Construction Planning of China’s Computing Power Center from the Perspective of Electricity–Computing Synergy

Abstract

Against the backdrop of the energy crisis and a rapidly advancing digital economy, electricity–computing synergy has become a strategic pathway to resolve energy constraints in computing power center and overcome renewable energy consumption challenges. This study breaks through the existing single-factor fragmented analysis method and systematically constructs a vertically progressive and horizontally coupled electricity–computing synergy planning model to deconstruct the core elements of computing power center construction and reconstruct the path of electricity–computing value co-creation. It proposes a multi-objective site selection decision-making method for computing power center based on linear weighting and the principle of reusability and universality, effectively avoiding the problem of overall system efficiency loss caused by single-objective optimization. Based on the empirical results from data from 31 provinces in China, this study classifies the endowments for computing power center construction, conducts targeted analyses of each province’s situation, and finds three major contradictions facing China’s computing power center: a spatial mismatch between green energy resources and service demand, a dynamic imbalance between electricity price advantages and comprehensive costs, and structural contradictions between talent reserves and sustainable development. Finally, a multi-dimensional integrated strategy was systematically constructed, encompassing demand-driven initiatives, electricity price adjustments, talent innovation, natural cold-source activation, and network upgrades, to provide guidance and policy toolkits for the government planning of computing power infrastructure development.

1. Introduction

In recent years, China’s computing power demand has sustained rapid growth, driven by both an active data factor market (the national data production volume reached ZB 32.85 in 2023, representing a year-on-year increase of 22.44%, while total data storage amounted to ZB 1.73) and robust expansion in the cloud computing sector (the public cloud market recorded a compound annual growth rate [CAGR] of 44.6% from 2021 to 2023, with the private cloud segment growing at a CAGR of 23.7%). Propelled by escalating computing needs, computing infrastructure has undergone rapid expansion: by the end of 2023, operational data centers in China totaled 8.1 million standard racks, marking a 24.2% year-on-year increase, and the aggregate computing power scale exceeded 230 EFLOPS (FP32), ranking second globally. Consequently, this growth has precipitated severe energy supply constraints, with total electricity consumption by computing centers surging to approximately 150 billion kWh in 2023—a 15% increase year-on-year. In the game between expanding computing power demand and energy constraints, there is an urgent need to break the deadlock through electricity–computing synergy mechanisms.

As dual-core infrastructure elements driving economic development, the symbiotic relationship between electricity and computing power has become a critical proposition in reconstructing the infrastructure system. In the context of the simultaneous development of energy reform and digital economy, countries around the world have accelerated the deep integration of power and computing power. In China, for example, in the “Implementation Opinions on Deeply Promoting the ‘East Data West Computing’ Project to Accelerate the Construction of a Nationally Integrated Computing Power Network”, it is explicitly put forward that, by the end of 2025, the initial establishment of a bidirectional power and computing synergistic mechanism and the proportion of green energy in the newly built computing power center of national hub nodes will reach more than 80%. The computing power center is a facility dedicated to providing computing resources such as high-performance computing, data processing, and analysis, focusing on the provision of powerful computing power and large-scale data-processing capabilities, in the broad sense that it contains a data center, a smart computing power center, and a super-computing power center. However, practical implementation faces challenges such as insufficient green energy utilization and inefficient regional coordination, reflecting deep-seated contradictions: structural imbalances between the exponential growth of computing power demand and rigid energy supply constraints, role mismatches between passive consumers and active regulators under the underutilized flexible-adjustment capabilities of computing systems, and value transmission blockages due to the ineffective activation of bidirectional data empowerment.

Existing research mainly focuses on single factors, such as electricity costs, when discussing development strategies for computing power centers regarding electricity–computing synergy, ignoring the fact that the construction of computing power centers is a multidimensional interactive system. This separates the coupling relationship between factors such as user demand and service quality, greatly reducing the research’s actual application value. Therefore, it is urgent to construct a comprehensive and dynamic adjustment-mechanism-based electricity–computing synergy planning model that covers multiple levels and perspectives, including energy supply, service quality, cost control, and talent supply, in order to promote the optimal allocation of electricity and computing resources and provide a solution for constructing an electricity–computing synergy network that combines theoretical depth and practical feasibility. Therefore, from a comprehensive perspective and dynamic adjustment mechanism, it is urgent to construct a multi-level and comprehensive electricity–computing synergy planning model covering energy supply, service quality cost control, talent supply, and other factors to promote the optimal allocation of electricity and computing resources and provide a solution for constructing an electricity–computing synergy network that combines theoretical depth and practical feasibility.

2. Literature Review

Throughout the world, related research and studies on electricity–computing synergy have primarily focused on three key areas: the competitiveness of computing power centers, the relationship between electricity costs and computing costs, and the collaborative planning of electricity–computing systems.

2.1. Computing Power Center Competitiveness

The identification of competitiveness enhancement paths helps us to target the construction of the computing power center so as to significantly enhance its sustainable development capabilities, which is a key step in the planning of electricity–computing synergy. At the technical level, Li et al. pointed out that technical capabilities, such as hardware-basic technology, software ecosystem, and information security, determine the speed of computing power development, and high investment in computing power technology research and development and infrastructure construction must be strengthened to promote the high-quality development of China’s computing power [1]. At the cost level, Ke et al. pointed out that energy costs account for the highest proportion of operating costs, so the level of green energy saving and intelligent management capabilities are key elements of the computing power center to maintain a competitive advantage, of which 40% of the energy is used for cooling, so the use of air-side economizers in cold climates can greatly reduce energy consumption and operating costs [2,3,4]. At the macro layout level, Christensen et al. pointed out that the competitiveness of a computing power center is shaped by a variety of factors such as political stability, national supportive policies, the level of industrial clusters, and the spatial layout of circles [5,6]. Existing studies have analyzed paths to enhance the competitiveness of computing power centers, focusing on factors like energy costs, technical capabilities, and supportive policies. However, discussions on the synergy among these factors are relatively limited.

