Introduction

The digital economy, which uses big data, artificial intelligence, 5 G, cloud computing, blockchain, the Internet of Things, and other emerging technologies to boost productivity and level of economic development, is a new type of economic form after the agricultural economy and industrial economy. Since it was first formally proposed in 1996, the concept of the digital economy has generated considerable discussion among academics (Adams and Akobeng, 2021; Akerman et al., 2015; Mueller and Grindal, 2019). It is widely accepted that the digital economy has the potential to reduce production costs, upgrade industrial structures, decrease carbon emissions, and encourage economic development (Cheng et al., 2023; Guo et al., 2023; He et al., 2022; Wu and Yang, 2022). Meanwhile, government management departments have also paid full attention to the layout of the digital economy. A study has pointed out that the total value of China’s digital economy has reached 50.2 trillion yuan in 2022, occupying 41.5% of the Gross Domestic Product (GDP) and has been playing a critical role in the high-quality development of the national economy (Wang et al., 2022).

The manufacturing industry reflects a country’s productivity level, which is an important factor that differentiates developing and developed countries (Liu et al., 2024). Meanwhile, the geographic pattern of manufacturing reflects the degree of spatial agglomeration and agglomeration pattern of manufacturing enterprises, both are important indicators of regional economic development. In the digital economy era, data and information, as new production factors, have brought disruption and impact to traditional enterprises’ production and spatial organization methods. In the process of continuous deep integration with traditional enterprises, the digital economy has been changing the constraints of factor endowment conditions and geographical location on enterprises significantly and promoting enterprises to relocate, migrate, flow and gather together to form a new economic geographical pattern to a great extent. Therefore, it can be inferred that the digital economy has been reshaping the geographical patterns of enterprises in China. However, there are still differences in existing research on the impact of the digital economy on the geographical pattern of enterprises.

There are two completely opposite views in academia on how the digital economy affects the enterprises’ geographical patterns. Some scholars hold the opinion that the digital economy can exert a dispersion effect on enterprises. Cairncross (1997) proposed the “death of distance” hypothesis, considering that the digital economy has broken the physical space-time constraints and built a bridge between enterprises in different geographical locations. The advent of the digital economy amplifies the autonomy of enterprises in terms of their geographical positioning, and weakens the importance of geographical factors (Banalieva and Dhanaraj, 2019; Goolsbee, 2000; Tranos and Nijkamp, 2013). According to the “global village theory”, compared to clustered enterprises, geographically isolated enterprises benefit more because good communication conditions would reduce the operating costs of enterprises (McLuhan, 1994). Although the digital economy has produced a decentralized mechanism for enterprises to some degree, the digital economy ca not completely eliminate the influence of geographical factors. The other scholars believe that the digital economy can generate agglomeration effects on enterprises. On the one hand, digital technology can significantly lower the cost of cross-regional flow and spatial agglomeration of enterprises. The network’s strong externality and positive feedback effects make it possible to form scale effects when the scale of Internet users reaches a critical value. Therefore, enterprises will further cluster in central cities with higher levels of the digital economy development. On the other hand, due to the uneven distribution of advanced information and communication technologies in geographical space, there is a significant digital divide between regions. Developed regions rely on their digital information advantages to attract talents and funds from neighboring underdevelopment regions, resulting in the polarization and siphon effect. Edward and Michael (2001) noted that the growth of the Internet can not only disperse certain daily activities, but also increase the complexity and time dependence of production activities, making agglomeration more critical in the information age. Wang et al. (2003) re-explored the distance death hypothesis in the information age and concluded that it is too early to affirm the distance death hypothesis. Some scholars pointed out that in the absence of transaction costs, Americans prefer to buy electronic products from nearby countries (Bernardo and Avi, 2005). As time goes on, the barriers of geographical distance to bilateral trade are rising, not falling (Faqin and Nicholas, 2012). The complex uncompilable information and other factors make the importance of geography can’t be completely ignored (Edward and Michael, 2001).

However, to the best of the authors’ knowledge, there are few papers that has thoroughly explored the influences of digital economy on the evolution of enterprises’ geographical patterns from a micro enterprise perspective. Many researches have looked into the effect of ICT technology on enterprise distribution (Lendle et al., 2016; Piet and Roger, 2004). Different from ICT, as a broad concept, the digital economy broadly contains all economic forms that employ data to lead resources to be effective and boost productivity development. Moreover, studying the impact of the digital economy on the geographical pattern of enterprises from a micro enterprise perspective can avoid both the “black box” problem from a macro perspective and the fragmentation defect from an individual perspective.

Another shortcoming of the existing literatures is that they all overlook the spatial correlation effects when exploring the impact of digital economy on enterprises’ geographical patterns. According to Tobler’s First Law of Geography, nothing exists in isolation (Tobler, 1970). The status that one thing is dependent on another is indicated by spatial correlation. Actually speaking, due to cooperation and competition mechanisms, the distribution of enterprises in one area may be influenced by the distribution of enterprises in its spatial-related regions. In that case, if we adopt traditional panel regression methods such as fixed/random effects and the GMM method, we can only obtain the direct effects of the digital economy on the enterprise’s geographical patterns, but fail to capture the spatial spillover effects of the digital economy on the enterprise’s geographical patterns, which would cause biased estimations.

To make up for the omissions in the existing literature, the study takes Chinese micro-enterprises as the research object and employs the spatial econometric technology to comprehensively probe into the far-reaching effects of the digital economy on Chinese manufacturing enterprises’ geographical patterns. We also investigate the heterogeneous effects and the impact mechanisms of the digital economy on the geographical patterns of enterprises.

