Can data center green reform facilitate urban green technology innovation? Evidence from China

Abstract

With the expansion of the digital information industry, the size of global data centers is exploding. As part of a high-energy-consuming industry, data centers can support sustainable urban development through green transformation. This study uses a quasi-natural experiment of China’s national green data center pilot policy implemented in 2015 to examine the impact of data center green transformation on green technology innovation in cities. Using a difference-in-difference (DID) analysis, this study finds that the national green data center pilot policy leads to higher levels of green technology innovation in cities, but the spillover effect of the policy on neighboring cities is insignificant. The mechanism test showed that the national green data center pilot policy could facilitate urban green technological innovation through capital deepening, market competition, and industrial agglomeration. In addition, the policy impact is more pronounced for cities with poorer environments, less civilization, high levels of network infrastructure, and more innovative endowments. The findings can provide new guidance for cleaner regional production from the standpoint of data center green development.

Introduction

There have been world-renowned accomplishments in China’s economic performance in recent decades. However, because of the relatively crude development model of the Chinese economy over time, the resource and environmental problems generated by China are threatening the sustainable global economy (Li et al. 2021b). The government has repeatedly emphasized efforts to develop the digital economy and promote high-quality economic progress to realize a mutual benefit of energy and economic development. The digital economy, as a new form of economy, is innovative and contributes to industrial synergies (Zhang et al. 2022c), as well as improving economic and environmental performance (Wang et al. 2022b), and it is essential to the quality of the economy’s growth (Balogun et al. 2020).

Owing to the advancement of information technology and the outbreak of COVID-19, the global digital economy has taken on considerable proportions. The rapid development of the digital economy has led to the digitization of an entire spectrum of economic activities, which has led to an unprecedentedly large scale of data volume. The existing storage capacity and application systems of enterprises can hardly meet the needs of enterprise development. In this development context, building efficient and large-capacity data centers is urgent. Following Chen et al. (2014), Mahbod et al. (2022), and Saunavaara et al. (2022), a data center is a multifunctional organization that is responsible for stockpiling large-scale data while managing it for the purpose of amplifying its value, which makes more sense than just having a full-fledged website. Therefore, exploring the relationship between data centers and green economic growth is essential.

In 2020, the construction of data centers was officially included in the new infrastructure strategy1 by the central government. As the basis of new infrastructure, data centers have ushered in more excellent development opportunities and have achieved phased results. China’s data center industry market revenue reached 150 billion yuan in 2021, with a high compound annual growth rate of 30.69% over the previous three years. In contrast, the global data center market size exceeded $67.9 billion in 2021, up to 9.8% from 2020. However, as high energy-consuming organizations, data centers have become an obstacle to conserving energy and reducing emissions (Han et al. 2021; Ichinose et al. 2022). Power costs typically account for more than 50% of data centers’ operating costs. This means that the development of data centers generates large amounts of greenhouse gasses while consuming many electricity resources, threatening the sustainable development of society (Xu et al. 2022a). To take the path of sustainable development, enterprises have begun to pursue the green transformation of data centers. International internet giants actively engage in facilitating the use of green energy, energy conservation, and emission reduction in data centers. Google, Apple, and Facebook actively announced the progress of renewable energy use and achieved 100% renewable energy use in their operating systems in 2017, 2018, and 2020, respectively. Microsoft and Amazon plan to achieve 100% renewable energy use by 2025. The concept of a green data center has yet to continuously gain influence albeit at a gradual pace.

Unlike traditional data centers, green data centers emphasize efficiency and environmental protection. Internationally, the PUE (power usage effectiveness) value is usually used as a measure of the power utilization efficiency of a data center. Nevertheless, as the world’s second-largest economy, China lags far behind international advancements in terms of the PUE value of data centers (Li and Porter 2019). Moreover, the popularization of green data centers in China is still in its infancy, and there are still many problems in development. For example, a basic data monitoring mechanism for data center energy conservation and environmental protection has not been established, current standards and evaluation systems are not perfect, green technologies are unevenly popularized and applied, the energy management system is not sound enough, and the financing model needs to be innovated. Developing green data centers is fraught with difficulties. The research of Guo et al. (2021) noted that because of the cross-regional nature of data center operations and the time-sensitive nature of revenue and price of the grid, data center vendors face certain operational risks, making it difficult to obtain the maximum expected profit.

In response to the above problems, the Ministry of Industry and Information Technology (MIIT) and other departments jointly launched the national green data center pilot in 2015 and set up three batches of national green data centers in 2018, 2020, and 2022, covering almost all provinces in China. With the gradual advancement of the national green data center policy and local policies, the PUE of China’s data centers has improved significantly. According to data released by MIIT, the PUE of most data centers in China in 2015 was still generally greater than 2.2. By the end of 2017, the average PUE of the hyperscale and large data centers had dropped to 1.64 and 1.63, respectively, and the average design PUE for above-large data centers under construction had reached below 1.3 by 2022.

The healthy development of national green data centers will undoubtedly contribute to alleviating resource and environmental problems and promoting urban development. This article analyzes the impact of national green data center pilots on urban green technology innovation in China for three reasons.

First, with the information age approach, data centers are filling an increasingly crucial role in the economic growth of cities. There is no shortage of scholars who have examined in-depth studies on the theme of data centers, but unfortunately, most of these studies are in the fields of engineering or computer science, and these scholars have conducted in-depth discussions on data center chips, server technology architecture, and operational efficiency (Bast et al. 2022). Few scholars have focused on the social impact of data centers or conducted simple theoretical analyses rather than rigorous empirical examinations on the impact of data centers on real estate, urban economies, climate, and energy politics (Brodie and Velkova 2021; Pan 2022).