2.2. Relationship Between Electricity Tariffs and Costs

The computing power industry is a high-energy-carrying industry, and the cost of electricity accounts for about 60% of the operating cost, and low electricity costs have become one of the core driving forces for the development of the computing power industry. In terms of power price optimization, Rao et al. point out that, in the context of soaring power consumption in computing power centers, in addition to reducing the power consumption of the equipment, power management that minimizes the total power cost is the core of the development of computing power, and they propose a strategy to optimize the power cost by combining the diversity of electricity prices in terms of time and location [7]. Zhang et al. point out that electricity prices are closely related to time, and that computing power centers should reduce their processing speed in the high-price period and take full advantage of the low-priced period to reduce the cost, and propose a strategy to optimize the cost according to the price of electricity, and they proposed an optimization algorithm to dynamically adjust the processing speed of tasks according to the time of high and low electricity prices [8]. In terms of the utilization of energy storage devices, Guo et al. point out that cloud service providers can adopt a strategy of charging at low tariffs and discharging at high tariffs to fully utilize energy storage devices, thus significantly reducing the actual tariffs and achieving the optimization of the total operating costs of the computing power center [9,10]. In terms of green energy introduction and the context of the energy crisis, Lei et al. point out that the introduction of green energy into computing power centers has become a trend, and that an intelligent, green, energy-saving scheduling strategy can effectively optimize the operating costs of computing power centers [11]; in addition, cooling system electricity is an important source of energy consumption in computing power centers, and Chen et al. point out that the cost structure of computing power centers can be effectively optimized through direct liquid-cooling technology, waste-heat recovery technology, and dynamic temperature regulation capability, for example, using its waste heat for tobacco-drying can increase energy reuse efficiency by 64.92%, and using its waste heat for eco-farms can save 676 MWh of electricity annually [12,13,14,15]. Existing studies have initially explored the relationship between power cost and computing power cost from the perspectives of cost structure, energy structure, etc. However, there is a lack of in-depth research on the impact of factors such as power structure, energy efficiency, and regional differences on the operation of computing power centers, as well as the reverse optimization of the computing power side to the power side. There is also a lack of research on the closed-loop value flow between power and computing power.

2.3. Electricity–Computing Synergy Construction

The rapid growth in energy use in the computing power industry and the need for the transformation of new power systems make electricity–computing synergy a key development trend. At the level of internal optimization of computing power, Ding et al. point out that energy efficiency is the main goal of computational collaboration, based on the integrated energy supply system within the computing power park and the optimization planning of power supply structure, to expand the overall benefits of computing power centers and power grids simultaneously [16,17]. At the level of computing resource scheduling, Ding et al. point out that making full use of the operational flexibility of a computing power center is the basis of computing convergence, based on the matching and real-time adjustability of computing power resources; constructing a bidirectional collaborative scheduling architecture of adjustable resources for computing nodes can realize the efficient use of computing resources [18,19]. At the level of the overall layout of the country, Erdem et al. point out that the construction of electricity–computing synergy needs to take into account multiple factors such as energy distribution, regional synergy, population and industrial layout, political risk, legal risk, etc., so as to build a synergistic framework such as the computing power channel [20,21,22,23]. Existing research focuses on key nodes of electricity–computing synergy, but lacks analysis from the perspective of the electricity–computing value co-creation on how computing power centers can achieve breakthroughs from single points of economic value to social value, as well as the deep integration of computing power infrastructure with urban and national development.

Based on existing research, although the interactive relationship between electricity and computing power has been preliminarily revealed from multiple dimensions, analysis of the two-way coupling relationship between electricity and computing power is insufficient and lacks systematicity and comprehensiveness. Previous research separates the coupling between factors such as energy, services, and costs, and ignores the value conversion path from energy to data. The fragmented research paradigm can easily lead to a decision-making dilemma where “energy priority leads to service degradation, cost optimization leads to talent loss.” The construction of computing power centers is a multidimensional interactive system that depends not only on internal computing power factors, but also on external market factors. Internal and external factors need to be comprehensively considered to construct an electricity–computing synergy planning layout.

Therefore, this study starts from a comprehensive perspective and dynamic adjustment mechanism to construct a multi-level comprehensive electricity–computing synergy planning model covering energy supply, service quality, cost control, and talent supply. It proposes a multi-objective decision-making method based on linear weighting and the method of least squares, with the aim of providing a data-driven decision-making tool for selecting a computing power center. Using the empirical results of the computing power construction endowment and the suitability of provinces, it analyzes the bottlenecks currently encountered in the development of the computing power industry and further formulates targeted strategies for the development of computing power centers. This paper aims to provide theoretical insights and policy guidance for promoting the sustainable development of computing power centers and the construction of electricity–computing synergy. It hopes to help computing power centers and governments take more effective measures to achieve sustainable development while responding to the global challenges of the energy crisis.

Compared to previous studies, the main contributions of this study are as follows:

(1) In terms of theoretical contributions, it breaks through the existing single-factor fragmented analysis method and systematically constructs a vertically progressive and horizontally coupled electricity–computing synergy planning model, deconstructing the core elements of computing power center construction and reconstructing the path of electricity–computing value co-creation. It helps to clarify the spiral upward synergistic development mechanism of “energy value foundation—data value appreciation—ecological value feedback”, providing theoretical support for the symbiotic evolution of electricity–computing systems.

(2) In terms of methodological contributions, single-cost optimization is prone to selecting cost-depressed areas and facing talent shortages. This study establishes a multi-objective decision-making method based on linear weighting and the law, balancing multi-dimensional objectives such as energy supply, service quality, cost control, and talent factors, avoiding the problem of overall system efficiency loss caused by single-objective optimization and providing data-driven, reusable, and universal decision-making tools for the selection of computing power centers. Based on empirical results from data from 31 provinces in China, this study categorizes the endowments for computing power center construction for the first time, provides a targeted analysis of the situation in each province, and proposes three major bottlenecks that urgently need to be addressed in the current development of the computing power industry: green energy–service mismatch, electricity price–cost imbalance, and talent–development contradictions.