Based on this, this paper utilizes spatial econometrics techniques to comprehensively explore the impact of digital economy on the geographic pattern of Chinese manufacturing enterprises. By systematically deconstructing the heterogeneous effect and influence mechanism of digital economy on the enterprises’ geographical patterns, this paper reveals the internal logic and evolution law of spatial reconfiguration of enterprises driven by digital economy, and provides a theoretical paradigm and a practical path for the optimization of the location and the synergistic development of the region in the global digital transformation. In short, the contributions of our study to the existing literature are as follows. First, this study breaks through the traditional industrial geography research paradigm and systematically explains the theoretical mechanism of digital economy reconfiguring the geographical pattern of manufacturing enterprises. Compared with the existing literature adopting the macro perspective of industrial agglomeration (Yang et al., 2025), this paper innovatively integrates the theories of new economic geography and corporate behavioral geography to systematically reveal the heterogeneous conduction network of micro enterprises’ location decision-making, which fills in the research gap of the theory of micro enterprises’ digital location choice. Second, our paper adopts “Tianyancha” survey data of Chinese enterprises, which covers the basic information of Chinese enterprises at all levels and is conducive to identifying the regional and industrial distribution status of different types of enterprises. Existing researches on industry distribution are mainly based on industry or regional macro-level data. However, industry or regional data are often aggregated by the data of large-scale enterprises, and it ignores the heterogeneity of industrial distribution due to the lack of exploration of small-scale enterprise data. Third, the study explores the direct and spatial spillover effects of the digital economy on the enterprises’ geographic patterns by using the Spatial Durbin Model (SDM), which can reduce bias caused by ignoring the spatial correlation of digital economy and geographical pattern between cities in previous research.

This study’s remaining sections are organized as follows. Section 2 reviews relevant literature. Section 3 recommends the research methods and the data. Section 4 discusses the empirical findings. The paper is concluded in Section 5.

Literature review and research hypothesis

The impact of the digital economy on the geographical patterns of enterprises

The geographical pattern of manufacturing enterprises refers to the spatial distribution and combination structure of manufacturing enterprises within a specific range (Lavoratori et al., 2020). As the primary manifestation of enterprises’ geographical patterns, enterprises’ migration, mobility and agglomeration are widespread and frequent in economy activity. Enterprises migrate and converge results in the production factors flowing and gathering in a specific geographical area. The enterprise’s geographical patterns are driven or influenced by factors such as market potential, resource allocation, and economies of scale (Barrios et al., 2009; Ellison and Glaeser, 1999). According to the theory of new geoeconomics, the digital economy mainly reshapes the geographical pattern of enterprises through three mechanisms. Firstly, the digital economy not only expands the omni-channel consumption scenarios by reconfiguring the supply and demand matching mechanism, but also relies on big data analysis to accurately capture niche market demand and activate the potential of the long-tail market, which promotes the exponential expansion of the consumer market scale. Secondly, the digital economy has upgraded from the “invisible hand” to the “intelligent hand”, realizing accurate matching of factors at the micro level, reconfiguring the shape of industrial organization at the meso level, and pushing the boundary of production possibilities to continue to move outward at the macro level to continuously optimize the allocation of resources. Finally, the digital economy breaks through the physical limitations of traditional economies of scale through data element reuse, network effects, and intelligent technologies, achieving exponential growth of “larger scale, lower cost, and higher value”. Therefore, the following explains the impact of the digital economy on the geographic pattern of enterprises from the three channels of market potential, resource allocation, and scale economies.

Market potential is one of the core forces driving the agglomeration of manufacturing enterprises. It provides enterprises with significant competitive advantages through mechanisms such as reducing transportation costs, realizing economies of scale and scope, sharing labor pools, facilitating knowledge spillovers and innovation, reducing market risks, and attracting supporting services and infrastructure (Artz et al., 2016). As a result, regions with greater market potential tend to cluster more enterprises. Many studies believed that the digital economy has unleashed economic vitality and expanded market potential (Wu and Yang, 2022). The digital economy is driving the geographic pattern of manufacturing enterprises to take on a clustered appearance by expanding market potential. Market potential is the core driving force for enterprise agglomeration, and the digital economy has become an important factor for expanding market potential (Jolivet and Turon, 2019). On the one hand, the digital economy promotes the transparency of market information (Afolabi, 2023; Li et al., 2020). With the development of Internet and big data technology, access to information has become unprecedented convenient (Chen, 2020). Consumers can easily obtain a large amount of information about goods and services through channels such as search engines, social media, and e-commerce platforms. This trend of information transparency greatly weakens the monopoly position of traditional enterprises in information, forcing them to pay more attention to product quality and service level, thereby enhancing market potential. On the other hand, the digital economy has driven the rise of the platform economy. The platform economy achieves effective resource matching by building online platforms to connect supply and demand sides. These platforms not only provide channels for enterprises to showcase and sell products, but also create convenient shopping, travel, and entertainment experiences for consumers (Carlsson, 2004). The rise of the platform economy has restructured the traditional industrial chain, enabling small and medium-sized enterprises to quickly reach a massive user base at a lower cost, thereby increasing market potential (Cette et al., 2019). Therefore, the digital economy can expand the market potential and enhance the agglomeration scale of manufacturing enterprises.