Second, in response to global environmental pollution and limited resources, data centers are transforming toward energy conservation and environmental protection (Jin et al. 2019; Wang et al. 2022c), which can be confirmed in engineering-related papers. The implementation of numerous policies has played a role in coordinating the order of market operation and capital support for the healthy development of related industries, ultimately leading to the emergence of new industries and technologies (Saunavaara et al. 2022). Shahbaz et al. (2022) identify the intermediary role of government governance in the mechanism of the digital economy promoting renewable energy transition. Prodi et al. (2022) state that a sound policy should ensure that the infrastructure is functioning properly while improving the energy efficiency of data centers. Similarly, in terms of data infrastructure, China has made considerable policy efforts in environmental protection, such as the development of the “Broadband China” strategy and the “Smart City” strategy, which have yielded good results. Studies have shown that these policies positively impact enterprise productivity, energy efficiency, and urban innovation (Chu et al. 2021; Dong et al. 2022; Tang et al. 2022; Zhang et al. 2022d).

Third, reviewing the articles on the factors influencing green technology innovation notes that environmental regulation has received more attention from researchers. Most academics have demonstrated that environmental regulation fosters the growth of green technology innovation (Cui et al. 2022; Liu et al. 2020). Furthermore, Ouyang et al. (2020) highlighted a U-shaped connection between technological innovations and environmental regulation. Zhang et al. (2022a) and Li et al. (2022b) argued that strict environmental policies can indeed spur companies to green innovation. Du et al. (2021) showed that environmental regulations are not beneficial to the upgrading of regional green technology innovation in the early stage of economic development. Analogously, the national green data center pilot policy can also be considered environmental regulation. According to the research conclusions of Brodie (2020) and Zhang et al. (2022e), the key to the transformation of data centers to achieve energy efficiency and environmental protection lies in the improvement in innovation capabilities and technological breakthroughs. However, it remains to be studied whether building green data centers can significantly raise the whole level of green innovation in cities.

This paper contributes to the extant literature in many aspects. First, this study presents the first empirical evidence on the social impact of data center industry development from the perspective of analysis in the social science field. As mentioned above, topics related to data centers are scarce in the social sciences, but with the booming digital economy, the role of data centers is becoming increasingly important. Thus, research on the data center industry should not only exist in the field of engineering but also be taken seriously by scholars in the field of social sciences. Second, this study examines the impact of data center greening reforms on urban green technology innovation for the first time. The study closely integrates the requirements for green economic development and focuses on the future direction of data center transformation to green. Third, the study uses an empirical analysis based on the national green data center pilot policy launched in China using the difference-in-differences model, which partially mitigates the endogeneity problem.

The structure of the paper is as follows. The research hypotheses are presented in Section 2. Section 3 provides a statistical description of the key variables and a summary of data center policies. In addition, the research design and data are introduced in Section 4, and the empirical analysis is provided in Section 5. The recommendations and conclusions are outlined in Section 6.

Policy background

With the rapid development of the digital economy in China, data centers, as the material carriers of information systems operating in various industries, have rapidly grown in scale year by year. It has become urgent to promote the transition of data centers to energy conservation and environmental protection. Over the past few years, central and local governments have developed many data center-related policies (Fig. 1).

Fig. 1.

National policies related to data centers

To accelerate the construction of a resource-efficient and environmentally friendly social development system, China launched the Twelfth Five-Year Plan for Industrial Energy Conservation in 2012, which explicitly states the importance of developing green data centers. Subsequently, guidance was provided on the construction and layout of data centers, as well as on technology application and management evaluation. However, due to problems such as slow progress in the green development of enterprise data centers, China began the deployment of pilot national green data centers in March 2015. The pilot construction is a new manifestation of energy-saving measures in the information age and is intended to respond to the call for global environmental protection. In May of the same year, the MIIT announced and identified 84 pilot national green data centers. In 2018, 2020, and 2021, three batches of national green data centers were officially established, totaling 153. Since then, China has been formulating policies to encourage the development of green data centers. For example, in July 2021, the Three-Year Action Plan for the Development of New Data Centers (2021–2023) was released, and in February 2022, the East Digital and Western Computing project was officially launched. In general, the Chinese government attaches great importance to developing green data centers. Constantly exploring green technology innovation, optimizing management and layout, and reducing the level of energy consumption of data centers are the current top priorities for developing the data center industry.

Literature review and theoretical hypothesis

The impact of national green data center pilots on urban green technology innovation

Because of the contradictory relationship between the economic interests of enterprises and the environmental interests of society, government control has a decisive effect on the level of green technology innovation of enterprises (Liu et al. 2020; Shang et al. 2022). The definition of a green data center given by Bird et al. (2014) points out that the construction of green data centers requires the guidance of professional engineers, manufacturers, and service providers before the development of future green data center technologies can proceed. This means that green data centers have higher requirements for green technology than traditional data centers. Additionally, many scholars have stated that high levels of ICT infrastructure are associated with high levels of innovation (Blyth 2020; Chandra et al. 2020; Lee et al. 2016). Furthermore, Tang et al. (2021) showed that telecommunication infrastructure can foster green technology innovation. Therefore, the following hypothesis for the research is presented.

• H1: The national green data center pilots can significantly promote urban green technology innovation.