(3) Practical contributions: This study proposes targeted planning strategies based on development bottlenecks, providing the national computing power construction initiative with directly implementable technical solutions and policy toolkits.

3. Theoretical Model

3.1. Construction of Electricity–Computing Synergy Planning Theoretical Model

As a dual-core driving element of digital economic development, electricity–computing synergy planning is essentially a value co-creation system construction based on complex systems theory. The two have resource symbiosis, spatial and temporal complementarity, and value co-creation, and the deep coupling of electric power supply capacity and computing power demand characteristics helps to realize the three-dimensional balance of resource efficiency, economic value, and social benefits. Through the closed-loop system of “input-conversion-output-re-input”, electricity and computing power presents a two-way value flow cycle. On the electric power side, electric power provides an energy base and task guidance for computing power. Electricity price signals guide computing tasks to shift in the time dimension and migrate in the space dimension so as to form the adaptive adjustment ability of computing load to the fluctuation of renewable energy. On the computing power side, computing power creates data value to feed electric power. The computing power center promotes green energy consumption and grid stability through scale procurement while transforming computing power into digital economic productivity, forming a chain of leaps from watts to value. Electricity–computing synergy planning realizes the paradigm upgrade from physical siting to value co-creation through the systematic matching of electricity resource availability, computing power service effectiveness, and operation cost control, as shown in Figure 1.

3.2. Operational Mechanisms of Electricity–Computing Synergy Planning Theoretical Model

The Green Energy Access Circle constitutes the core layer of energy control, the High-Quality Service Circle constitutes the layer of value creation and empowerment, and the Cost Optimization Circle constitutes the layer of synergy for efficiency improvement. The three circles are vertically progressive and horizontally coupled. Vertically, the Green Energy Access Circle provides the energy base for computing power centers, the High-Quality Service Circle realizes the value transformation, and the Cost Optimization Circle feeds the sustainable operation of the first two circles through performance improvements. Horizontally, the electric power procurement strategy affects the service pricing ability, and the network service quality determines the cost recovery cycle, forming a dynamic equilibrium through a triangular model of cost–benefit-risk. And with the Green Energy Circle as the dominant one in the early stages of construction, it becomes driven by the service circle in the operation period, and finally realizes the system maturity leap through the cost circle to form an orderly spatiotemporal evolution.

Under the synergistic promotion of the three circles, the computing power center’s value will break through from the economic value to the social value system reconstruction, and realize the deep integration of infrastructure and urban development through the synergistic ecology of multi-subject participation. This is embodied as a triple value leap. First, the energy value closed loop: The Green Energy Access Circle, through large-scale new energy consumption capacity and grid frequency modulation, service a two-way empowerment of computing power load-flexible adjustment capacity into power frequency modulation auxiliary service revenue. Second, the data value spillover: The High-Quality Service Circle, through latency-sensitive business agglomeration and digital innovation ecological cultivation, catalyzes the regional digital industry snowball effect. Third, ecological value circulation: The Cost Optimization Circle drives multi-directional interaction involving regional industries and livelihood services, such as waste-heat recovery, to create ecological practices that promote the transformation of computing power centers from energy-consumption entities to urban carbon-neutral nodes.

China’s “East Data West Computing” represents the spatial expression of the electricity–computing synergy planning model, where a mapping relationship exists between the model and the eight hub nodes of this initiative. This relationship manifests through differentiated functional emphases of the three circles across hubs. Specifically, Inner Mongolia, Gansu, and Ningxia operate as Green Energy Access Circle-dominant hubs, underpinning the “West Computing” cost advantage. Relying on the abundant wind and solar resources in Western China, these hubs reduce electricity prices through a direct green power supply and give priority to non-real-time computing power tasks with a latency tolerance of more than one hour. These hubs attract computing power demand based on their low energy costs and convert renewable resources into stable energy flows, transforming large amounts of wasted wind and solar power into increased revenue from ancillary power services.

Meanwhile, Beijing-Tianjin-Hebei, Yangtze River Delta, and Guangdong-Hong Kong-Macao function as High-Quality Service Circle-dominant hubs. With network latencies ≤ 5 ms and coverage of 60% of national computing demand, these hubs specialize in latency-sensitive services like financial transactions and real-time rendering. Through continuous technological advancement, these hubs will become the driving force behind the transformation of GDP growth per unit of computing power into a leap in regional digital industry penetration, and will promote the transition from computing power efficiency to economic value-added growth. Finally, Guizhou and Chengdu-Chongqing serve as Cost Optimization Circle-synergistic hubs, balancing energy and locational advantages to accommodate transitional East–West demand and bridge value chains. Through measures such as the regional reuse of waste heat, these hubs will help transform computing power centers from energy consumers into urban energy nodes, promoting a shift from carbon emission reduction to lower costs for public services, ultimately achieving a socialized cycle of ecological value.

Electricity–computing synergy exhibits a spiral upward pathway of “energy value foundation—data value appreciation—ecological value feedback.” Through multi-stakeholder collaboration among governments, power grids, enterprises, and the public, it achieves a Pareto improvement in power resources, computing capabilities, and social welfare.

3.3. Influencing Factors of Computing Power Center Planning from the Perspective of Electricity–Computing Synergy

Computing power demand is the primary element. The amount of demand that a computing power center is able to capture is the basic variable for resource allocation, and its scale and structure directly affect the scale, functional design, capital expenditure, and operational efficiency of the computing power center, which need to be fully assessed and quantified first at the planning stage. Computing demand also determines power demand: on the one hand, the growth in demand requires the deployment of more hardware devices, which will generate more electric power consumption; and on the other hand, the increase in the number and density of devices will generate more cooling demand. With the chain reaction of computing power demand, the computing power center needs to manage power resources based on demand forecasting to ensure efficient, stable, and sustainable operation.