Resource allocation is the foundation of manufacturing enterprises agglomeration. Key production factors such as labor, capital and technology are the prerequisites for enterprises agglomeration. The efficiency of resource allocation directly affects the production cost, innovation ability, response speed and overall competitiveness of enterprises in the region. Optimizing factor allocation can alleviate the problem of labor and capital mismatch, improve the efficiency of factor synergy, and promote enterprises agglomeration (Yin et al., 2019). The digital economy improves resource allocation in three ways. Firstly, the digital economy has significantly enhanced the elasticity and inclusiveness of the labor market through technological empowerment and model innovation, and built a multi-level and specialized employment “reservoir” mechanism (Mouelhi, 2009; Castiglione and Infante, 2014). Secondly, the digital economy promotes the agglomeration of capital in areas such as technology research and development, industrial upgrading, and data factor markets by optimizing the efficiency of capital flows and innovating investment and financing modes, so that capital shifts from the traditional crude allocation to precise, dynamic, and cross-domain efficient allocation. Thirdly, the digital economy significantly accelerates knowledge spillover effects through technology driven and ecological reconstruction. Digital technology can efficiently collect, store, process, and analyze data, greatly reducing the cost of data sharing and promoting the widespread dissemination of knowledge (Chen, 2022). A specialized labor market, efficient capital allocation, and intensive knowledge elements are conducive to reducing the search and research and development costs of enterprises, thus promotes the geographic agglomeration of enterprises (Mbanyele et al., 2024).

Economies of scale effect promotes the geographic agglomeration of manufacturing enterprises by lowering their production costs, improving resource utilization efficiency, and attracting the concentration of upstream and downstream affiliates, forming a positive feedback loop. The digital economy can drive manufacturing enterprises to exhibit a geographic pattern of agglomeration through economies of scale. On the one hand, the core infrastructure of the digital economy has the characteristics of high fixed costs and extremely low marginal costs. Once built, the cost of service for new users is almost zero, attracting a large number of enterprises to move in and share resources. On the other hand, e-commerce, fintech and other platform-type enterprises realize diminishing marginal costs through the expansion of user scale, further attracting upstream and downstream enterprises to concentrate in their periphery. The network effect of digital platforms leads to higher platform value with more users, which in turn attracts more enterprises and users to join, forming a “snowball” agglomeration. We formulate the following two hypotheses in light of the study above.

Hypothesis 1. The digital economy is conducive to driving manufacturing enterprises to show a geographic pattern of agglomeration.

Hypothesis 2. The digital economy facilitates manufacturing enterprise to exhibit a geographic pattern of agglomeration by increasing market potential, optimizing resource allocation, and leveraging economies of scale.

The spatial spillover effect of the digital economy on the geographical patterns of enterprises

The digital economy possesses immediacy, sharing, and high dissemination features. High efficiency digital platforms can decrease the innate spatial rampart of market players and increase the breadth and depth of interregional trading activities (Yi et al., 2022). Digital economy not only speeds up the flux of knowledge, decreases the expend of information delivery and eliminates information asymmetry, but also innovates new commerce to promote the efficiency of trading and forms a sharing mechanism for knowledge and technology (Zhang et al., 2022). The digital economy has promoted a combination of online and offline information dissemination channels. The development of the digital economy has broken the rule that knowledge and information spillovers decrease with the increase of geographical distance, improved the dissemination of technology and knowledge, and produced spillover effects on surrounding areas. Related literature confirmed that the digital economy has spatial spillover effects. Ma and Zhu (2022) pointed out that the digital economy influenced high-quality green development in adjacent regions through spatial spillover effects. Xue et al. (2022) argued that the digital economy held a significant spatial spillover effect on energy consumption. Cheng et al (2023), based on urban-level data in China, discovered that the spatial spillover effect of the digital economy on the carbon emission intensity of neighboring cities presented a U-shaped pattern. Therefore, we synthesize the above analysis and put forward hypothesis 3 as follows.

Hypothesis 3. The digital economy exhibits a significant spatial spillover effect on the geographical patterns of enterprises.

Regression specification and data

Regression specification

Elhorst (2010) pointed out that ignoring spatial correlations in an econometric model would lead to biased results if variables have spatial correlations. The flow, transfer, and addition of enterprises will impact the distribution of enterprises in neighboring areas. Numerous prior pertinent researches offer strong justifications for the spatial dependency of enterprises distribution among cities (Fan and Wang, 2016). Moreover, the Spatial Durbin Model (SDM) can convert to the spatial autoregressive model (SAR) and the spatial error model (SEM). The SDM is more popular than both the SAR and the SEM (LeSage, 2014). Therefore, in order to avoid the problem of spatial dependence caused by omitted variables, we include a spatial lag term for explanatory variables, endogenize exogenous interaction effects, use the SDM to build the benchmark regression model with the method of Wang et al. (2018).

$$ag{g}_{it}={\alpha }_{0}+\rho {W}_{ij}ag{g}_{jt}+{\alpha }_{1}dig{i}_{it}+{\alpha }_{2}{W}_{ij}dig{i}_{jt}+\mathop{\sum }\limits_{k=1}^{K}{\beta }_{k}{X}_{kit}+\mathop{\sum }\limits_{k=1}^{K}{W}_{ij}{X}_{kjt}{\theta }_{k}+{\mu }_{i}+{\mu }_{t}+{\varepsilon }_{it}$$
(1)

where \(i\) and \(t\) denotes regions and years. \(K\) represents the number of control variables. \(agg\) denotes the level of agglomeration of manufacturing enterprises, which is used to reflect changes in the geographic pattern of manufacturing enterprises. \(digi\) stands for the development level of digital economy. \(X\) includes a vector of control variables. \({\alpha }_{0}\) and \({\varepsilon }_{it}\) are intercept term and random disturbance term. \({\alpha }_{1}\),\({\alpha }_{2}\),\(\rho\),\(\beta\) and \(\theta\) are the corresponding regression coefficients. \({u}_{i}\) and \({\mu }_{t}\) is the city and time fixed effect, which are used to control the unobserved city and time heterogeneity. And \(W\) is the spatial weight matrix.