An analysis of the mechanism by which national green data center pilots affect urban green technology innovation

Capital deepening mechanism

The data center industry is not only energy intensive but also technology intensive, especially in the case of green data centers; therefore, new and highly technical tools are constantly emerging in the data center green reform process (Li and Porter 2019). However, while companies deal with the high financial pressures associated with green technology innovation, they also undertake the risk of unknown time benefits. As a result, enterprises need significant funds to address market failure in green technology research and development (Zhao et al. 2021). Scholars have demonstrated the positive contribution of investment to corporate green technology innovation (Jiang and Bai 2022; Zhang et al. 2022b). Considering the current situation of international cooperation, the green investment market has huge potential and urgent demand and is an indispensable component of international economic and financial cooperation. The data center green reform is in line with the development trend of the times and is conducive to attracting foreign capital and expanding local investment. In addition, national policies, such as government subsidies, can address the lack of R&D investment and high R&D costs caused by externalities (Chang et al. 2022). Adequate R&D funds and special subsidies can stimulate green technology innovation of enterprises, alleviate the contradiction between the economic interests and environmental costs of enterprises, and form capital deepening. According to the National Green Data Center Pilot Work Plan issued by the MIIT, the national green data center pilots will receive policy and financial support in the construction of pilot-related projects to create a better investment climate for the pilots. Shen et al. (2019) showed that the field of cleaner production is a capital-intensive industry, which means that capital deepening can promote green technology innovation and thus boost the efficiency of enterprises’ green production. In light of the preceding analysis, the following hypothesis is proposed.

• H2: The national green data center pilots promote the green technology innovation of cities through a capital deepening mechanism.

Market competition mechanism

The construction of the national green data center pilot project is conducive to promoting market competition. On the one hand, the construction of the pilot project intensified the “survival of the fittest" effect in the local data and information industry, resulting in the continuous elimination of businesses that consume much energy and pollute (Shao et al. 2020). Ultimately, the entire industry has seen improvements in energy efficiency and green technology innovation. On the other hand, increasing attention is being paid to the green development of data centers (Huang et al. 2020). The development pattern of enterprises characterized by high energy consumption and pollution is the main obstacle to attracting foreign investment and achieving sustainable development (Lee et al. 2023). In an environment with fierce market competition, green development needs of low energy consumption and high environmental protection force enterprises to strengthen their green technology innovation skills (Ouyang et al. 2020). A comparative analysis of monopolistic and competitive markets by Hahn (2010) found that a competitive market environment leads to more innovative behavior, while monopolistic industries are discouraged. In a competitive market, the key to determining whether an industry innovates is whether the risk cost of being substituted is greater than the cost incurred by R&D (Shao et al. 2020). Furthermore, Song and Wang (2018) pointed out that the relationship between Chinese market competition and green technology innovation has passed the inflection point of the U-shaped curve, where the more competitive the market is, the better the progress of green technology. For these reasons, the following hypothesis is proposed in this study.

• H3: The national green data center pilots promote green technology innovation in cities through a market competition mechanism.

Industrial agglomeration mechanism

The construction of national green data center pilots is conducive to industrial agglomeration. Enterprises selected as pilots have a strong presence in the local data center industry. If companies can be selected as national green data center pilots, it implies that these enterprises are also models for regional green data center development. Due to policy support, the national green data center pilots rely on resources and other advantages to form an industrial agglomeration effect in the surrounding areas. In addition, industrial agglomeration usually has a spatial spillover effect (Han et al. 2018; Wang et al. 2022d), and geographical proximity allows companies to learn from each other’s green knowledge and technologies, increasing their enthusiasm to become involved in green innovation (Peng et al. 2021a). Industrial agglomeration can be divided into specialized agglomeration and diversified agglomeration models, depending on whether an internal enterprise belongs to the same industry (Han et al. 2018), while different types of agglomeration produce different innovation effects (Li et al. 2021a). Yang et al. (2022) claimed that industrial agglomeration and regional green innovation have an inverted U-shaped relationship, that is, green innovation is hindered by excessive industrial development. It is necessary to note that the green data center industry remains in the exploration stage, whether in China or internationally, and will not form a sufficiently high industrial agglomeration level in the short term. In light of the preceding analysis, this article proposes the following research hypothesis.

H4: The national green data center pilots promote green technology innovation in cities through an industrial agglomeration mechanism.

Research design

Sample selection

Although China’s national green data center pilots were established at the end of 2015, deployment started as early as the beginning of the year. In addition, to date, the MIIT and other entities officially established a total of 153 national green data centers covering 59 cities in three batches in 2018, 2019, and 2022. It is remarkable that the first batch of national green data centers, established in 2018, was all included in the pilot list in 2015. Therefore, to guarantee the reliability of the policy analysis results, the sample period of this paper is from 2011 to 2019. The treatment group is a total of 31 cities2 with national green data center pilots, and the control group is a total of 152 cities without national green data center pilots.

Model setting

Following Yuan et al. (2021), the effect of the national green data center pilots on urban green technology innovation is examined using a DID methodology. The specific regression model is as follows.

$${\mathit{agrivt}}_{\mathit{it}}={\alpha }_0+{\alpha }_1{\text{did}}_{\mathit{it}}+\beta {\text{control}}_{\mathit{it}}+{\sigma }_t+{\mu }_i+{\epsilon }_{\mathit{it}}$$ 1

where i is the city, t is the time, and agrivt_it represents green technology innovation. did_it is the core variable that this study focuses on. During the sample period, if the city has a national green data center pilot, then did_it = 1; otherwise, it is 0. control_it denotes control variables; ε_it is a random disturbance term; μ_i is an individual fixed effect; σ_t is a time fixed effect. The coefficient α₁ is the most critical estimated value in this study. If it is significantly positive, it means that the national green data center pilot policy helps drive the green technology innovation of cities.