Electric power cost is a basic element. As an energy-intensive industry, computing power centers have electric power costs accounting for more than 50% of their operating costs, so reducing power costs and energy conservation is an important direction for computing power infrastructure development. On the one hand, through the proximity to a renewable energy power generation area, the direct supply of green energy can be utilized to significantly reduce the cost of electric power procurement and reduce the loss of electric power transmission and efficiency loss, so as to form a long-term cost advantage. On the other hand, natural cold-source advantages can be utilized to create structural cost advantages by reducing cooling costs and depreciation costs through reduced cooling equipment usage and extended server life.

Computing power performance is the core element. When a computing power center provide services to consumers, performance metrics, such as mathematical processing power, response speed, and resource utilization efficiency, are decisive factors for service quality and user satisfaction, which depend on key factors such as CPU and other hardware performance, technical architecture, and the number of network hops. In particular, the number of network hops has an exponential amplification effect on end-to-end latency. Taking computing power quality of service as the focus and addressing developmental constraints by moving closer to the user side, optimizing network topology and deploying edge nodes are the core of realizing improved quality of service and supporting the high value-added needs of the digital economy.

Labor cost is an important element. Computing power centers are high-tech-level field projects with a high degree of dependence on core technical personnel. From the perspective of labor supply and demand, the short supply of highly educated talent in remote areas will not only lead to a rise in salary levels and an increase in manpower costs, but may also trigger competition for talent, exacerbate employee mobility, and increase recruitment and training costs. From the perspective of the industrial chain, the quality and stability of core technical talents directly affects production capacity and service quality. Therefore, improvements in the talent cultivation, introduction, and retention mechanism will have an important impact on sustainable development.

As the core link of electricity–computing synergy planning, the decision-making process of computing power center construction site selection needs to weigh energy supply, service quality, cost control, and talent elements, as well as other multi-dimensional objectives, to achieve an optimal end result, which essentially belongs to the multi-objective optimization problem. In the electricity–computing synergy system, regarding the asymmetric coupling relationship between the objectives, energy-advantageous areas are prone to service performance degradation, cost-depressing areas are prone to face a lack of supply of talent, and single-objective optimization is very easy to trigger the overall effectiveness of the system compromised. Multi-dimensional and multi-scale interaction determines that computing power center siting must adopt a multi-objective decision-making framework and construct a siting decision-making system with a dynamic adaptive capacity through reasonable weight configuration between objectives.

4. Methodology

4.1. Multi-Objective Decision System for Siting the Computing Power Center

The determination of decision-making objectives followed the principle of quantifiable in elements. First, the network latency minimization target, based on the response speed, is the core element of computing power center in the construction process used to determine the effectiveness of the computing power service using the user cluster-weighted node number modeling, directly mapping the computing power center site on the latency-sensitive business support capacity. Second, the electricity price minimization target reflects the integrated grid power purchase price and wind power supply price under the difference between the cost of electric power, again being the wage level. The minimization target reflects the demand for cost control of human resources, and the premium coefficient is measured through the proportion of the highly educated population to transform the supply and demand of talents into quantifiable indicators. Finally, the minimization target of average annual air temperature is used to reflect the regulation of operating costs when using natural cold sources. Through the four key decision-making dimensions that independently characterize the electricity–computing synergy system and the decision-making objectives that form a complex feedback network with the value transmission chain of energy–service–cost–talent, a multi-objective decision-making system for the computing power center location was constructed under the perspective of electricity–computing synergy in order to ensure a systematic and comprehensive location selection, as shown in Figure 2.

Objective function I: Minimum network latency. The network latency of the computing power service mainly depends on the number of weighted network nodes to be crossed by the computing power center siting area from each user cluster. And, considering that the Internet industry occupies 50% of the downstream industrial demand of the computing power center, and it is the industry with the largest computing power demand, the relative weight of each user cluster is thus expressed by the normalized value of the business revenue of the Internet and related services industry.

$$f_1=min\sum _{k=1}^K{n}_{node}(x,k)·w_k=min\sum _{k=1}^K{n_{node}(x,k)·I}_k/\sum _{j=1}^K{I}_j$$

(1)

where $w_k$ is the relative weight of the user cluster, $I_k$ is the business revenue of the user cluster from Internet and related services, and $n_{node}(x,k)$ is the number of nodes required to span from the siting region x to the user cluster k.

Objective function II: Minimum use of electricity price. The price of electricity used by the computing power center depends on the price of electricity sold from the regional grid and the price of green energy, which is negatively correlated with the regional green energy generation capacity under the scale effect.

$$f_2=min\sum _{t=1}^T\left[{\lambda }_{gri}{·P}_{gri}\right(x)+{(1-{\lambda }_{gri})·P}_{green}(x\left)\right]$$

(2)

where ${\lambda }_{gri}$ is the grid dependency, $P_{gri}$ is the grid purchase price, and $P_{green}$ is the direct wind power supply price.

Objective function III: Minimum wage level. High-quality technical talents are the core competitiveness of the computing power center, so regions with a low percentage of highly educated population need to pay a certain percentage of wage premium to attract talents and reduce staff turnover.

$$f_3=min\left(1+\mu \right)·C\left(x\right),\mu =\left\{\begin{array}{c}\begin{array}{c}20\%, M_{hig}\left(x\right)/M_{pro}\left(x\right)<0.4\\ 10\%, 0.4\le M_{hig}\left(x\right)/M_{pro}\left(x\right)<0.8\end{array}\\ 0, 0.8\le M_{hig}\left(x\right)/M_{pro}\left(x\right) \end{array}\right.$$

(3)

where $\mu$ is the wage premium, $M_{hig}$ is the number of graduates from general higher education institutions, $M_{peo}$ is the number of resident population, and $C$ is the average annual labor cost of the information industry.