Data

The empirical analysis makes use of panel data of 279 Chinese prefecture-level cities from 2011 to 2020. The reason for starting from 2011 is that inclusive finance data was released in 2011. All the data are collected from “Tianyancha” survey data platform of Chinese enterprises, The China Urban Statistical Yearbook, and The Research Center for Digital Finance of Peking University. The selection and introduction of variables are as follows:

  1. (1)

    Explained variable. The critical challenge in empirical studies is choosing a good indicator to comprehensively measure manufacturing enterprises agglomeration level. In existing research, the total output value of the manufacturing industry is often used to measure the agglomeration status of the manufacturing industry. However, the total output value of the manufacturing industry published by the National Bureau of Statistics of China is compiled using data from manufacturing enterprises above the designated size. If the total output value of the manufacturing industry is used to measure the geographical distribution of enterprises, then small-scale enterprises will be missed. Therefore, the variable of manufacturing enterprises agglomeration level (\(agg\)) is calculated by the ratio of industrial enterprises in each region to the overall number of industrial enterprises, which can mitigate the limited impact of explained variable brought by the positive integer nature of the number of enterprises to a certain extent.

  2. (2)

    Core explanatory variable. Following Zou and Deng (2022), we select five indicators from two levels to build a comprehensive digital economy indicator system. The indicators from the level of Internet development contain telecom business income, number of employees in information transmission, computer services and software industries, number of Internet broadband access users and number of mobile phone users. The inclusive finance index is selected to represent the level of digital financial inclusion. To reduce subjective factors, we adopt the relatively objective entropy weight method to obtain the comprehensive index of digital economy (\(digi\)).

  3. (3)

    Control variables. To obtain an unbiased estimation, following Liu et al. (2022) and Steijn et al. (2022), economic development level (\(lnpgdp\)), fixed asset scale (\(fix\)), degree of openness (\(fdi\)), education level (\(edu\)), degree of government intervention (\(gov\)) and financial development level (\(fina\)) are selected as the control variables. Economic development level (\(lnpgdp\)), is measured as the logarithm of per capita GDP; fixed asset scale (\(fix\)), is represented by the ratio of total fixed assets to GDP; degree of openness (\(fdi\)), is calculated as the ratio of foreign direct investment to GDP; education level (\(edu\)), is measured as the number of students in higher education per 100,000 population; degree of government intervention (\(gov\)), is expressed as the per capita government expenditure; financial development level (\(fina\)), is represented by the ratio of financial institutions’ deposit and loan balances to GDP at the end of the year. The calculation methods for all variables are shown in Table 1.

    Table 1 Calculation of variables.
    Full size table
  4. (4)

    Spatial weight matrix (\(W\)). Before empirical testing, the choice of the specific spatial model and the selection of the spatial weight matrix should be resolved. The spatial weight matrix shows the degree of spatial connectivity between cities. The estimated results may vary with different spatial weights. Generally, the binary matrix and geographic distance weight matrix are more common in prior relevant literature. The connections between different spatial units are often affected by geographical distance and economic disparity (Chen et al., 2022). Considering the weight matrix built only by geographical distance or economic disparity can’t accurately reflect the sophisticated connections between regions, we conduct a hybrid economic distance weight matrix as follows:

$$W=\left\{\begin{array}{l}(\overline{{Q}_{i}}\times \overline{{Q}_{j}})/{d}_{ij}^{2},i\ne j\\ 0,i=j\end{array}\right.$$
(2)

where \(i\) and \(j\) refer to regions. \(Q\) indicates the actual per capita GDP. \(d\) represents geographical distance.

Table 2 displays the descriptive statistics for pertinent variables. Among them, the mean value of \(agg\)(manufacturing enterprises agglomeration level) is 8.598, with the lowest (highest) values of 2.833 and 12.312, respectively. The corresponding standard deviation is 1.33, indicating that \(agg\) exhibits high volatility.

Table 2 Statistical description of variables.
Full size table

Results

Spatial autocorrelation test

The Moran’s I values, which range from -1 to 1, are used to quantify the degree of spatial self-correlation. Spatial clustering is indicated by a significantly positive value, and a significantly negative value shows spatial dispersion. A larger absolute Moran’s I value represents a more serious spatial agglomeration or dispersion. The absence of spatial dependence is demonstrated by a value of zero. The Moran’s I value is calculated as presented in Table 3. We observe that the Moran’s I index of manufacturing enterprises agglomeration level is always positive and significant at the level of 1%, manifesting that manufacturing enterprises agglomeration level exhibits significant positive spatial autocorrelation. We create Moran’s I value scatters plot of manufacturing enterprises agglomeration level in 2011, 2014, 2017, and 2020 for comparison, all of which are presented in Fig. 1. As depicted in Fig. 1, most cities lie in the first (high–high) and third (low–low) quadrants, indicating that the traditional econometric models ignoring spatial dependency may lead to biased estimators.

Table 3 Moran’s I spatial autocorrelation tests.
Full size table
Fig. 1
Fig. 1
Full size image

Moran’s I scatter chart of manufacturing enterprises agglomeration level.

The spatial distribution of Chinese manufacturing enterprises is shown in Fig. 2, from which we find that adjacent cities prefer showing a similar agglomeration level of manufacturing enterprises. In general, the agglomeration level of manufacturing enterprises in China is gradually reducing from south to north and east to west. China’s manufacturing enterprises are still mainly distributed in the southeast, which illustrates the southeast area has a leading economic and financial development level, education level and industrial support policies. These indicate that the distribution of manufacturing enterprises is characterized by spatial agglomeration. Therefore, it is indispensable to employ spatial econometrics to investigate the influence of the digital economy on manufacturing enterprises agglomeration level.