When studying the innovation effect of the pilot construction of the national green data center, the possibility of spillover effects should be taken into account. This paper redefines the selection method of the experimental group to measure the spatial spillover of the policy effect. Neighboring cities are set as the treatment group where the policy occurs, and cities that exclude the policy occurrence area and neighboring areas are set as the control group. On this basis, the policy variable (didspill_it) is constructed to evaluate the policy spillover effect of the pilot construction of the national green data center.

Variable description

Explained variable: in the literature, R&D expenditure and patent counts in the environmental technology field are widely used to measure green technology innovations (Zhang and Li 2022). R&D expenditures are generally considered inputs to innovation activity, while patent counts are considered outputs. Compared with innovation input, innovation output can better reflect the innovation capability of a region. Therefore, this study uses the number of granted green patents3 to measure the level of regional green technology innovation as most studies do (Peng et al. 2021b; Wurlod and Noailly 2018). Specifically, in this study, agrivt and agrpra represent green invention patents and green utility patents, respectively. agrivt is used for the benchmark regression, and agrpra is used for the robustness analysis, in which green invention patents have the highest innovation value. In addition, the patent data are divided by the total population measured in 10,000 people.

A box plot of the variable agrivt is drawn for 283 cities in China from the period 2011–2019 to present the time series development trend of the data, as shown in Fig. 2. The graph shows that green technology innovation in all regions has long been characterized by low values. The development of urban green technology innovation has undergone two stages: the first stage (2011–2014) is a period of stable growth, during which the value of agrivt is basically “constant” across cities. The national average level of agrivt is relatively low and stable from year to year, and even cities with high levels of green innovation generally have values of agrivt below 10. The low value of agrivt during this period is perhaps because regions had neglected the environmental benefits of economic development before 2015. The second stage (2015–2019) is a period of rapid growth. During this period, the level of urban green technology innovation began to show a marked upward trend. This fast development trend is evident in both the numerical magnitude of the average growth rate and the size of the high-value cities. It is notable that during this period of rapid development of green technology innovation in cities, the national green data center pilot is also being implemented in parallel. This in turn provides a plausible rationale for this paper to explore the question of whether the establishment of national green data center pilots contributes to urban green technology innovation.

Fig. 2.

Green invention patent boxplot

Figure 3 shows the core density analysis of green technology innovation in cities in 2011, 2014, 2017, and 2019, from which the evolution of green technology innovation development in prefecture-level cities can be observed intuitively. In 2011, regional green technology innovation capabilities were inadequate, and cities with strong green technology innovation capabilities were concentrated in the urban circles centered on Beijing, Shanghai, and Guangzhou. In 2014 and 2017, the progress of green technology innovation in various regions accelerated significantly, with Guangdong Province leading the country in green technology innovation, and the trend of coordinated development in the Yangtze River Delta region was obvious. However, the development in the western and northern areas was still lagging. By 2019, green technology innovation was flourishing in the central and eastern regions, and the value of agrivt was significantly increased in the western and northern regions. From 2011 to 2019, the progress of green technology innovation in cities was very rapid. This may have been due to the increasing emphasis on green development in regions and the rising investment of enterprises and governments in green technology innovation.

Fig. 3.

Heat map of green technology innovation from 2011 to 2019

Core explanatory variable: variable did was derived based on the national green data center pilot list and the timing of the pilot list announcement.

Mediating variables: (1) capital deepening (cadeep): in real economic operations, capital deepening is mainly reflected in the increase in capital supply, bank credit expansion, the decrease in capital price, and the increase in capital stock per capita (Chen et al. 2020). In this paper, capital deepening is measured by the capital stock per capita. The stock of fixed assets, which is calculated using the perpetual inventory method, is used to measure capital stock. (2) Market competition (pete): privatization is one of the important parts of competitive markets, and state-owned enterprises can create higher economic efficiency through privatization (Ners 1995). Private companies can represent the economic dynamism of a region and are explorers of market development (Ramasamy et al. 2012). Therefore, the market competition degree of urban enterprises is measured by the number of private listed companies existing in the city in that year. (3) Industrial agglomeration (agg): location entropy, regional average concentration rate, and economic or employment density are often used to measure industrial agglomeration (Liu and Zhang 2021; Wang et al. 2022a; Wu and Lin 2021). Since the information industry is mainly included in the secondary and tertiary industries, this study uses the ratio of the sum of the output value of the secondary and tertiary industries to the urban construction land area to measure the degree of information industry agglomeration.

Control variables: (1) industrial structure advancement (indus): the upgrading of industrial structure will promote green development by improving energy efficiency and energy structure, while the improvement of technology level plays a key role in this impact (Su and Fan 2022). indus is expressed by the ratio of the added value of the tertiary industry to the added value of the secondary industry; (2) innovation vitality (inva): resource endowments can influence green innovation behavior (Qi et al. 2023). One of the most direct influences on a city’s ability to innovate in green technologies is the city’s innovation endowment. inva is measured by the ratio of the number of private enterprise employees and self-employed in cities to the total number of employees; (3) the level of urbanization (urb): Xu et al. (2022b) indicated a strong link exists between urbanization and green technology innovation. urb is measured by the proportion of the urban population to the total population; (4) population density (popd): Li et al. (2022a) showed that population density is positively associated with green innovation. popd is expressed by the proportion of the total urban population in the administrative area; (5) per capita science and education expenditure (edout): since green technology innovation is costly, the level of regional green technology innovation depends to a large extent on the scale of capital investment (Yumei et al. 2022). edout is measured by the government’s per capita scientific expenditure and education expenditure; (6) industrial pollution intensity (env): if enterprises cause serious pollution, they will face stricter environmental regulations, forcing them to carry out green technological innovation (Cai et al. 2020). env is measured in terms of industrial sulfur dioxide emissions per unit of GDP.

