Cross-Institution Reweighting of National Green Data Center Indicators: An AHP-Based Multi-Criteria Decision Analysis with Consensus–Divergence Diagnostics

Abstract

Evaluating green data centers is a multi-attribute decision problem. To enhance the rigor and precision of green data center assessment, this study verifies the weighting of the national green data center evaluation index system using the Analytic Hierarchy Process (AHP) with the participation of 19 domain experts from various data center sectors. The aim is to gain an in-depth understanding of the perspectives and priorities of different types of institutions regarding evaluation indicators and to investigate the underlying reasons for these perspectives and priorities. Through an analysis of expert sample distribution, this paper reveals the preferences of financial, internet, research, and design, as well as technical consulting service institutions, regarding indicators such as energy-efficient utilization, computational resource utilization, green low-carbon development, scientific layout, and intensive construction. Specifically, financial institutions tend to place a relatively lower emphasis on energy efficiency due to their focus on transaction speed and security. In contrast, internet companies prioritize efficient utilization of computational resources. Research and design institutions consider scientific layout and intensive construction more crucial, while technical consulting service institutions emphasize green and low-carbon development. Meanwhile, we identified substantial discrepancies among experts in determining the weights of specific indicators, suggesting a lack of consensus within the industry about the correlation between these indicators and green data centers. To propel the sustainable development of green data centers, future assessments should refine evaluation dimensions, consider disparities such as data center types and embrace regional differences, actively adopt novel technologies and innovative practices, and establish mechanisms for long-term monitoring and evaluation.

1. Introduction

Given the growing global momentum for carbon reduction, data centers, as an energy-intensive industry, are being compelled to pursue low-carbon and clean development, making green transformation an inevitable path forward [1]. Since 2021, data centers have been officially classified as the ninth category of “high energy consumption and high emissions” industries, alongside the traditional eight sectors [2]. The continually increasing electricity consumption of data centers has set new records, placing considerable pressure on national efforts to control energy consumption and reduce carbon emissions. The United States, the European Union, and other regions began developing green data center evaluation frameworks as early as 1998, though the focus of these frameworks varies. Some emphasize green building and energy efficiency, while others concentrate on design, operations, and management practices. In 2021, Chinese authorities officially released a national evaluation index system for green data centers and launched the corresponding recommendation program. This study aims to apply the Analytic Hierarchy Process (AHP) to determine the weightings of the evaluation indicators and to conduct a comparative analysis with the existing national evaluation weights. The objective is to gain deeper insights into the perspectives and priorities of different types of institutions regarding the evaluation criteria and to propose potential improvements to the current green data center evaluation system.