Fig. 2
Fig. 2
Full size image

Spatial distribution of manufacturing enterprises.

Regression results

The LM test is often used to decide the necessity of spatial econometrics models. Table 4 shows that the LM (robust) test results with spatial error and lag are both significant at the 1% level. Thus, it is necessary to employ spatial econometric models. In addition, we use the LR and Wald tests to check whether the SDM degrades to SAR or SEM. As can be seen that the \(p\) value of the Wald and LR test for spatial lag is 0.0000, which indicates that SDM is better than SAR. The spatial error test results show that SDM outperforms SEM. Thus, the SDM is considered more suitable than SAR and SEM. Furthermore, the Hausman test is used to decide the applicability of the fixed or random effect models. The value of the Hausman test statistic is 68.04, and the corresponding \(p\) value is 0.0000. Therefore, the SDM with fixed effect model is selected to conduct further study. Besides, we carry out the LR tests to examine whether the city-fixed effect or year-fixed effect is commonly insignificant. The results show that we should reject the city fixed effect (305.49, \(p\) < 0.01) and the year fixed effect (7798.60, \(p\) < 0.01). Overall, the above results suggest that we should adopt the SDM with city and year period fixed effect to conduct the research.

Table 4 Applicability tests of the spatial panel model.
Full size table

In line with LeSage and Pace (2014), we choose the maximum likelihood technique (MLE) to estimate model (1) since the use of OLS method may result in biased estimates because of the spatial lag of manufacturing enterprises agglomeration level. Table 5 displays the SDM results. We find that the spatial autocorrelation coefficient \(\rho\) is significant at 1% level, manifesting that manufacturing enterprises agglomeration level of different regions have a significant spatial dependence. Because SDM comprises the spatial lag terms of explanatory variables and explained variable, the explanatory variables exert an impact on both the local explained variables and the adjacent explained variables. Therefore, the regression coefficients of the SDM model can’t accurately depict the impact of every explanatory variable on the explained variable (Nan et al., 2022). Following LeSage and Pace (2014), we use partial differential to divide the SDM model’s total effects into direct and indirect effects. In the following analysis, we focus on the values of direct effects, indirect effects, and total effects.

Table 5 Results of Spatial Durbin model.
Full size table

The direct, indirect, and total effects of all explanatory variables are presented in Table 6. Generally speaking, a direct effect has two components. The first section denotes the direct impact of explanatory variables on the local explained variable, while the second section pertains to the feedback effects, which illustrate the repercussions of the local explanatory variables on the explanatory variables of neighboring regions and subsequently affect the local explanatory variables in return. Table 6 indicates that the direct effect estimation value of the digital economy is 0.0990, demonstrating statistically significant at the 1% level. This finding is comparable to the coefficients observed in the SDM of the digital economy. This regression results provide empirical support for hypothesis 1. The direct effect of digital economy is 0.0990, and the corresponding coefficient is 0.0559, so its feedback effect is 0.0431. It is obvious to deduce that the impact of local digital economy on the locally explained variable and the spatial feedback effect are both the major factors of the direct effect. The indirect effect means the effects of neighboring areas on the local explained variable, which is known as spatial spillover effects. More importantly, we observe that the indirect effect (6.196) of the digital economy is much larger than the direct effect (0.0990). We can conclude that promoting the development of local digital economy not only affects the local manufacturing enterprises agglomeration level, but also influences the manufacturing enterprises agglomeration level of adjacent regions. This indicates that the digital economy exhibits a significant spatial spillover effect, which validates hypothesis 3. Meanwhile, the spatial spillover effects are much bigger than the direct effects. The main reason is that the indirect effect denotes the total amount of spillovers in all neighboring areas, whereas the direct effects simply take into account the impact of local digital economy development on the manufacturing agglomeration level of local enterprises. In addition, the total effect is 6.295 and significant, denoting that a unit change in a city about the core explanatory variable could significantly increase 6.295 units of enterprises agglomeration across all the cities. The findings of this paper achieve theoretical deepening and paradigm breakthroughs while supporting the digital economy agglomeration effect, which supports the findings of the research conclusion of Wu and Li (2024). This study finds that under the vigorous development of the digital economy, enterprises are still further gathering, and it is too early to talk about the death of distance. On the one hand, the high fixed cost characteristics of digital infrastructure and data elements drive enterprises to concentrate in areas of computing power and data enrichment. On the other hand, in order to improve the resilience of the industrial chain and create digital agglomeration dividends, the digital economy drives material flow through data flow, promoting the manufacturing industry to cluster in the “digital physical” dual space.

Table 6 Direct, indirect and total effects.
Full size table

The results for other control variables are also informative. For economic development level, the direct effect coefficient is positive and significant, whereas the spillover effect coefficient is insignificant, implying that economic development level has an important and positive impact on the location selection of local enterprises, but has no influence on the adjacent areas. As for the fixed asset scale, the direct effect coefficient is negative and significant, and the spillover effect coefficient is insignificant, indicating that cities with excessive fixed asset investment crowd out other investments, which is unsuitable for enterprises aggregation. For the degree of openness, the direct effect and indirect effect coefficients are positive at the significant level of 5%, showing that opening to the outside world benefits enterprises by introducing foreign capital, technology and advanced management experience, improving enterprises’ value. Therefore, it is easier for domestic and foreign enterprises to gather in areas with a high level of opening-up. The direct effect and indirect effect coefficients of the financial development level are positive and significant, suggesting that a good financial environment is conducive to reducing transaction costs, easing financing constraints and improving the competitiveness of enterprises (Huang et al., 2025), thus attracting enterprise agglomeration. The direct effect and indirect effect of education level are insignificant, implying that the improvement of education level has not increased the degree of enterprises agglomeration. The direct effect coefficient of the degree of government intervention is statistically insignificant, whereas the indirect effect coefficient is significantly negative, which indicates that local government intervention crowds out enterprises’ investment, reduces investment efficiency (Liu et al., 2025), and decreases the degree of agglomeration of enterprises, consistent with the conclusion of Deng et al. (2020).

