Greening China’s digital economy: exploring the contribution of the East–West Computing Resources Transmission Project to CO2 reduction

We employed both a scenarios approach and fsQCA to estimate the impact of digital infrastructure on carbon emissions in China. While the scenarios approach involves a comprehensive exploration of key factors (e.g., PUE, green electricity, etc.) that influence carbon emissions across the entire EWCRT Project, the fsQCA focuses on the configurational relationships of these key factors by undertaking a review of 10 data center clusters as cases. By integrating these two methods, we aim to provide a more comprehensive understanding of how different conditions, such as the amount of green energy and the number of racks, would coexist in different scenarios within China’s EWCRT Project, as well as how the different conditions would lead to distinct carbon emission reduction outcomes.

Scenarios approach

We used the scenarios approach to estimate the carbon emissions using the low emissions analysis platform (LEAP) software (version 2020.1.0.64), as proposed by Heaps (2022). The data generated by our analysis can be found in supplemental material (see Data S1–S8).

Step 1: The bottom-up method

Following prior studies (Ziegler et al. 2019; Saunois et al. 2020; Greenpeace 2021), we adopted a bottom-up method to estimate the carbon emissions levels from the data centers required for each scenario using the following equations:

where E_i denotes the total electricity consumption of the data centers; N_i is the number of racks; P_i is the purposely designed power of a single rack; L_i is the average IT load utilization rate (i.e., the ratio of the actual load of the IT equipment to the purposely designed full load of the IT equipment); U_i is the average PUE (determined by dividing the total energy consumption of a data center by the total energy consumption of its IT devices); t is the working hours; G_i is the proportion of green electricity; and F_i denotes the emissions factors (based on those of the local or national grids where the data centers are located). For further details, see Data S3, S6, and S8.

We analyzed the eight national-level data center hubs included in the EWCRT Project. The four data center hubs in China’s western region are the Inner Mongolia Hub (including the Helingel Cluster), Gansu Hub (including the Qingyang Cluster), Ningxia Hub (including the Zhongwei Cluster), and Guizhou Hub (including the Guian Cluster). The four data center hubs in China’s eastern region are the Beijing-Tianjin-Hebei Hub (including the Zhangjiakou Cluster), Yangtze River Delta Hub (including the Wuhu Cluster and Demonstration Zone of the Yangtze River Delta Cluster), Greater Bay Hub (including the Shaoguan Cluster), and Chengdu-Chongqing Hub (including the Tianfu Cluster and Chongqing Cluster). According to policy requirements, the Chengdu-Chongqing Hub is considered one of the four hubs in the east despite its actual geographical location, which is in the western part of the country.

Step 2: Scenarios and data sources

To explore the EWCRT Project’s contribution to the greening of China’s digital economy, we established three scenarios: a business as usual (BAU) scenario, an EWCRT Project (PRO) scenario, and an advanced improvements (ADV) scenario with additional greening measures. Using available data, we assumed each parameter of the BAU and PRO scenario. The data sources are found in Table 2.

Data center hub managers often consider a mix of actions to reduce CO₂ emissions, such as simultaneously improving PUE and increasing the proportion of green electricity. We assumed that the data center hubs could combine PUE improvement with green electricity use in the ADV scenario: (i) PUE for all data center hubs would fall to 1.1 after 2030 because a PUE value close to 1.1 would be the ultimate improvement, demonstrating true technological progress (Masanet et al. 2020). (ii) The proportion of green electricity in China’s eastern and western hubs would gradually increase to 100% (up from 20% initially) by 2040. Based on China’s net-zero carbon emissions policy, all digital infrastructure is to be fully powered by clean energy in the future (NEA 2021; Qiu et al. 2021); moreover, some data center hubs in China have committed to increasing their use of green electricity to 100% by 2030. Otherwise, the values of all of the other parameters (i.e., the number and distribution of racks, the purposely designed power of a single rack, the average IT load utilization rate, the working hours per rack/year, and the average CO₂ emissions factor) are the same as those in the PRO scenario (see Table 3).

Step 3: Calculation of CO₂ reductions in the different scenarios

Following previous studies (Liang et al. 2019; Ouedraogo 2017), we used the LEAP model to explore the optimum decarbonization path for China’s data centers under the EWCRT Project (Heaps 2022), as well as to generate the predictive values of the number of racks in the BAU and PRO scenarios (see Data S2 and S3). Then, by taking 2020 as the baseline year, we predicted the parameters of the other scenarios (i.e., the P1 scenario, P2 scenario, G1 scenario, G2 scenario, M1 scenario, and ADV scenario) and calculated their respective carbon emissions between the years 2020 and 2050.

fsQCA

We used both fsQCA and an empirical method based on Boolean algebra to analyze our cases, as such methods allow researchers to analyze combinations of several features, and because they also allow one to theorize the configuration of patterns between cases by identifying similarities and differences (Ragin 2006; Xie and Wang 2020). For example, in previous work, one QCA study used 10 cases from Africa, Asia, and Latin America to find crucial preconditions for community forest management performance and to provide a theoretical lens different from most community forest management research (Arts and de Koning 2017). The scenario analysis focused on the trend of the data centers’ carbon emissions over the long run, while the fsQCA results provided details on the low-carbon configurations in different data centers that have been overlooked in previous analyses. We thus adopted the fsQCA methodology to analyze the configuration of patterns between the 10 data center clusters included in the EWCRT Project. This framework comprised an outcome variable and five conditions, as discussed below.