In the current body of research on green data center evaluation, the literature can broadly be categorized into two domains: investigations into the developmental pathways for green data centers and methodological studies on their evaluation frameworks. In the research domain concerning the green development pathways of data centers, numerous scholars have proposed specific improvement strategies for the design, construction, and operation of green data centers. For example, Xu et al. [3] optimized the development directions of green data center design and construction with the goal of achieving high reliability and cost-effectiveness, and what they identified as four major trends in data center development are reliability, efficiency, greenness, and intelligence. Zhu et al. [4] analyzed what are expected to be the five potential opportunities that will influence the future development of data centers, namely the digital revolution, market diversification, methods of integrating renewable energy into data center construction and post-construction operation and management, advanced cooling technologies, and the heterogeneity of the public in the information age. Shao et al. [5] analyzed the influencing factors of data center energy consumption, and what they proposed is that appropriate indicators should be selected for optimization according to the functional characteristics of data centers, based on the classification of evaluation indicators by priority. Hu et al. [6] emphasized that developing green and energy-efficient modular data centers, which represent a new type of facility, has significant advantages when compared with traditional construction methods In the research domain of green data center evaluation frameworks, Li et al. [7] emphasized that the consideration of operational reliability, resource conservation, energy efficiency improvement, environmental friendliness, and human well-being in green data centers is what will significantly contribute to advancing both urban digitalization and green development. Zhou et al. [8] proposed what they regarded as a comprehensive carbon reduction pathway, which includes green planning, green design, green construction, green operation and maintenance, and green energy utilization. Hu et al. [9] conducted a detailed analysis of the critical factors for constructing green and highly available data centers in the financial sector, and what they constructively suggested is that financial institutions should develop multilayer data centers that are efficient, secure, green, energy-saving, resiliently scalable, intelligent, and user-friendly. Uddin et al. [10] developed a sustainability evaluation system for data centers, which established an indicator framework from six perspectives: site location and outdoor environment, energy use, water use, material resource use, indoor environmental quality, and improvement and innovation. Wang et al. [11], drawing on the evaluation system of green hospital buildings, reviewed the existing grading standards and technical guidelines for green data centers in China, and what they proposed is that specific scoring should be conducted in five dimensions: energy efficiency, energy-saving technologies, green management, innovative exploration, and green building performance. Li et al. [2] compared the Guidelines for the Evaluation of Green Data Centers with the Evaluation Index System, noting that the former stresses energy-saving technologies while the latter emphasizes life-cycle progress. Both, however, neglect human development factors, indicating the need for refinement and unification. Cai et al. [12], examined domestic and international green building evaluation systems and proposed a framework tailored to data centers, consisting of nine overall and seven energy indicators. Their findings suggest that the Chinese Assessment Standard for Green Building emphasizes indoor environmental quality, management, and passive design, whereas international systems focus more on energy performance and active design [13]. However, although these contributions are valuable, what the existing studies on development pathways often lack is comparative analysis across different institutional contexts, and what they have failed to achieve is the integration of these pathways into a comprehensive evaluation framework, which is exactly the gap this study aims to address.

In evaluating data centers and sustainable infrastructure, commonly used multi-criteria decision-making (MCDM) methods include the analytic hierarchy process (AHP), best-worst method (BWM), TOPSIS/VIKOR, PROMETHEE/ELECTRE, and analytic network process/decision-making Trial and evaluation laboratory (ANP/DEMATEL). In addition, hybrid schemes that combine subjective and objective weighting, such as integrating entropy-based weights, have been developed [14]. The different methods emphasize different aspects, as illustrated by AHP, which, as a technique that derives weights through hierarchical modeling and pairwise comparisons, incorporates a consistency ratio (CR) check and is particularly suitable when samples are small and interpretability is required. Moreover, it has been applied to examine what factors organizations consider when selecting cloud data-center solutions [15]. That BWM compares all criteria to the identified best and worst ones means that the number of pairwise judgments is substantially reduced, which typically yields higher consistency and stability [16]. TOPSIS and VIKOR rely on an alternative-by-criterion performance matrix, which makes them effective for ranking options and identifying compromise solutions. They are suitable when data are ample and the comparators are well defined, for example, for cross-site evaluations of candidate data-center locations [17]. PROMETHEE and ELECTRE are based on outranking relations and preference functions; they can handle non-compensatory and threshold preferences, but have been used relatively little in comprehensive evaluations of data centers [18]. ANP and DEMATEL, which model interdependencies and feedback among indicators and identify causal pathways, require more demanding data and modeling capacity [19]. Entropy weighting and hybrid weighting, which extract objective weights from data dispersion and combine them with AHP or BWM, can enhance robustness [20]. With respect to methodological fit, when the objective is to obtain baseline weights for a multilevel indicator system and expert judgment is the primary evidence, AHP or BWM is preferable; when cross-center ranking of alternatives is required, TOPSIS, VIKOR, PROMETHEE, or ELECTRE is more suitable; and when the coupling among energy use, water use, and reliability needs to be represented explicitly, ANP or DEMATEL is appropriate. Given the context of this study—namely a hierarchical indicator system, a small cross-industry expert panel, and the need for interpretable and communicable weights—AHP is selected as the core method.