Robustness tests

Several robustness tests are carried out to check the accuracy of the above conclusions. First, we use the proxy indicators of manufacturing enterprises agglomeration level and digital economy to check whether the primary conclusions will change. Second, we employ an inverse distance weight matrix to confirm the stability of the benchmark regression results. Lastly, the potential endogeneity problem caused by omitted variables in model (1) is addressed by adding control variables.

Replacing the explained variable

The results about the impact of the digital economy on manufacturing enterprises agglomeration level may differ from different manufacturing enterprises agglomeration level indicators. Therefore, we consider alternative measures of manufacturing enterprises agglomeration level. For this purpose, following to Li et al. (2024), two methods are used to measure the level of manufacturing enterprise agglomeration, respectively. One is using the proportion of total industrial output value in each region (\(idr\)), and the other is measured by locational entropy (\(loe\)), as described in Eq. (3).

$$loe=\frac{in{d}_{i}/gd{p}_{i}}{ind/gdp}$$
(3)

Where, \(ind\) denotes total industrial output above scale, \(gdp\) is gross domestic product. The results in Table 7 are broadly accordance with the estimates of direct, spillover, and total effects in Table 6. In particular, the direct and indirect effects of the digital economy are positive and significant, illustrating that our findings are reliable.

Table 7 Robust test with the alternative explained variable.
Full size table

Replacing core explanatory variables

The logarithm of the word frequency of digital economy policies in prefecture-level cities (\(pdigi\)) is used to measure the digital economy and the effect of the digital economy on the agglomeration level of manufacturing enterprises are shown in Table 8. It can be seen that the coefficients of the direct effect, indirect effect and total effect of the digital economy are all positive. It shows that the findings of the paper are robust.

Table 8 Robust test with the alternative core explanatory variable.
Full size table

Replacing spatial weight matrix

Following You and Lv (2018), we perform an additional robustness test using an inverse distance weight matrix. Specifically, the weight satisfies \({w}_{ij}=1/{d}_{ij}\) when city \(i\) not belongs to city\(j\); otherwise, \({w}_{ij}=0\). Table 9 displays the regression results. Thus, applying various spatial weight matrices does not alter our primary conclusions. The above results indicate that adopting a different spatial weight matrix doesn’t alter our preliminary conclusions.

Table 9 Robust test using a different weight matrix.
Full size table

Adding control variables

The endogeneity problem caused by omitted variables may lead to biased estimation results. Therefore, industrial structure (\(stru\)) and infrastructure level (\(infr\)) are added to model (1) to test the impact of digital economy on the agglomeration level of manufacturing enterprises. The industrial structure is measured by the ratio of the added value of the tertiary industry to GDP; The level of infrastructure is measured by per capita road area. The regression results are shown in Table 10. It can be seen that the coefficients of the direct effect, indirect effect, and total effect of the digital economy are all significantly positive, indicating that the conclusions of our study are reliable.

Table 10 Robust test adding control variables.
Full size table

Heterogeneity analyses

Heterogeneity analyses of different enterprises’ scales

The digital economy is likely to exert heterogeneous effects on different-scale manufacturing enterprises. To compare the heterogeneous effects, we separate all manufacturing enterprises into two categories: large-scale and small-scale. Specifically, as stated in the “Regulations on the Classification Standards for Small and Medium sized Enterprises”, enterprises with a registered capital exceeding 5 million are classified as large-scale enterprises, otherwise, as small-scale enterprises. Table 11 displays the regression results of the digital economy on the manufacturing agglomeration level of different scales enterprises. The coefficients of the direct and indirect effects of the digital economy for large-scale and small-scale enterprises are positive and statistically significant at 1% significance level. However, it is remarkable that the direct and indirect effects coefficients of large-scale enterprises are greater than those of small-scale enterprises, suggesting that the digital economy appears to be more appealing to large-scale enterprises. From a direct effect perspective, the digital economy enhances enterprises’ division and collaboration efficiency, and greatly reduces the production costs of large-scale enterprises through the scale effect. The scale economies and network effects of the digital economy make it easier for large-scale enterprises to absorb and utilize the dividends brought by the digital economy during the agglomeration process. Therefore, the digital economy has a greater direct effect on the agglomeration of large-scale enterprises. From the perspective of indirect effects, data flow replaces traditional factor flow, and large-scale enterprises layout based on comparative advantages, promoting the formation of specialized clusters in neighboring areas. And the core areas of the digital economy attract leading enterprises with technological advantages, generating a “siphon effect” that radiates and drives the agglomeration of surrounding supporting large-scale enterprises. Therefore, the digital economy has a greater indirect effect on the agglomeration of large-scale enterprises.

