, offering a foundation for enterprises to enhance R&D investment and develop market strategies.

7.3. Policy Recommendations

Research indicates that developing computing infrastructure can significantly boost manufacturing innovation, suggesting that initiatives like the “East-West Computing” project are feasible and can address regional disparities in computing supply and demand. Establishing a resource-sharing market and differential subsidy mechanisms can reduce entry barriers for small and non-state enterprises. However, building such infrastructure demands substantial investment and has a long timeline, making nationwide coverage challenging in the short term, and regional imbalances may persist.

Research findings indicate that manufacturing enterprises can fully harness the innovation potential of computing infrastructure only when they possess robust R&D capabilities and a significant market share. This underscores the need for increased R&D investment and the use of supercomputing centers or regional computing platforms for applications like digital twins and complex simulations, thereby enhancing innovation scale and quality. Nonetheless, the long-term and uncertain nature of R&D investment may constrain some enterprises, particularly small- and medium-sized ones, due to challenges in funding and talent acquisition.

This study reveals that computing infrastructure significantly boosts both the quantity and quality of innovation in manufacturing enterprises. Thus, optimizing the allocation of computing resources on the supply side holds practical importance for developers. Notably, enterprises in the eastern region and non-state-owned enterprises experience greater benefits. Developers should therefore focus on regional coordination and the specific needs of different enterprises. This involves increasing the supply of high-performance and intelligent computing resources, enhancing the integration of computing, network, and cloud services, and fostering resource sharing. Additionally, establishing joint R&D mechanisms with manufacturing enterprises to create tailored application solutions is crucial. However, challenges remain in cross-regional coordination and standardization, and regional disparities may temporarily limit the efficient use of computing resources.

7.4. Limitations and Future Research

This paper has several limitations. One limitation concerns the data and measurement methods. The distribution and number of China’s National Supercomputing Centers are limited. Although using these centers as a quasi-natural experiment provides some representativeness, it does not fully capture the extent of computing infrastructure development. To address this, robustness tests and heterogeneity analysis were conducted in this study. Nonetheless, future research should develop a more systematic computing index encompassing dimensions such as scale, industrial foundation, development environment, and application scenarios for a more comprehensive assessment.

Another limitation lies in the explanation of mechanisms. It primarily verifies the moderating effect through enterprise R&D investment and market share. However, it falls short in deeply exploring the specific pathways through which computing infrastructure influences manufacturing enterprise innovation. To address this, the paper employs a moderation test as a partial remedy. Future research could enhance theoretical explanatory power by integrating analyses of micro-mechanisms, such as organizational structure adjustments, factor allocation efficiency improvements, knowledge spillover, and collaborative innovation.

A further limitation is related to the generalizability of the findings. This study primarily centers on listed manufacturing enterprises in China, rendering its conclusions less universal in an international context. While this focus aids in identifying policy shocks, future research should expand to include small and medium-sized enterprises and draw comparisons with developed economies like the United States, the European Union, and Japan. Such comparisons would illuminate the similarities and differences in computing infrastructure across various institutional and market environments, thereby increasing the international relevance of the findings.