The existing national evaluation framework has revealed several limitations in practice. On the one hand, its fixed weight assignments may be biased, placing excessive emphasis on energy-efficiency indicators (e.g., PUE) while underestimating criteria such as scientific layout and efficient utilization of computing resources. On the other hand, the diverse operational objectives and environmental responsibilities across sectors (finance, internet, research, consulting, etc.) mean that a uniform weight system may not adequately reflect actual sector-specific priorities. Thus, the central research question of this study is as follows: To what extent do the nationally prescribed indicator weights reflect sectoral needs, and should the framework be rebalanced and optimized according to industry characteristics?

To address this question, the study pursues three objectives:

(1)

Validation of rationality: to test the applicability of national indicator weights across different industry perspectives through expert assessment;

(2)

Identification of divergence points: to apply a dispersion-based diagnostic method to capture consensus and divergence, highlighting key indicators such as “cabinet resource utilization” where disagreement is most significant;

(3)

Development of improvement measures: to propose sector-specific and operable design recommendations for refining the green data center evaluation framework on the basis of the national scheme.

Methodologically, this study applies the Analytic Hierarchy Process (AHP) to construct a hierarchical indicator model and elicits pairwise comparisons from experts in finance, internet, research, and consulting sectors to derive weights. To go beyond the traditional AHP, which outputs only averaged weights, a dispersion and coefficient of variation (CV) diagnostic mechanism is introduced to quantify consensus and divergence across industries. By benchmarking the results against the national baseline weights, the analysis not only identifies alignment and misalignment between national assignments and sectoral preferences but also provides an empirical foundation for reweighting and framework optimization.

The main contributions of this paper are as follows:

(1)

Cross-institutional revalidation: using AHP, the study re-estimates national indicator weights and finds that PUE is overweighted, while “scientific layout and intensive construction” and “efficient utilization of computing resources” are underweighted;

(2)

Sectoral preference profiling: mapping indicator weight distributions across different institution types and management levels, and proposing scenario-specific weighting recommendations;

(3)

Consensus diagnostics: identifying indicators with the greatest divergence—particularly “cabinet resource utilization” (dispersion = 0.20)—which require clarification and long-term monitoring.

The structure of this paper is as follows: Section 2 introduces the national evaluation framework for green data centers; Section 3 develops the multi-attribute evaluation model based on AHP with the integration of consensus diagnostics; Section 4 presents empirical results and sector-specific recommendations; and Section 5 concludes with policy implications.

2. Green Data Centers: Concepts and Frameworks

2.1. Fundamental Connotation of Green Data Center

Due to differences in practical conditions and development priorities, the definition and interpretation of “green” vary across countries and regions. In the United States, the emphasis is placed on energy-saving technologies and the adoption of renewable energy. Policies primarily promote the deployment of high-efficiency equipment and technologies, and encourage enterprises to increase the use of renewable energy sources [21,22]. The European Union mandates that data centers comply with the EN 50600 standard, which outlines specific requirements concerning energy efficiency, carbon emission thresholds, and environmental safety [23]. In Singapore, the definition of green data centers extends beyond energy efficiency and the use of renewable energy, placing additional emphasis on the adoption of high-efficiency equipment and management strategies to achieve low carbon emissions and minimal resource consumption [24]. In Japan, the concept of green data centers also encompasses requirements for optimizing both the design and operation of data center facilities [25]. In China, the definition of green data centers typically adopts a life-cycle perspective, emphasizing enhanced energy efficiency and reduced environmental impact while ensuring the reliability and stability of information systems [26]. Notably, different government bodies and organizations present varying emphases. The Ministry of Housing and Urban-Rural Development defines green data centers primarily from the perspective of green building standards. In contrast, organizations such as the China Electronics Society focus more on improving energy efficiency based on the foundational requirements of information system security and operational stability [27,28,29].