Objective function IV: Minimum annual mean temperature, calculated from mean monthly temperature data.

$$f_4=min\sum _{i=1}^{12}{F}_i\left(x\right)/12$$

(4)

where $F_i$ is the average temperature of the site area in the i-th month.

4.2. Multi-Objective Decision Solving for Siting the Computing Power Center

This study employs the linear weighting method to achieve manifested weight and deliver rankable results. The weights of indicators were calculated using the CRITIC method, which objectively assigns weights based on the comparative intensity of indicator data and inter-indicator conflict. A sensitivity analysis was conducted to test the stability of the final siting decisions under different weight combinations (perturbed within ±10% of baseline weights) via a Spearman correlation analysis, thereby verifying the robustness of the solutions derived from the linear weighting method and ensuring that decision outcomes remain stable despite minor weight fluctuations. The specific computational steps are as follows:

(1) Dimensionless processing was performed using the min–max normalization method to eliminate dimensional influences on evaluation results, as shown in Equation (5).

$${\mathrm{x}}_{\mathrm{j}\mathrm{i}}^{\mathrm{’}}={\displaystyle \frac{x_{ij}-\mathrm{m}\mathrm{i}\mathrm{n}\{x_{1,}{x}_2,\dots ,x_n\}}{\mathrm{m}\mathrm{a}\mathrm{x}\{x_{1,},\dots ,x_n\}-\mathrm{m}\mathrm{i}\mathrm{n}\left\{x_{1,\dots ,}{x}_n\right\}}}$$

(5)

(2) Indicator variability is represented by the standard deviation $S_j$. A larger standard deviation indicates greater dispersion in the indicator’s values, reflecting richer information and stronger discriminatory power for evaluation. Indicator conflict $R_j$ is measured by the inter-indicator correlation coefficient $r_{ij}$. A higher correlation coefficient implies lower conflict between indicators, signifying greater redundancy in their evaluation content, thereby justifying reduced weight allocation for such indicators. The specific calculations are detailed in Equation (6).

$$S_j=\sqrt{{\displaystyle \frac{{\sum }_{i=1}^n{(x_{ij}-{\overline{x}}_j)}^2}{n-1}}} , R_j={\sum }_{i=1}^p(1-r_{ij})$$

(6)

(3) Indicator weighting $w_j$ was computed according to Equation (7), where a larger indicator information quantity $C_j$ signifies a more significant role of the j-th indicator in the system, warranting a higher weight.

$${w_j={\displaystyle \frac{C_j}{\sum _{i=1}^p{C}_j}} , C}_j=S_j\times R_j$$

(7)

(4) Based on the linear weighting method, the multi-objective function was transformed into a single-objective function using Equation (8) to obtain the decision value $Z\left(X\right)$.

$$Z\left(X\right)=W^T·F={\sum }_{i=1}^4{w}_i·f_i\left(x\right)$$

(8)

(5) Disturb the weight $w_j$ within ±10% of the benchmark value to generate a new weight combination and calculate the disturbance value ${Z\left(X\right)}^{\prime }$ and the difference $d_i$ between the sample ranking under the benchmark weight and the disturbed weight. Then calculate the Spearman rank correlation coefficient $\rho$ according to Equation (9) to test the stability of the ranking.

$$\rho =1-{\displaystyle \frac{6\sum d_i^2}{n(n^2-1)}}$$

(9)

5. Empirical Analysis

5.1. Data Sources and Organization

Relevant data for China’s 31 provinces come from the National Energy Administration, the National Bureau of Statistics, China’s Ministry of Industry and Information Technology, and the China Academy of Information and Communications Technology and its computing power center official website, cloud computing platform, and other websites, and the number of cross-domain nodes is based on China Unicom’s CHINA169 backbone network as a reference, and the proportion of green energy used is based on the “In-depth Implementation of the ‘East Data West Computing’ Project to Accelerate the Construction of the National Integrated Computing Power Network Implementation Opinions”. The ratio of green energy usage is set ${\lambda }_{gri}$ to 0.2 based on the “Implementation Opinions on Deepening the Implementation of the ‘East Data West Computing’ Project and Accelerating the Construction of a Nationally Integrated Computing Power Network”.

Based on the two principles of actual grounded operation and computing power scale over 40PFLOPS, 35 computing power center in China were selected to obtain relevant service and operation data, which were crawled from the technical specification pages, service documents, or news announcements of the official websites of the computing power center and cloud computing platforms using Python 3.9 crawler technology, and were formed into an empirical dataset after data cleansing and structured processing.

5.2. Empirical Results of Siting the Computing Power Center

First, based on STATA MP 18 software, four indicators of 35 computing power centers in China in 2023—namely, network latency, electricity prices, wage levels, and average annual temperature—were normalized, as shown in

Table A1. Standardized data of 35 computing power centers in China.