Data are acquired from the China Urban Statistical Yearbook, the statistical yearbooks of provinces or prefecture-level cities, and the CSMAR Database from 2011 to 2019. The descriptive statistics of the variables are shown in Table 1.

Table 1.

Descriptive statistics

Variable name	Obs.	Mean	S.D.	Min	Max
agrivt	2547	0.718	1.754	0	26.818
did	2547	0.061	0.239	0	1
cadeep	2547	1827.345	1356.126	112.241	10926.347
pete	2547	6.278	17.808	0	226
agg	2547	3078.771	7415.579	19.756	137000.000
indus	2547	0.982	0.546	0.114	5.168
inva	2547	0.489	0.158	0	0.945
urb	2547	54.127	14.801	0.355	118.840
popd	2547	438.002	345.903	5.093	2759.139
edout	2547	1820.098	1569.217	358.093	25802.068
env	2547	2.867	4.080	0.001	48.632
patent	2547	10.681	23.230	0.059	307.121

Empirical analysis

Analysis of benchmark regression and spatial spillover effects

To test Hypothesis 1, model 1 is empirically tested, and the results are shown in Table 2. The estimated value of the did coefficient decreases slightly after the addition of control variables and is significantly positive. Comparing column 2 and column 3, it can be found that the implementation of the national green data center pilot policy promotes green invention patents more than green utility patents. The results show that the national green data center pilot policy can effectively improve the level of urban green technology innovation, with green invention patents serving as the primary focus, which is consistent with H1.

Table 2.

Results of benchmark regression and spatial spillover effects

	agrivt	agrivt	agrpra	agrivt	agrivt
	(1)	(2)	(3)	(4)	(5)
did	1.9266*** (0.4996)	0.6365*** (0.1818)	0.6184*** (0.2144)
didspill				0.1576 (0.1194)	0.0705 (0.0686)
indus		0.1329 (0.0980)	-0.1891** (0.0828)		0.0211 (0.0550)
inva		0.0865 (0.1113)	0.1332 (0.0884)		0.0335 (0.0869)
urb		-0.0148** (0.0068)	-0.0085 (0.0060)		-0.0091* (0.0050)
popd		0.0005 (0.0009)	0.0011 (0.0007)		-0.0001 (0.0007)
edout		0.0009*** (0.0001)	0.0005*** (0.0001)		0.0006*** (0.0002)
env		0.0256*** (0.0062)	0.0117* (0.0061)		0.0198*** (0.0054)
patent		0.0324*** (0.0072)	0.0457*** (0.0079)		0.0325*** (0.0082)
constant	0.2599*** (0.0476)	-0.6892 (0.4245)	-0.5935* (0.3401)	0.1354*** (0.0310)	-0.2497 (0.4323)
Time effects	Yes	Yes	Yes	Yes	Yes
City effects	Yes	Yes	Yes	Yes	Yes
N	2547	2547	2547	2268	2268
R2	0.8319	0.9386	0.9466	0.7290	0.8305

(1) ^⁎⁎⁎, ^⁎⁎, and ^⁎ represent significance levels of 1%, 5%, and 10%, respectively. (2) Values in parentheses are standard errors

Columns 4 and 5 of Table 2. indicate that the pilot construction of the national green data center does not have significant spillover effects on surrounding cities. This is perhaps because a city’s social environment and development model determine the level of green technology innovation, and the impact of green technology innovation produced by a city that can be effectively absorbed by surrounding cities is relatively limited.

Robustness tests

Through the above analysis, it can be found that the pilot construction of national green data centers can significantly contribute to the innovation of urban green technology, but the robustness of the conclusion remains to be investigated. Therefore, to confirm the validity of the empirical findings of this study, the tests described below are carried out.

Parallel trend assumption

For reliable conclusions to be drawn using the DID approach in policy analysis, it is necessary for the model to satisfy the parallel trend assumption. Following Jacobson et al. (1993) and Bertrand et al. (2004), the result of the test using the event study method is shown in Fig. 4. The coefficient is not significant before 2015, but after 2016, it is significantly positive. This indicates little evidence for a pre-change trend in did. The graph also shows that the impact of the national green data center pilot policy on urban green technology innovation is evident, with the greatest impact in the first year of implementation. In addition, the result also indicates that the policy effects of the national green data center pilot flattened after the second year of implementation. The complexity of the factors influencing the green transformation of data centers and the fact that the green reform of data centers cannot rely solely on government control may account for this outcome.

Fig. 4.

Parallel trend assumption

Excluding other policy influences

Urban green technology innovation may be impacted by more than one policy over the sample period examined in this paper. According to the research of Tang et al. (2021), the pilot of “Broadband China” can foster urban green technology innovation. In addition, 2013 was a critical time when China began to promulgate many environmental protection policies (such as the Air Pollution Prevention Action Plan), which may interfere with the research conclusions of this paper. Two policy dummy variables, did _ gr and did _ wb, are added to the benchmark regression model to test whether the findings still hold after excluding the influence of other policies. As shown in Table 3, the regression coefficient between did and agrivt is significantly positive, that is, the research conclusion of this study remains valid after excluding the interference of other policies.

Table 3.