It is worth noting that prior comprehensive evaluations of green data centers often adopt an aggregated end-user perspective (general users, enterprises, government) and thus do not fully capture heterogeneity in compute demand across industries. In the context of the rapid increase in the scale and digitization of the industry, if universal indicators and weights continue to be used, it may lead to redundancy and a shortage of computing power in some industries, making it difficult to align the requirements of the construction side accurately. Accordingly, it is necessary to recalibrate evaluation emphases and weighting schemes by industry characteristics. The objective of this study is to utilize the Analytic Hierarchy Process (AHP) to elicit industry-specific preference weights and establish sector-oriented baseline weights for evaluating green data centers, thereby laying the groundwork for a more targeted assessment framework.

2.2. Comparison of Evaluation Frameworks

The evaluation of green data centers originated in the United States. In 1998, the U.S. Green Building Council (USGBC) released the first edition of the Leadership in Energy and Environmental Design (LEED) evaluation guidelines, which have since evolved to the fifth edition [30]. LEED certification takes buildings as the primary object of evaluation and emphasizes environmental protection and sustainable development in both the structure and its surrounding context [31]. Figure 1 presents the comprehensive building evaluation indicators and their weights as published by the USGBC in the LEED guidelines. Within this building-oriented scheme, the weights denote the relative importance of each dimension for assessing a project’s overall environmental and sustainability performance. USGBC establishes these weights based on green-building practice and expert judgment, and applies them to both existing buildings and new construction. In the context of evaluating green data centers, the relevant certifications are LEED BD + C (Building Design and Construction) and LEED O + M (Building Operations and Maintenance). This study uses LEED O + M as the exemplar for discussing the evaluation framework.

The assessment framework encompasses eight major categories of indicators (as illustrated in Figure 1), including but not limited to the following:

(1)

Sustainable Sites: This category focuses on the environmental impact of site selection. It assesses whether the building effectively utilizes existing infrastructure and minimizes disruption to the surrounding ecosystem.

(2)

Water Efficiency: Emphasizes strategies to reduce water consumption in cooling systems and encourages the adoption of water management practices such as rainwater harvesting and reuse.

(3)

Energy and Atmosphere: Concentrates on energy efficiency and carbon emissions. It highlights energy management in data centers, including the efficiency of servers, equipment, and cooling systems.

(4)

Materials and Resources assess the use of renewable or recycled materials and strategies for reducing construction and operational waste within the data center.

(5)

Indoor Environmental Quality: Focuses on maintaining a stable and appropriate indoor environment to ensure the optimal operation and longevity of data center equipment, particularly servers.

(6)

Innovation and Design Process: Recognizes the application of innovative technologies, design strategies, or performance achievements that exceed standard LEED requirements [30].

(7)

Regional Priority: Provides additional credits for projects that address regional environmental priorities, as defined by local climate, geography, and regulatory needs.

(8)

Location and Transportation: Evaluates the building’s location in terms of accessibility to sustainable transportation options and its integration with surrounding communities.

As an internationally recognized certification system, LEED provides a standardized reference framework for the evaluation of green data centers. It facilitates the adoption of consistent standards by data center operators and designers worldwide, thereby promoting the industry’s transition to a greener approach. However, certain limitations remain. Given the unique characteristics of the data center sector, operators often prioritize hardware stability and reliability over energy efficiency. Moreover, the technological advancements and evolving standards in the data center industry tend to progress more rapidly than those in traditional building sectors. Since LEED is fundamentally centered on green building assessment, it still requires adaptations to address the specific needs of data centers, ensuring its effectiveness and applicability in this context.

In addition to LEED, other widely adopted international evaluation systems for green data centers include the BREEAM Data Centers (BREEAM DC) standard developed by the Building Research Establishment (BRE) in the United Kingdom, and the Energy Star Score for Data Centers issued by the U.S. Environmental Protection Agency (EPA) [32]. The BREEAM Data Centers evaluation system, developed by the Building Research Establishment (BRE), is a green building assessment method specifically designed for data centers. It focuses on evaluating the design, construction, and operational phases of data centers. Similarly to LEED, BREEAM DC adopts a building-centered assessment approach; however, key differences exist in the indicator structure. Notably, BREEAM DC incorporates additional evaluation dimensions related to operational management and social impact, placing greater emphasis on resource management, waste treatment, and overall sustainability practices. In contrast, LEED tends to prioritize energy efficiency and reducing the carbon footprint. The BREEAM DC framework covers multiple dimensions, including energy performance, water resource management, material selection, waste management, operations and management, and social impact. The ENERGY STAR for Data Centers program, developed by the U.S. Environmental Protection Agency (EPA), focuses specifically on evaluating the energy efficiency of data centers. It emphasizes the optimization of hardware and equipment performance to reduce overall energy consumption.