Outcome

We used two sub-variables to measure the carbon emissions intensity of the digital economy at the regional level. The present variable (Carbon Emissions Intensity) means that the digital economy produces more carbon dioxide per unit; otherwise, the absent variable (~Carbon Emissions Intensity) means that the digital economy produces less carbon dioxide per unit.

Conditions

Prior studies related to digital infrastructure, the digital economy, and low-carbon development have discussed various financial, institutional, digital, and energy conditions (Fedorowicz et al. 2018; Manny et al. 2021). Following this literature, we selected five representative conditions to capture the antecedents of a low-carbon digital economy, as follows. (1) The amount of public green investment at the regional level (Green Investment) is an important factor for strengthening pollution control and improving the environmental effects of the digital economy (Ding et al. 2023). (2) The number of racks in data centers at the regional level (Racks) is a key underlying parameter in prevailing data center energy models and is used for estimating carbon emissions (Lei and Masanet 2020). (3) The proportion of green electricity consumption at the regional level (Green Energy) can reduce data centers’ energy costs, as well as their carbon emissions (Masanet et al. 2020; Ziegler et al. 2019). (4) The regional government’s attention to the environment (Green Attention) influences the low-carbon transition of most industries, including data centers and related enterprises (Tang et al. 2023). (5) The digitalization index score at the city level (Digital Level) reflects the quality of digital infrastructure in a given city (e.g., the pilot cities of the “Broadband China” strategy have better digital infrastructure than non-pilot cities) (Feng et al. 2023). Table 4 shows the measurement of all variables. Based on the theoretical framework, our fsQCA procedure involved three key steps.

Step 1: Calibration

Following prior literature (Howell et al. 2022; Jia et al. 2023), we employed the direct calibration method to transform the continuous quantitative data to fuzzy-set memberships based on three qualitative thresholds: “fully in” (i.e., more than the 75th percentile), “crossover point” (i.e., the 50th percentile), and “fully out” (i.e., less than the 25th percentile). For example, we coded a data center cluster with green investment in the upper quartile (the biggest scale) of all cases as “fully in” the set of high green investment. As shown in Table 5, we use those three thresholds to calibrate each condition and outcome.

Step 2: Analysis of necessary conditions

In fsQCA, an outcome does not exist without a necessary condition, and the value of the necessary condition must be higher than 0.9 (Rihoux and Ragin 2008). Our results indicated that there was no necessary condition—that is, none of the five conditions could fully produce a non-green outcome (i.e., Carbon Emissions Intensity) or an expected green outcome (i.e., ~Carbon Emissions Intensity). Table 6 shows the results of the necessary condition analyses.

Step 3: Configuration analysis

The results of this step are provided in Table 7. Either the present outcome (i.e., Carbon Emissions Intensity) or absent outcome (i.e., ~Carbon Emissions Intensity) has three different configurations; hence, we attained Eqs. (3) and (4):

where * represents Boolean AND (i.e., intersection), and~represents Boolean NOT (i.e., non-membership). The conditions of each configuration are as follows: Configuration 1 (GI*R*~GA*~DL), Configuration 2 (~GI*~R*GE*GA), and Configuration 3 (GI*~R*GE*~GA*DL) are the low-carbon configurations; and Configuration 4 (~GI*~R*~GE*GA), Configuration 5 (~GI*R*~GE*~GA*DL), and Configuration 6 (GI*R*~GE*GA*~DL) are the high-carbon configurations.

Although this study was focused on how to reduce the carbon emissions intensity of the digital economy in China, we also analyzed the conditions for the unexpected outcome of high-carbon emissions intensity: Configuration 4 indicates that the government’s green attention should be present, while public green investment, the number of racks in data centers, and regional green energy should be absent in order to contribute to the high-carbon emissions intensity (~GI*~R*~GE*GA). This configuration is exemplified by the Helingel Cluster and Zhongwei Cluster, both of which are located in China’s western region. Configuration 5 suggests that the racks and the digital activity at the city level should be present, while green investment, green energy, and the government’s green attention should be absent in order to increase the carbon emissions intensity (~GI*R*~GE*~GA*DL). The example case for Configuration 5 is the Guian Cluster, located in China’s western region. Configuration 6 shows that green investment, the number of racks, and the government’s green attention should be present, while the regional green energy and the city’s digital level should be absent (GI*R*~GE*GA*~DL). The example case for Configuration 6 is the Zhangjiakou Cluster, located in China’s eastern region.