Excluding other policy influences

	Broadband China policy	2013 and beyond	Combined effect
	(1)	(2)	(3)
did	0.6103*** (0.1826)	1.0350*** (0.2635)	0.9826*** (0.2593)
did _ gr		0.4490*** (0.1521)	0.3815*** (0.1402)
did _ wb	0.1367* (0.0734)		0.2232** (0.0961)
constant	-0.5488 (0.4546)	-1.3066 (0.8553)	-1.2889 (0.8529)
controls	Yes	Yes	Yes
Time effects	Yes	Yes	Yes
City effects	Yes	Yes	Yes
N	2547	2547	2547
R2	0.9387	0.9394	0.9466

(1) ^⁎⁎⁎, ^⁎⁎, and ^⁎ represent significance levels of 1%, 5%, and 10%, respectively. (2) Values in parentheses are standard errors

PSM-DID estimation

In 2015, the work plan for the national green data center pilot issued by the MIIT pointed out the assessment and application conditions of the pilots. Therefore, the establishment of the national green data center pilot in China is not a random process, and there are some selectivity biases, which can easily lead to errors in the estimation results. In view of this, the propensity score matching difference-in-differences method (PSM-DID) is adopted to estimate the model. Urban per capita GDP (pgdp), population density (popd), per capita expenditure on science and education (edout), industrial pollution intensity (env), industrial structure upgrading (indus), and patents per 10,000 people (patent) are selected as matching characteristic variables. The matching variables are used to predict the probability of cities being selected to pilot national green data centers. On this basis, the nearest neighbor matching method is used to match the control group of cities that set up pilot national green data centers. This ensures that the treatment and control groups can pass the parallel trend test to the maximum extent possible. The matched data are re-estimated with the difference-in-differences method, and the estimation results are shown in column 1 in Table 4. The result indicates that the driving effect of the national green data center pilot policy on urban green technology innovation remains robust after matching.

Table 4.

Other robustness test results

	PSM–DID	Change in treatment group selection method	City × time fixed effects	Tail 1%
	(1)	(2)	(3)	(4)
did	0.6365*** (0.1818)		0.6199*** (0.1742)	0.5232*** (0.1725)
did _ 1		0.8555*** (0.3287)
constant	-0.6892 (0.4245)	-0.6235 (0.4309)	158.3868*** (42.8203)	0.5858 (0.4540)
controls	Yes	Yes	Yes	Yes
Time effects	Yes	Yes	Yes	Yes
City effects	Yes	Yes	Yes	Yes
N	2547	2448	2547	2269
R2	0.9386	0.9394	0.9403	0.8909

(1) ^⁎⁎⁎, ^⁎⁎, and ^⁎ represent significance levels of 1%, 5%, and 10%, respectively. (2) Values in parentheses are standard errors

Placebo test

To prevent other factors from affecting the robustness of the conclusions, this paper takes a cue from the research method of Moser and Voena (2012) to construct counterfactual policy dummy variables to re-estimate the benchmark regression by DID. If the estimated value of the counterfactual policy dummy variable is significantly 0, it can indirectly indicate that the findings of this study are reliable. In this paper, the random sampling process of the placebo test is repeated 500 times, and the probability density distribution of the coefficient estimates of the counterfactual policy dummy variables obtained is shown in Fig. 5. As shown in Fig. 5, the coefficient estimates of the randomly assigned policy dummy variables are concentrated approximately around 0, which indicates that the randomly set up national green data center pilot has no promoting impact on the urban green technology innovation level. In conclusion, the positive impact of setting up a national green data center pilot on urban green technology innovation is not disturbed by unobservable omitted variables.

Fig. 5.

Nuclear density curve

Change the selection method of the treatment group

To prevent the bias of the treatment group selection method on the policy analysis results, the treatment group selection method is changed in this study. In the sample period of this paper, the Chinese government formally established the first batch of national green data centers in 2018. Therefore, to minimize the error of policy analysis, this paper redefines the treatment group based on pilot cities, excluding cities that were not among the first batch of national green data centers. The estimated results, which can be found in column 2 of Table 4, demonstrate that the new estimation is extremely compatible with the outcomes of the benchmark regression. This further confirms that the conclusion is robust.

Other inspection methods

Different dynamics of policy adjustments by local governments in different cities may also impact the development of urban green technology innovation, leading to model estimation bias. To this end, column 3 in Table 4 further controls the city × time fixed effect and finds that the estimation results remain robust. In addition, the benchmark model is regressed by reducing the sample tail by 1% and obtains column 4 in Table 4. The regression results indicate that the original conclusion does not change significantly.

Influence mechanism test heterogeneity analysis

Analysis of the impact mechanism

To analyze the mediating mechanism of the pilot policy, according to the research of Feng et al. (2022), this study builds on model 1 to develop the following models.

$${\text{media}}_{\mathit{it}}={\gamma }_0+{\gamma }_1{\text{did}}_{\mathit{it}}+\phi {\text{control}}_{\mathit{it}}+{\sigma }_t+{\mu }_i+{\epsilon }_{\mathit{it}}$$ 2

$${\mathit{agrivt}}_{\mathit{it}}={\delta }_0+{\delta }_1{\text{did}}_{\mathit{it}}+{\delta }_2{\text{media}}_{\mathit{it}}+\vartheta {\text{control}}_{\mathit{it}}+{\sigma }_t+{\mu }_i+{\epsilon }_{\mathit{it}}$$ 3

where media_it represents the mediating variable (cadeep; pete; agg). The results of the mediation effect test, shown in Table 5, show that through capital deepening, market competition, and industrial agglomeration, the construction of a national green data center pilot can boost urban green technology innovation. On the one hand, at least at the significance level of 10%, the construction of the national green data center pilot has a positive impact on capital per capita, the number of private listed companies, and the economic density of the secondary and tertiary industries. This shows that the establishment of the national green data center pilot is advantageous for increasing the industrial agglomeration of data centers, enhancing capital deepening, and encouraging market competition. On the other hand, at least at the significance level of 1%, capital deepening4, market competition, and industrial agglomeration all exert a significant positive effect on urban green technology innovation capacity, demonstrating the existence of the mediating effect. These results are consistent with H2, H3, and H4.