Table 11 Heterogeneity analysis results of different enterprises scale.
Full size table

Heterogeneity analyses of different regions

To further study the heterogeneous influences of the digital economy on manufacturing enterprises agglomeration level in various areas, we divide all manufacturing enterprises into three areas including eastern, central and western, according to their registered address in China. The heterogeneous influences of the digital economy on manufacturing enterprises agglomeration level in different areas are displayed in Tables 12–14. As shown in Tables 12–14, the direct effect in the eastern region is positive and significant, whereas the indirect effect is positive and insignificant. Meanwhile, the direct effect in the western region is greater than that in the other two regions, whereas the indirect effect in the central region is the biggest. From the perspective of direct effects, the digital economy amplifies policy effects. With the implementation of the “East Calculation, West Calculation” strategy, data center industry chain enterprises are gradually migrating to the western region. Therefore, compared to the eastern and central regions, the increasing digital economy in western region has more fabulous attractions for enterprises. For the indirect effects, the central region, with its location hub, industrial adaptability, and policy leverage, has become an amplifier for the indirect effects of the digital economy. Therefore, the digital economy has a greater indirect effect on the agglomeration of enterprises in the central region.

Table 12 Heterogeneity analysis results of the eastern region.
Full size table
Table 13 Heterogeneity analysis results of the central region.
Full size table
Table 14 Heterogeneity analysis results of the western region.
Full size table

Heterogeneity analyses of whether to establish high-tech zones

High-tech zones are the core carriers of digital economy development, and their digital infrastructure construction is the key foundation to support technological innovation, industrial upgrading and ecological synergy. The impact of the digital economy on the agglomeration level of manufacturing enterprise may vary depending on whether cities have established high-tech zones. Therefore, based on whether a city has established a high-tech zone, the sample is divided into cities with high-tech zones and cities without high-tech zones. The regression results are display in Table 15. The regression results show that regardless of whether there is a high-tech zone or not, the digital economy has a positive promoting effect on the agglomeration level of manufacturing enterprises, while the digital economy has a greater direct and indirect effects on the former. On the one hand, it shows that the development of digital economy and high-tech zone presents a deep mutual promotion relationship. The digital economy injects innovative kinetic energy and optimizes the industrial structure of the high-tech zone, while the high-tech zone integrates upstream and downstream enterprises and promotes the agglomeration of manufacturing enterprises through digital technology. On the other hand, after the top enterprises settle in the high-tech zone, they drive upstream and downstream enterprises to gather in neighboring areas through technology spillover and order correlation.

Table 15 Heterogeneity analysis results of whether to establish high-tech zones.
Full size table

Mechanism analyses

As mentioned in previous sections, the digital economy has positive direct and indirect effects on manufacturing enterprises agglomeration level. To further explore the influencing mechanism, we draw lessons from Zhang et al. (2022) and build model to conduct mechanism analyses as follows.

$$mec{h}_{it}={\alpha }_{0}+\rho {W}_{ij}ag{g}_{jt}+{\alpha }_{1}dig{i}_{it}+{\alpha }_{2}{W}_{ij}dig{i}_{jt}+\mathop{\sum }\limits_{k=1}^{K}{\beta }_{k}{X}_{kit}+\mathop{\sum }\limits_{k=1}^{K}{W}_{ij}{X}_{kjt}{\theta }_{k}+{\varphi }_{i}+{\varphi }_{t}+{\varepsilon }_{it}$$
(4)

where, \(mec{h}_{it}\) denotes a set of variables that catch market potential, resource allocation or scale economies. The other variables have the same meanings as above.

Mechanism: effect of market potential

Table 16 reports the mechanism test results for increasing market potential. Referring to Ding and Niu (2019), the market potential is calculated by the logarithm of the total retail sales of consumer goods. Estimation results demonstrate that the direct and indirect effects coefficients of the digital economy on market potential are positive and significant. The results illustrate that the digital economy increases the agglomeration level of manufacturing enterprises by improving the market potential, which is consistent with the theoretical analysis mentioned above. It manifests that the digital economy breaks geographical boundaries, activates long tail markets, reduces information asymmetry, and unleashes potential demand. By radiating to neighboring areas through core cities, the digital economy enhances the market potential of both local and neighboring regions. As the market potential increases, many enterprises will be established and transferred, increasing the level of enterprise agglomeration.

Table 16 Mechanism: effect of market potential.
Full size table

Mechanism: effect of resource allocation

Table 17 reports the mechanism test results for optimizing resource allocation. Referring to Hsieh and Peter (2009), we measure resource allocation by calculating the reciprocal of factor price distortion in different regions. Specifically, we use formula (5)-(7) to measure factor price distortion.

$${\gamma }_{kit}=\frac{{k}_{it}}{{k}_{t}}/\frac{{s}_{it}{\beta }_{kit}}{{\beta }_{kt}}$$
(5)
$${\gamma }_{lit}=\frac{{l}_{it}}{{l}_{t}}/\frac{{s}_{it}{\beta }_{lit}}{{\beta }_{lt}}$$
(6)
$${\gamma }_{it}={\gamma }_{kit}\times {\gamma }_{lit}$$
(7)

where, \(\gamma\) is the factor price distortion coefficient. \(k\) stands for capital stock, measured by the scale of fixed assets in each region. \(l\) refers to labor input, calculated by the number of employees in each region. \({s}_{it}\) represents the output contribution degree of prefecture-level for city \(i\) in \(t\) period. \({\beta }_{kit}\) denotes the elasticity of capital output. \({\beta }_{lit}\) indicates the output elasticity of labor and \({\beta }_{lt}=\sum {s}_{it}{\beta }_{lit}\).