The ENERGY STAR for Data Centers certification developed by the U.S. Environmental Protection Agency (EPA) primarily focuses on key performance indicators related to energy efficiency. Among these, Power Usage Effectiveness (PUE) is a core metric. Additional factors commonly considered include

(1)

Data Center Infrastructure Efficiency (DCIE): The reciprocal of PUE (i.e., 1/PUE), DCIE measures the proportion of total energy consumption that is actually used to power IT equipment.

(2)

Energy Use Profile: A detailed analysis and reporting of energy consumption across major components such as cooling systems, lighting, and IT equipment.

(3)

Cooling Efficiency: Assessment of the design and operational efficiency of the cooling infrastructure.

(4)

Server Utilization: Evaluation of how effectively IT hardware is utilized to avoid energy waste.

(5)

Energy Management Strategies: Examination of formal energy management plans, periodic energy monitoring and reporting, and ongoing efforts for improving energy efficiency.

In contrast to the LEED rating system, which provides a broader evaluation framework encompassing various aspects of the built environment—including energy, materials, water resource management, and site sustainability—ENERGY STAR places a stronger emphasis on operational energy efficiency, particularly metrics such as PUE. However, the green data-center evaluation frameworks still exhibit several limitations: LEED (BD + C/O + M) remains building-centric, such that operations- and utilization-oriented indicators—including compute-load utilization, cabinet occupancy, and network resource utilization—are insufficiently specified, and no weighting or threshold design is differentiated by industry; BREEAM, although it strengthens management processes, provides neither clearly quantified indicators and monitoring mechanisms for utilization and O&M performance nor sector-specific parameterization guidance; ENERGY STAR for data centers, insofar as it is tightly focused on operational energy efficiency, has relatively narrow coverage, pays limited attention to water use, siting/resilience, and utilization dimensions, and does not offer a multi-indicator weighting framework that reflects sectoral context.

China’s green data center evaluation system has undergone a gradual evolution from nonexistence to initial development, and subsequently toward refinement. In its early stages, the evaluation of green data centers in China drew upon international frameworks and, similarly, focused primarily on buildings as the object of assessment. The release of the Technical Guidelines for the Evaluation of Green Data Center Buildings in December 2015 marked the introduction of China’s first formal evaluation system specifically for green data centers. This framework categorizes evaluation indicators into seven dimensions: land conservation and outdoor environment, energy efficiency and energy utilization, water conservation and water resource utilization, material conservation and material resource utilization, indoor environmental quality, construction management, and operation management. Additionally, a bonus-point mechanism was established to incentivize improvements and innovations in data center construction. Similarly to the LEED rating system, this framework primarily emphasized the physical attributes of the building itself, rather than assessing the internal infrastructure of the data center.

These evaluations provide a basis for reconstructing China’s comprehensive evaluation system for green data centers. At the level of indicator weighting, the system should be rebalanced by moderating the exclusive reliance on PUE, increasing the weights for scientific layout and intensive construction, and for the efficient utilization of computing resources, and closing gaps in operations-and-maintenance indicators. At the methodological level, adopting a coupled framework that links AHP-derived weights with consensus–divergence diagnostics can distinguish indicators suitable for common guidance from those requiring sector-specific calibration, thereby improving the interpretability and generalizability of the evaluation process. The procedure is also portable: by incorporating contextual variables such as grid carbon intensity, climate conditions, water stress, and workload profiles, weights can be localized, thereby strengthening cross-country comparability.