Objective Function	I	II	III	IV
Beijing Digital Economy Computer Power Center	0.5776	0.7184	0.9513	0.4749
Chongqing Artificial Intelligence Innovation Center	0.0412	0.7916	0.2999	0.7542
Chongqing Jiayun Intelligent Computer Power Center	0.7588	0.7916	0.2999	0.7542
China Mobile Intelligent Computing Power Center (Qingdao)	0.2024	0.3878	0.1618	0.4469
China-ASEAN Artificial Intelligence Computing Power Center	0.1494	0.8039	0.1133	0.9497
China Telecom Beijing-Tianjin-Hebei Intelligent Computing Power Center	0.0635	0.7623	0.2847	0.5028
Changsha Artificial Intelligence Innovation Center	0.96	0.783	0.155	0.7263
Changsha 5A Intelligent Computing Power Center	0.1929	0.783	0.155	0.7263
Wuhan Artificial Intelligence Computing Power Center	0.6541	0.8286	0.2032	0.6983
Tianjin Artificial Intelligence Computing Power Center	0.9718	0.7623	0.2847	0.5028
Tencent Intelligent Industry Yangtze River Delta Smart Computing Power Center	0.5412	0.7342	0.1183	0.6425
Tencent Yangtze River Delta Artificial Intelligence Computing Power Center	0.4553	1	1	0.6704
Shenzhen Artificial Intelligence Convergence and Empowerment Center	0.0188	0	0.5663	1
Shanghai Artificial Intelligence Public Service Computing Platform	0.4753	1	1	0.6704
Shangtang Technology Artificial Intelligence Computing Power Center	0.4612	1	1	0.6704
Shanxi Advanced Computing Power Center	0.7094	0.4453	0.0468	0.3073
Nanjing Kunpeng-Shengteng Artificial Intelligence Computing Power Center	0.0588	0.6923	0.4193	0.6425
Baidu Intelligent Cloud Kunlun Core Yancheng Smart Computing Power Center	0.3635	0.6923	0.4193	0.5587
Qingdao Artificial Intelligence Computing Power Center	0.7871	0.3878	0.1618	0.4469
Ningbo Artificial Intelligence Super Computing Power Center	0.9871	0.8162	0.816	0.6704
Nanjing Intelligent Computing Power Center	0.8941	0.6923	0.4193	0.6425
Jinan Artificial Intelligence Computing Power Center	0.1659	0.3878	0.1618	0.5307
Hengqin Artificial Intelligence Computing Power Center	0	0.9691	0.5663	0.9777
Hefei Artificial Intelligence Computing Power Center	0.0118	0.7342	0.1183	0.6425
Hangzhou Artificial Intelligence Computing Power Center	0.6859	0.8162	0.816	0.6983
Harbin Advanced Computing Power Center	0.9812	0.6674	0.0293	0
Guangzhou AI Convergence and Empowerment Center	0.3859	0.9691	0.5663	0.9777
Gansu Qingyang Smart Computing Power Center	1	0.3205	0	0.2514
Fujian Artificial Intelligence Computing Power Center	0.6835	0.7082	0.247	0.838
Dalian Artificial Intelligence Computing Power Center	0.5047	0.5913	0.1562	0.3631
Chengdu Artificial Intelligence Computing Power Center	0.3859	0.5325	0.2815	0.6425
Beijing Rise AI Computing Power Center	0.5741	0.7184	0.9513	0.4749
Aliyun Ulanqab AI Computing Power Center	0.5024	0.487	0.11	0.0838
Taihu Quantum Computing Power Center	0.1106	0.6923	0.4193	0.6704
Shenyang Artificial Intelligence Computing Power Center	0.0235	0.5913	0.1562	0.2235

(Appendix A). According to Equations (5) and (6), the indicator variability, indicator conflicts, and indicator weighting were obtained to ensure that the weighting could truly reflect the importance of each target in the overall decision-making process. The decision objective weights results are shown in

Table 1. Calculation results of decision objective weights.

Decision Objective	Indicator Variability	Indicator Conflict	Indicator Information Quantity	Indicator Weighting (%)
Network Latency	0.331	3.285	1.088	37.905
Electricity Price	0.217	2.147	0.465	16.205
Wage Levels	0.316	2.2	0.695	24.205
Average Annual Temperature	0.236	2.639	0.622	21.685

.

Then, based on the indicator weighting, the multi-objective decision function for selecting the location of the computing power center can be converted into a single-objective function, as shown in Equation (10).

$$Z\left(X\right)=\mathrm{m}\mathrm{i}\mathrm{n}(0.379051f_1{+0.16205f}_2+0.24205f_3+0.21685f_4)$$

(10)

Consequently, the corresponding target values were calculated based on relevant data from China’s 31 provinces, and the provinces were ranked in ascending order according to their target values, as shown in Figure 3.

Finally, we completed a sensitivity analysis. By successively varying the indicator weighting of the four objectives by ±10% from the baseline values, the decision values and rankings under perturbation were obtained. According to Equation (8), the Spearman rank correlation coefficients between the baseline ranking ${\mathrm{D}}_0$ (results shown in

Table 2. Calculation results of Spearman rank correlation coefficient.

	Dw1+	Dw1−	Dw2+	Dw2−	Dw3+	Dw3−	Dw4+	Dw4−
D0	0.988 (0.000 ***)	0.978 (0.000 ***)	0.997 (0.000 ***)	0.999 (0.000 ***)	0.993 (0.000 ***)	0.998 (0.000 ***)	0.994 (0.000 ***)	0.995 (0.000 ***)

*** p < 0.01.

) and all perturbed rankings are ≥0.978 (p < 0.01), while pairwise correlations among all ranking sets exceed 0.946 (p < 0.01). This verifies that the results generated using the linear weighting method are insensitive to weight perturbations and exhibit stability.

5.3. Analysis of Computing Power Center Siting Results

5.3.1. Spatial Differentiation Characteristics Analysis

Based on the K-Means method, the target values for the location of computing power center were clustered, dividing China’s 31 provinces into five tiers and drawing a geographical distribution map of the tiers to achieve spatial visualization, as shown in Figure 4. On the gradient leap path, the geographic distribution of the echelons presents a continuous spectrum of the Western green-energy depression-central transition zone–Eastern service highland, and the difference between the echelons also confirms the necessity of a synergistic layout between Western bases and Eastern edges.

The first tier comprises two Central–Western provinces, Shanxi and Inner Mongolia, which leverage energy and temperature advantages to form a cost depression. With average wind and solar power generation (105.7 billion kWh) significantly exceeding the national mean (35.581 billion kWh), this tier achieves green energy procurement prices 43.52% and 56.85% lower than the national average, confirming the foundational role of the Green Energy Access Circle. Simultaneously, an average annual temperature below 9 °C (national mean is approximately 15 °C) enables computing power centers to substantially reduce cooling costs via natural cooling modes, optimizing the cost structure. However, the tier’s mean weighted node count (2.61) exceeds the eastern provincial average (2.03), reflecting a deeper network topology layer. Demand-side optimization is thus required to further transform energy advantages into service value.