Table 5.

Mediation test results

	cadeep	agrivt	pete	Agrivt	agg	agrivt
	(1)	(2)	(3)	(4)	(5)	(6)
did	208.1518* (120.1987)	0.6123*** (0.1862)	3.7084** (1.8279)	0.4172*** (0.1422)	770.6082* (399.8310)	0.5572*** (0.1667)
cadeep		0.0001* (0.0001)
pete				0.0592*** (0.0184)
agg						0.0001*** (0.0000)
controls	Yes	Yes	Yes	Yes	Yes	Yes
Time effects	Yes	Yes	Yes	Yes	Yes	Yes
City effects	Yes	Yes	Yes	Yes	Yes	Yes
N	2547	2547	2547	2547	2547	2547
R2	0.9289	0.9391	0.9713	0.9489	0.9871	0.9410

(1) ^⁎⁎⁎, ^⁎⁎, and ^⁎ represent significance levels of 1%, 5%, and 10%, respectively. (2) Values in parentheses are standard errors

Heterogeneity analysis

The mechanism by which the establishment of national green data center pilots promotes urban green technology innovation is verified above from both the theoretical and empirical perspectives. However, for cities with different characteristics, is the establishment of national green data center pilots still crucial for promoting urban green technology innovation? What differences are there? To answer these questions, this study discusses the heterogeneity of the effect of national green data center pilot construction on urban green technology innovation in four aspects: urban environment, comprehensive level of urban civilization, network infrastructure, and innovation endowment.

Heterogeneity of environmental conditions

The main purpose of the pilot construction of national green data centers is to respond to the global trend of green development. It is also an outstanding measure to cope with the power crisis and to improve the operational efficiency of the digital industry. To summarize, the policy was formulated to improve environmental and resource issues. Then, it remains to be seen whether the impact of the national green data center pilot policy on green technology innovation varies across cities with different environmental conditions. This paper selects urban CO₂ emissions (the CO₂ data come from China Emission Accounts and Datasets) as an indicator representing the urban environmental conditions. A dummy variable (CO₂) is set based on the average value for heterogeneity analysis, assigning cities with high CO₂ emissions the value 1 and cities with low CO₂ emissions the value 0. The analysis results are shown in columns 1 and 2 in Table 6. The findings show that the policy effect of cities with high CO₂ emissions is greater, and the significance level is higher than those of cities with low CO₂ emissions.

Table 6.

Environmental heterogeneity

	High CO2 emission cities	Low CO2 emission city	National civilized city	Non-national civilized city
	(1)	(2)	(3)	(4)
did	0.5212*** (0.1955)	0.2565* (0.1491)	0.3095 (0.2836)	0.6759*** (0.2336)
controls	Yes	Yes	Yes	Yes
Time effects	Yes	Yes	Yes	Yes
City effects	Yes	Yes	Yes	Yes
N	1278	1269	360	2187
R2	0.9509	0.8727	0.9633	0.8930

(1) ^⁎⁎⁎, ^⁎⁎, and ^⁎ represent significance levels of 1%, 5%, and 10%, respectively. (2) Values in parentheses are standard errors

In conclusion, the national green data center pilot policy has been more effective in promoting green technology innovation in cities with poor environmental conditions. This is perhaps because cities with poor environmental conditions lag in green development concepts and green technologies. The construction of a national green data center not only plays a leading role in policy but also provides financial support for enterprises to develop green technology. This policy effect is magnified in cities with relatively poor green technologies. Additionally, there are many key factors influencing green technology innovation in cities with good environmental conditions, which may obscure the green innovation effect of the pilot policy.

National civilized cities

To explore whether the difference in the degree of urban civilization influences the policy effect of the national green data center pilot policy, this paper uses the 2015 national list of civilized cities to conduct heterogeneity analysis. Among all the city brands in China, the national civilized city is a comprehensive honorary title indicating the whole civilization degree of a city. Therefore, using the national civilized city as the standard for measuring urban civilization is reasonable. A dummy variable (civi) is set, and cities on the national civilized city list are assigned a value of 1; otherwise, 0. The findings are depicted in columns 3 and 4 of Table 6. It can be concluded that the pilot construction of the national green data center can achieve a more significant innovation driving effect in non-national civilized cities. This is perhaps because non-national civilized cities tend to pay increased attention to economic development, and the national green data center pilot policy shifts the focus of local governments to green development. Furthermore, the national green data center pilots tend to be among the most influential enterprises in China, and such regional champions play an exemplary role in leading the demonstration.

Network infrastructure level heterogeneity

The ultimate goal of launching the national green data center pilot is to develop the green digital information industry. Therefore, the level of the city’s network infrastructure largely determines the impact of this policy on the development of urban green technology. The higher the network infrastructure level is, the faster the city’s digital economy develops, and the higher the efficiency of information integration into urban development. Accordingly, this study takes the number of broadband internet access users per 10,000 people as a measure of the urban network infrastructure level and further studies the policy heterogeneity of the level of urban network infrastructure. A dummy variable (net) is set, where 1 means that a city has a high level of network infrastructure, and 0 indicates a low level of network infrastructure. Columns 1 and 2 in Table 7 show that the pilot policy promotes green technology innovation significantly in cities with a high level of network infrastructure but not significantly in cities with a low level of network infrastructure, confirming the above analysis conclusion.