Subsequently, various professional associations and research institutions in China introduced their own standards for evaluating green data centers. Notable examples include the Evaluation Guidelines for Green Data Centers issued by the China Institute of Electronics, the Evaluation Standards for Green Data Centers published by the Architectural Society of China, and the Evaluation Indicators and Methodology for Green Data Centers released by the China Institute of Communications [13]. These evaluation frameworks were published during the 12th to 14th Five-Year Plans, aligning with the evolving national policies and energy-saving requirements of each stage. The specific evaluation indicators are summarized in

Table 1. The current index design for evaluating green data centers in China.

Standard Name	Technical Guidelines for the Evaluation of Green Data Center Buildings	Evaluation Guidelines for Green Data Centers	Evaluation Standards for Green Data Centers	Evaluation Specification for Green Data Centers
Indicators	Land Conservation and Environment	Power Usage Effectiveness	Planning and Design	Power Consumption Load Utilization Rate
	Energy Efficiency and Energy Utilization	Energy-saving Measures	Power Supply System	Renewable Energy Utilization
	Water Conservation and Utilization	Energy Management System	Cooling System	Water Resource Utilization
	Material Conservation and Utilization	Water Resource Utilization	Information Systems	Green Diagnosis Procurement
	Indoor and Outdoor Environmental	Resource Recycling	Innovation	Application of Technologies
	Construction Management	Hazardous Substance Control	Operation and Maintenance Management	Quality Management
	Operation Management	Renewable Energy Utilization		Business requirements
	Improvement and Innovation	Infrastructure Security		Green Product Management

.

3. Evaluation Models and Methods for Data Centers

3.1. Green Data Center Evaluation Model

Currently, the National Green Data Center Evaluation Index System is one of the most widely used frameworks for assessing green data centers in China. This evaluation system was officially released as part of the Notice on Organizing the 2021 National Green Data Center Recommendation Work, jointly issued by the General Offices of the Ministry of Industry and Information Technology, the National Development and Reform Commission, the Ministry of Commerce, the National Government Offices Administration, the China Banking and Insurance Regulatory Commission, and the National Energy Administration. The framework consists of four primary indicators—efficient utilization of energy and resources, green and low-carbon development, scientific planning and intensive construction, and efficient utilization of computing resources—as well as 15 secondary indicators. Detailed indicator descriptions and corresponding weights are shown in

Table 2. National green data center evaluation index system and explanations of indicators.

Indicator		Definition of Indicator	Weight Score
A1	Power usage effectiveness	Ratio of total annual electricity consumption to that of IT equipment	40
	Renewable energy utilization level	Proportion of renewable electricity used annually	10
	Reduction in comprehensive energy	Annual decrease in energy consumption per unit of information flow	2
	Intelligent energy management level	Use of smart systems and policies for energy monitoring and control	7
	Waste heat and cold recovery	Use of systems for waste heat/cold recovery and reuse	4
A2	Water resource utilization level	Ratio of annual water consumption to IT equipment electricity use	5
	Green procurement level	Implementation of a system prioritizing eco-friendly product purchases.	5
	Green operation and maintenance	Regular maintenance and proper waste management based on regulations	5
	Green retrofit improvement	Implementation of energy-saving or green upgrades within the past	3
	Green public service participation	Engagement in promoting green or participating in standard development	3
A3	Scientific layout level	In key regions, serving public or backend computing needs.	3
	Intensive construction level	Consolidated or reused site with natural or regulatory advantages.	3
A4	Cabinet resource utilization level	Share of cabinets with effective space and equipment use.	4
	Computing load utilization level	Ratio of average annual power consumption per cabinet to the average rated power.	3
	Network resource utilization level	Ratio of average annual network traffic to total installed network bandwidth.	3

. The first-level dimensions are A1 efficient energy utilization, A2 green and low-carbon development, A3 scientific layout and intensive construction, and A4 efficient utilization of computing resources.

Based on the above indicator system, judgment matrices can be constructed to provide initial data for multi-attribute evaluation. According to

Table 3. Judgment matrix for primary indicators of green data center.