The second tier is mainly categorized into Western energy-driven and Central and Eastern efficiency-driven models. As depicted in Figure 5, the labor cost (mean 0.08) in this tier is markedly lower than the national average (mean 0.23). Western energy-oriented provinces (e.g., Qinghai, Gansu) couple wind and solar resources with cold climate advantages to achieve cost control, while Central and Eastern efficiency-driven provinces (e.g., Shandong, Jiangxi) have leveraged their low-latency advantages to achieve service quality that is more than 60% higher than the national average. Among them, Shandong has a unique geographic location and industrial chain synergy composite value. On the geographic level, with the Beijing-Tianjin-Hebei and Yangtze River Delta convergence of regional advantages, not only can they undertake the spillover of computing power demand in the East, but also, through the province’s green energy depression, reduce the cost of energy consumption and the formation of spatial synergies of demand and energy; and on the industry chain level, with the help of the province’s Wave and other server manufacturing giants and through the localization of the supply chain, they can reduce the cost of equipment purchasing, operation, and maintenance, the formation of computing power, and the manufacturing of the advantages of vertical integration. However, compared to the first echelon of provinces with a natural cold-source disadvantage—needing to rely on liquid cooling and other artificial refrigeration technology, weakening the potential for cost optimization—computing power in Shandong Province is mainly used in the industrial manufacturing sector, the penetration rate of medical care, culture and tourism, governmental affairs, and other scenarios is low, and there is a structural imbalance of “heavy training, light application”, which fails to fully realize the value of computing power in multiple industries.

The third tier is mainly divided into Central and Eastern efficiency-driven and Northeastern natural-cold-source-driven types. Efficiency-driven provinces (e.g., Tianjin, Henan) still have high-quality service capability, but the cost is about 32% higher than that of the efficiency-driven provinces in the second echelon, which significantly reduces the excellence of site selection. Northeastern regions (e.g., Jilin, Liaoning) demonstrate pronounced advantages in natural cold-source utilization, with an average annual temperature of 7 °C (54.51% lower than the national mean), enabling a >30% reduction in cooling energy consumption.

The fourth and fifth tiers are mainly high-volume high-cost types whose Internet business revenue accounts for more than 50% of the national total. However, cost pressures, such as electric power and manpower, constitute a service premium trap, especially in the three Eastern provinces within the fifth echelon, all of which have the worst numerical terms in the cost target, putting the computing power center in the provinces in a cost dilemma.

5.3.2. Core Contradictions Analysis

Electricity–computing synergy planning needs to break through the logic of linear trade-offs between energy, service, and cost and turn to the construction of a nonlinear synergy network with the three contradictions at its core.

First, the spatial mismatch between green energy resources and service demand is contradictory. With abundant wind and solar energy resources (wind power generation in the West is 53.8% higher than that in the East), the Western region has lower electric power procurement costs (the East–West electricity price imbalance coefficient is 1.72:1). However, the West is far from the core area of computing power demand, which is mainly in the East (Western demand is 92.7% lower than Eastern demand), and these services suffer from excessive latency and inefficient data transmission (the East–West latency imbalance coefficient is 0.64:1), which leads to difficulties in guaranteeing the quality of real-time computing power services, i.e., overly focusing on energy advantages and neglecting service effectiveness may lead to a decline in user experience. And West–East transmission relies on long-distance ultra-high-voltage transmission, which undermines the cost advantage of green energy at high loss rates. In the case of double-value dissipation, a two-way flow of computing power to the West and data to the East is needed to reconfigure the resource-matching relationship.

Secondly, the dynamic imbalance between electricity price advantages and comprehensive costs is contradictory. Regions that have developed a significant electricity price advantage by virtue of their green-energy resource endowment have seen higher unit computing power costs than expected in actual operations due to escalating maintenance costs for hardware equipment repair, cooling system maintenance, regular inspections of the power supply system, software system optimization, and other maintenance costs, which have substantially undermined the benefits of electricity cost savings. In terms of labor costs, for example, the level of wage premium in the electricity price advantage zone is 30% higher than the benchmark level in other regions. Essentially, it is difficult to hedge the marginal cost increment of other elements by the single-dimension optimization of electric power costs, and there is an urgent need to balance the marginal cost increment effect of each segment through system cost optimization to achieve long-term economic benefits.

Third is the structural contradiction between the talent reserve and sustainable development. Take the Northeast region as an example. Its cold climate conditions can achieve natural cold-sources to reduce refrigeration energy consumption by 30%, which is a good cost optimization value, but the talent reserve in the region is seriously insufficient: the graduate talent turnover rate, according to the total number of graduates and Northeastern-origin caliber statistics, reached 63.46% and 26.45%, respectively. The scarcity of high-quality core technicians not only affects the daily operation of the computing power center, but also impedes technological upgrades, innovation, and service quality improvements, and cannot support the long-term stable development of the computing power industry. This contradiction stems from the disconnection between the talent supply system and the demand of the computing power industry, and there is an urgent need to build a flexible talent supply mechanism of localized training and cross-regional sharing.

6. Countermeasures

Electricity–computing synergy planning is a strategic approach to addressing energy constraints and value-transformation issues in computing power center. Aiming to optimize resource allocation efficiency and promote innovative development across the computing power value chain, this study proposes five recommendations based on the current state of the industry and three core contradictions. These recommendations highlight actionable pathways; however, their effectiveness needs to be verified through more in-depth and detailed analyses in the future in order to refine the solutions and assess their feasibility.