Table 17 Mechanism: effect of resource allocation.
Full size table

The results of Table 17 show that the direct effect coefficients of the digital economy on resource allocation are positive and significant, while the indirect effects coefficients are negative and significant. For the local area, the digital economy reduces enterprises’ search costs, enhances information transparency, makes more efficient resources flow to efficient industries, and orderly guides the rational allocation of resources. For neighboring regions, the siphon effect of the digital economy leads to efficient resource flow to the central city, reducing the efficiency of resource allocation in neighboring areas. A city with high resource allocation efficiency usually gathers high-quality elements, attracting enterprises to further gather. These results illustrate that the digital economy improves the agglomeration level of manufacturing enterprises by optimizing resource allocation.

Mechanism: effect of scale economies

Table 18 displays the estimation results of scale economies mechanisms. Referring to Wang et al. (2022), the economy of scale is measured by the ratio of main business revenue to main business cost of industrial enterprises above designated size. Estimation results demonstrate that the direct effect coefficient of the digital economy on scale economies is positive and significant, while the coefficient of indirect effect on scale economies is significantly negative. It indicates that the digital economy has exerted the effect of scale economies locally. With the accumulation of the number of users, data as well as technology, the unit cost of manufacturing enterprises has dropped significantly, further attracting the agglomeration of manufacturing enterprises. In addition, for neighboring regions, the siphon effect of the digital economy makes them lack core data resources and technological capabilities, making it difficult to leverage economies of scale. The results show that the digital economy increases the agglomeration level of local manufacturing enterprises through economies of scale effect, which is consistent with the theoretical analysis above. The conclusion of the above mechanism test validates hypothesis 2.

Table 18 Mechanism: effect of scale economies.
Full size table

Conclusions, policy implications, and limitations

Conclusions

With the booming information technology industry, the digital economy has incrementally developed into one of the primary elements influencing enterprises’ locations. Therefore, using the level of manufacturing agglomeration as a proxy variable for the enterprises’ geographical patterns, this paper applies the SDM to explore the influence of the digital economy on the geographical patterns of manufacturing enterprises. Empirical results show that the manufacturing enterprises agglomeration level and digital economy have evident spatial dependence. In general, the digital economy exhibits a positive spatial spillover effect, significantly facilitating manufacturing enterprises’ agglomeration in local and adjacent areas. The results have been subject to multifarious robustness tests, such as using alternative indicators for manufacturing enterprises agglomeration level and digital economy, different weight matrices, and the possible endogeneity. Results of scale heterogeneity regression results imply that the digital economy has a greater attraction to large-scale enterprises, and regional heterogeneity results reveal that the digital economy shows a greater direct agglomeration effect on enterprises in the western area and a greater indirect agglomeration effect on enterprises in the central region. The heterogeneity of high-tech zones indicates that the digital economy has a greater agglomeration effect on high-tech zones enterprises. Mechanism analyses reveal that the digital economy improves the spatial agglomeration degree of enterprises by increasing market potential, optimizing resource allocation and leveraging economies of scale. Our research conclusions demonstrate that the digital economy strengthens the imbalance of enterprise distribution and does not eliminate the impact of distance. It is too early to talk about the death of distance (Cairncross, 1997).

Policy implications

Some policy recommendations based on our findings are put forward. First, the government must strengthen the construction of new infrastructure to promote the development of the digital economy. Focusing on the core areas of digital infrastructure, such as broadband networks, 6 G base stations and data centers, the Government implement a plan for the coordinated deployment of “double gigabit” networks. In industrial parks, science and innovation corridors and other areas where enterprises are concentrated, priority should be given to the deployment of ultra-high-speed network systems. It is suggested to build a collaborative innovation mechanism of “government, industry, academia, research and finance”, set up a joint laboratory for digital economy, and comprehensively improve the data processing capability. Through these measures, a favorable digital ecological environment has been created for enterprise clustering.

Second, build a differentiated digital infrastructure gradient investment strategy. The digital economy has shown a huge agglomeration effect on manufacturing enterprises in central and western China, which is conducive to regional coordinated development. Therefore, the short board of network infrastructure construction in the central and western regions should be made up, and the investment in network infrastructure construction should be increased for the central and western regions and less developed regions. Set up a special R&D fund for digital technology in the central and western regions, focusing on supporting the research and development of special technologies such as plateau base stations and Gobi data centers. Building digital economy enclave parks in Xi’an and Chengdu to undertake digital service outsourcing business in the east, in order to promote the clustering of manufacturing enterprises in the central and western regions.

Finally, agglomerations foster differentiated development paths. The digital economy shows a greater agglomeration effect on large-scale enterprises, which is not conducive to the balanced development of the manufacturing industry. Therefore, in order to crack the pain points of SMEs such as difficult to obtain data elements, expensive to use, and slow to realize the value, the government should take the lead in building a regional manufacturing digital technology sharing platform, and promote large enterprises to open low-code tools, algorithm model libraries, and other resources to SMEs. Through institutional innovation and technology empowerment dual-wheel drive, to reduce the cost of SME data call. In addition, focusing on the core needs of small and medium-sized enterprises, through the integration of government, industry, academia and research resources, to create a public service hub for digital transformation, to enhance the core competitiveness of small-scale enterprises.

Limitations

Although this paper uses a variety of methods to conduct an in-depth analysis of the digital economy on the geographic pattern of manufacturing enterprises, there are still some directions that deserve further research. On the one hand, due to the large sample size, this article did not consider the distance between enterprises in measuring the explained variable. Future research can further calculate the spatial agglomeration of manufacturing industry based on the geographical coordinates of enterprises. On the other hand, we conducted multiple robustness tests to mitigate potential bias, however, some unavoidable factors still affect our empirical conclusions. Therefore, future research may find it worthwhile to explore alternative selective measurement methods to address potential endocrine issues.