Table 7.

Other heterogeneity test results

	Cities with a high level of network infrastructure	Cities with low level of network infrastructure	High innovation endowment	Low innovation endowment
	(1)	(2)	(3)	(4)
did	0.6860*** (0.2373)	0.0166 (0.0399)	0.7163** (0.3006)	0.2302** (0.0967)
controls	Yes	Yes	Yes	Yes
Time effects	Yes	Yes	Yes	Yes
City effects	Yes	Yes	Yes	Yes
N	873	1674	550	1997
R2	0.9442	0.6853	0.9413	0.8425

(1) ^⁎⁎⁎, ^⁎⁎, and ^⁎ represent significance levels of 1%, 5%, and 10%, respectively. (2) Values in parentheses are standard errors

Innovation endowment heterogeneity

Although the national green data center pilot policy implementation can affect the urban green technology innovation level, it must be based on the city’s own innovation endowment. In other words, a city must have a certain innovation ability, and the stronger its innovation ability is, the richer the green technology innovation results. Therefore, to verify this idea, this study takes the number of patent authorizations per 10,000 people as an indicator to measure the urban innovation endowment. A dummy variable (pat) is set, where 1 indicates cities with a high innovation endowment, and 0 indicates cities with a low innovation endowment. The findings are displayed in columns 3 and 4 of Table 7. It can be concluded that the pilot establishment of national green data centers has a promoting effect on cities with different innovation endowments, and they are all significant at the 5% significance level. Moreover, the policy has a stronger impact on cities with high innovation endowment than on cities with low innovation endowment.

Conclusions and implications

Data centers are the carriers of the digital economy and are constantly growing in size. Meanwhile, the energy consumption and carbon emissions associated with data centers are increasing rapidly because of the large amount of storage equipment, backup equipment, cooling systems, and other infrastructure required to operate a data center. As China strives to achieve carbon reduction, data centers are facing severe challenges to elevate energy efficiency and reduce greenhouse gas emissions. Special attention must be given to understanding how to bring about high-quality growth in the data center industry.

Therefore, a DID model is employed to verify the impacts of the national green data center pilot policy on urban green technology innovation and to evaluate the action mechanism.

The empirical analysis results show that China’s national green data center pilot policy significantly contributes to the level of urban green technology innovation. The effect is strongest in the first year after the policy is implemented. Moreover, the pilot policy has a greater impact on urban invention-based innovation than on utility innovation. The spillover effect analysis reveals that green technology innovation in the areas surrounding the pilot cities is not affected by the national green data center pilot policy. Furthermore, the analysis of mediating effects shows that the pilot establishment of the national green data center promotes urban green technology innovation through capital deepening, market competition and industrial agglomeration. In addition, the policy effects of the national green data center pilot policy differ across types of cities. Specifically, the pilot establishment of national green data centers has a stronger impact on boosting green technology innovation in cities with a poor environment, low level of civilization, high level of network infrastructure, and high innovation endowment.

To speed up the expansion of green data centers and promote green growth in countries, the following policy recommendations are made based on the research results.

First, local governments should actively guide the data center industry’s green transition through financial support and tax relief. When building large-scale data centers, green and energy-saving technologies should be actively introduced, because the high-quality development of green data centers can drive up green technology innovation in cities, which contributes to the city’s low-carbon development. Additionally, the government should encourage the development of the data center industry chain and fully use the development advantages of industrial agglomeration. Simultaneously, local governments should formulate appropriate policies to ensure good competition in the market.

Second, the heterogeneity of the urban network infrastructure indicates that cities with a higher network infrastructure level are better equipped to develop green data centers. This means that the development of the green information industry needs to be based on the traditional information industry, and green development is a higher requirement for enterprise development. The government should perfect urban network infrastructure, and enterprises should reasonably introduce digital infrastructure in accordance with their own development goals.

Third, the heterogeneity of urban innovation endowment shows that urban innovation ability is crucial to the urban green technology innovation level. The higher a city’s innovation endowment is, the more likely it is to produce high-quality innovations with a “green” nature. On the one hand, local governments should increase incentives for green innovation and tailor talent introduction policies to local circumstances. On the other hand, instead of focusing on benefits and investments in the short term, businesses should prioritize long-term green development plans, considering that both the environmental and economic benefits of the development model are the ultimate development direction for excellent enterprises.

For future research, this study opens up a number of intriguing new vistas. Although the development of data centers has attracted increasing attention in society, there are too few empirical studies on data centers. In addition, most of the policies related to the development of green data centers have been introduced recently, which is not conducive to policy evaluation. Therefore, scholars can focus more on the development of the data center industry from a social sciences perspective in the future, taking into account issues such as enterprise transformation and social welfare.

Author contribution

Yuanhong Liu conceived the topic, combed the relevant literature, and wrote the first draft. Zhihui Yang collected the data and corrected this paper.

Funding

This research was supported by the National Natural Science Foundation of China (Grant No. 71974001), the Major Research Projects of Natural Science in Universities of Anhui Province of China (Grant No. KJ2017ZD35), and the Top Talent Project in Anhui Province (No. gxyqZD2020087).

Data availability

Data and materials are available from the authors upon request.

Declarations

Ethics approval

Not applicable.

Consent to participate

We confirm that the manuscript has been read and approved by all named authors and that there are no other persons who satisfied the criteria for authorship but are not listed. We further confirm that the order of authors listed in the manuscript has been approved by all of us.

Consent for publication

The authors agree to publish this article in the Environmental Science and Pollution Research.

Competing interests

The authors declare no competing interests.