(1) Cultivate a computing power application ecosystem with demand-driven orientation: In view of the problems of the single-computing-power demand scenario and insufficient industrial penetration, we should take demand as the basis to build a trinity-computing-power application ecosystem of “scenario expansion-ecological synergy-resource scheduling”. First, implement the “computing power +” program to strengthen the penetration of vertical industries through the promotion of AI quality inspection, digital twins, medical imaging cloud platforms, and virtual reality fusion interaction scenes to focus on breakthroughs in the three major vertical fields of industry—medical care, culture and tourism, and the deployment of the “1 + N” computing power application matrix—to form multi-industry computing power. Second, build an industrial synergy platform for computing power suppliers, platform operators, and industry users, and promote the cross-domain circulation of computing power resources through standard interfaces and computing power coupons to form a demand-driven benign ecological cycle. Third, set up a cross-regional trading market for computing power resources, and realize nationwide computing power scheduling based on blockchain technology to decouple the computing power tasks by layers and establish a Western training market to meet the requirements of time latency. Establish a division of labor model of training in the West and reasoning in the East to achieve a dynamic match between real-time business-oriented model in the East and non-real-time business-oriented model in the West.

(2) Take the tariff signal as a lever to build a dynamic cost-adjustment mechanism: In response to the problem of the electricity price advantage being eroded by operation and maintenance costs, the whole life cycle economy of computing power center should be started, and the price signal theory should be used to guide the optimal allocation of factors. First, implement wind computing three-dimensional dynamic pricing during the peak hours of wind power generation to reduce the price of green energy and provide computing power migration incentives, such as real-time power over 80% of the rated capacity of the East back to the 1 PB hard-drive to give 0.8 CNY/kWh tariff subsidies, and the implementation of an initiative to reduce the non-emergency computing power load incentives in the summer peak hours of electricity, such as per MW cuts to obtain compensation of CNY 800. Second, develop load-flexible regulation products by training computing power cluster intelligent regulation algorithms to achieve 15 min level electric power demand-side response commands in order to participate in the grid frequency modulation auxiliary service market to obtain compensation revenue. Third, construct modular small- and medium-sized computing power centers in green-energy-rich areas through the accurate matching of the scale of computing power with the supply of green energy, expanding the flexibility in the expansion of the capacity, and reducing the cost of construction and the duration of construction, as well as other strategies to enhance the ability to regulate dynamic costs.

(3) Take the talent ecology as the core and innovate the flexible supply system: In order to break the contradiction between cost advantage and talent shortage, we should take the innovation of talent supply mode as a strategic breakthrough and build a three-dimensional driving mechanism of “local cultivation + flexible sharing + policy to retain talents”. First, deepen the industry–education fusion cultivation system, the directional cultivation of high-level technicians in the computing power industry, set up additional specialties—such as smart computing operation and maintenance, green energy management, and other specialties—in colleges and universities, and solve the problem of the structural mismatch of local talents through “order-type cultivation”. Second, implement the flexible sharing mechanism of talents and synchronize the implementation of the computing power engineers’ working mechanisms on the cloud and short-term assignment mechanism so as to realize the cross-regional equipment touring mechanism. Third, optimize a special policy of stabilizing employment and living in a secure environment for special talent, and implement a long-term strategy of retaining talent by means of equity incentives for the core talent who are capable of key technological research, and share the revenue from the conversion of achievements, as well as provide differentiated benefits, such as full subsidies for the purchase of house deeds and taxes, to keep the rate of talent turnover at 15%.

(4) Reshaping the competitiveness of Northeast computing power by taking the value of natural cold-sources as a fulcrum: In order to activate the latecomer advantages of regions such as Northeast China, the in-depth development of the value of natural cold-sources should be taken as the starting point, and the closed-loop energy strategy of utilizing natural cold-sources and tapping the value of waste heat should be implemented. First, stimulate the value of natural cold-sources, optimize the design of refrigeration architecture, realize the stacked use of natural cold-sources and mechanical refrigeration in the transition season to extend the use of natural cold-sources throughout the year, and take advantage of the annual average of 8 months of cold climate in the Northeast region, building an ultra-low-Power Usage Effectiveness (PUE) data center in Harbin, Changchun, and other cities. Second, promote the computing power center’s waste-heat reuse in multiple cross-industry applications, such as district heating, agricultural greenhouses, and industrial drying. This will focus on breaking through the coupling technology of waste-heat recovery and the regional heating system, realizing the waste heat to cover the heating demand of the surrounding 500,000 square meters of buildings, and feeding back the cost of computing power with the heat price gain to build a closed loop of energy of cold-source–computing power–thermal energy. Thirdly, create a natural cold-source economic ecosystem, joining hands with manufacturers and scientific researchers to form a cold region computing power industry alliance, and output targeted solutions for high-latitude countries through the output of products, such as cold-source technology patent packages and cold operation and maintenance services, so as to realize the transformation of natural advantages into technological advantages.

(5) Build a foundation of network infrastructure to promote ubiquitous access to computing power resources: In order to crack the latency predicament of computing power services, we should start from the network capacity, which is the prerequisite for releasing the potential of computing power services, and improve the basic support system of “backbone capacity expansion + edge deployment + protocol innovation”. First, the backbone cable expansion project should be implemented to accelerate the deployment of a 400 G/800 G ultra-high-speed transmission network, relying on China’s advantages in optical communication technology, to build an all-optical switching non-blocking intelligent computing power center and a highly reliable three-dimensional computing backbone network, especially to optimize the all-optical interconnection architecture between computing power hubs, under the strategy of “East Data West Computing”. Second, promote the ubiquitous deployment of edge computing power, deploy a miniature modular data center within a radius of 30 km in demand-intensive areas, such as the Yangtze River Delta and the Pearl River Delta, promote the sinking of edge computing nodes to industrial parks and urban communities, and form a three-tier computing power service system of core, edge, and terminal. Third, create an intelligent network protocol system, perform research and development on new types of network protocols to support the resolution of computing power identifiers and the awareness of resources, and realize the linkage between task demands and resources via computing power routing optimization technology. This will involve innovation toward an intelligent network protocol system, developing new network protocols that support computing power identification resolution and resource sensing, and achieving millisecond matching between task demand and resource supply through computing power optimization technology.