A	A1	A2	A3	A4
A1	1	4/3	4	2
A2	3/4	1	9/8	8/9
A3	1/4	8/9	1	3/5
A4	1/2	9/8	5/3	1

, which outlines the green data center evaluation indicator system, a total of five judgment matrices are developed. Judgment Matrix 1 represents the pairwise comparisons among the first four-level indicators: A1, A2, A3, and A4, as shown in Table 3. Subsequently, judgment matrices are constructed for the second-level indicators under each primary category: a11–a15 under A1, a21–a25 under A2, a31–a32 under A3, and a41–a43 under A4. All matrices are formed using the 1–9 scale method for expert scoring and evaluation.

In this study, a total of 19 experts were invited to assign weights to the evaluation indicators. The experts were drawn from leading internet data centers, financial data centers, data center research and design institutions, technical consulting firms, and testing and evaluation agencies, ensuring a high level of representativeness and professional diversity. Among them, five were management-level personnel. Detailed information is provided in

Table 4. Expert composition by institution type.

Number	Type of Institution	Number of Experts	Number of Management-Level Experts
1	Internet data centers	4	1
2	Financial institutions	4	2
3	Research and design institutions	3	1
4	Technical consulting firms	7	1
5	Testing and evaluation agencies	1	0
Total		19	5

.

3.2. Green Data Center Evaluation Method

The Analytic Hierarchy Process (AHP) is a methodological approach that combines both qualitative and quantitative analysis to determine decision weights for complex multi-objective problems. Initially developed by Saaty (1980), AHP integrates subjective expert judgments with objective data to address complex decision-making scenarios [33]. The method requires experts to perform pairwise comparisons among evaluation criteria to assess their relative importance. A consistency check is then conducted to ensure the logical coherence of expert judgments. Once consistency is verified, the corresponding weights of each indicator can be derived. AHP has proven particularly effective in domains where purely quantitative methods are insufficient or impractical. Given the hierarchical indicator structure, the small expert panel, and the need for readily interpretable results, AHP was chosen because it derives weights via pairwise comparisons, enforces consistency through the CR check (CR < 0.10), and yields transparent global weights that can be benchmarked against national or industry frameworks.

The weight vector can be determined through expert consultation, empirical and/or field research, or by adopting existing theoretical frameworks. The steps of the AHP are as follows:

(1)

Establish the hierarchical structure of the AHP model.

The first step is to clearly define the research objective and construct a hierarchical model that captures the influencing factors. For most systems, the hierarchy can be divided into three levels: the overall goal at the top, the first-level criteria (primary indicators) in the middle, and the second-level criteria (secondary indicators) at the bottom. This structure helps organize complex decision-making problems systematically and logically, facilitating pairwise comparisons and weight derivation in subsequent steps.

(2)

Constructing the pairwise comparison judgment matrix.

Based on the established hierarchical structure model, a pairwise comparison is performed among the factors within each level, and corresponding judgment matrices are constructed. The relative importance or preference of each factor is assessed and assigned a quantitative value, typically using the 1–9 scale method. One of the main advantages of the Analytic Hierarchy Process (AHP) is that it does not require experts to assign exact weights to each evaluation factor directly; instead, they are asked to evaluate the relative importance between factors. Assume there are evaluation factors, and the intensity of importance of factor iii relative to factor j is represented by a_ij. Therefore, a pairwise comparison matrix A can be obtained as follows:

$$A=\left[\begin{array}{cccc}{a}_{11}& a_{12}& \dots & a_{1n}\\ a_{21}& a_{22}& \dots & a_{2n}\\ ⋮& ⋮& \dots & ⋮\\ ⋮& ⋮& \dots & ⋮\\ a_{n1}& a_{n2}& \dots & a_{nn}\end{array}\right]$$

(1)

(1)

When it indicates that it is more important than $a_j$;

(2)

$a_{ij}=1$ indicates that $a_i$ and $a_j$ are of equal importance;

(3)

$a_{ij}={\displaystyle \frac{1}{a_{ij}}}$ and $a_{ij}\ne 0$.

(3)

Calculation of indicator weights.

The most commonly used methods are the square root method, summation method, and eigenvector method. This paper primarily adopts the summation method, which involves dividing each element of a column by the sum of that column and then averaging the rows of the resulting matrix. The weight vector can be obtained by normalizing each column vector of the matrix A.

$$w_i={\displaystyle \frac{1}{n}}{\displaystyle \sum _{j=1}^n\frac{a_{ij}}{{\displaystyle {\sum }_{i=1}^n{a}_{ij}}}}(i,j=1,2,\dots ,n)$$

(2)

(4)

Consistency check of the judgment matrix.

To verify the rationality of the weight allocation, a consistency check of the judgment matrix is required. Is the consistency ratio of the judgment matrix, while is the consistency index of the judgment matrix. The formula for calculating $CI$ is provided:

$$CI={\displaystyle \frac{{\lambda }_{\max }-n}{n-1}}$$

(3)

$${\lambda }_{\max }={\displaystyle \sum _{i=1}^n\frac{{\left(A\overrightarrow{w}\right)}_i}{{\left(nw\right)}_i}}$$

(4)

$RI$ is the average random consistency index of the judgment matrix. When it is considered that the judgment matrix has satisfactory consistency and the weight assignments are deemed reasonable, the judgment matrix is considered satisfactory. Otherwise, the judgment matrix needs to be adjusted and rechecked. After passing the consistency check, the comprehensive weight of the indicator can be obtained. If the data fails the consistency check, it is necessary to examine for potential logical issues, and the judgment matrix should be re-entered for further analysis. The formula for calculating $CR$ is as follows:

$$CR={\displaystyle \frac{CI}{RI}}$$

(5)

If the weights have been calculated and the judgment matrix satisfies the consistency check, a conclusion can be drawn, and further analysis can proceed.

In this study, the Analytic Hierarchy Process (AHP) is chosen as the primary evaluation method for its suitability with small expert samples and strong interpretability. AHP provides a structured framework to decompose complex problems into a hierarchy of criteria, enabling experts to make pairwise comparisons with consistency checks. This choice aligns with our goal of verifying and rebalancing national weight allocations for green data center criteria, where expert knowledge is essential but data are limited. Compared to data-driven or “black-box” techniques (e.g., machine learning), AHP ensures transparency and an explicit rationale in weight derivation, facilitating communication and validation of results. We recognize that alternative multi-criteria evaluation methods exist, such as the entropy weight method, fuzzy comprehensive evaluation, and the Best–Worst Method (BWM), each with distinct merits. The entropy weight method (EWM) objectively derives weights from data by calculating the entropy of each indicator’s distribution, which works well when abundant quantitative data are available. However, EWM alone lacks incorporation of expert insight and may not be reliable with a small sample of qualitative judgments. Fuzzy evaluation methods introduce fuzzy set theory to capture uncertainty in expert assessments, often complementing AHP; yet a purely fuzzy approach can reduce the interpretability of how specific expert inputs determine final weights. BWM is a newer technique requiring fewer comparisons by focusing on the best and worst criteria. It improves elicitation efficiency but still relies on expert preferences and produces weights comparable to AHP; indeed, studies have noted that AHP offers a more context-specific prioritization, whereas BWM relies more on abstract mathematical optimization. Considering our context and the need for clarity, AHP’s well-established procedure and consistency verification made it the most appropriate choice. Importantly, this study’s contribution lies not in proposing a new algorithm, but in augmenting the traditional AHP with a “consensus–divergence diagnostic mechanism” to enhance discovery. After deriving criteria weights via AHP, we introduced a diagnostic step to examine the level of expert consensus on each criterion by analyzing the dispersion of their judgments. We quantify dispersion using the coefficient of variation (CV), defined as the standard deviation of experts’ scores for a criterion divided by their mean. A higher CV indicates greater relative variability in expert opinions (i.e., lower consensus), whereas a lower CV signifies that experts’ opinions are more aligned (higher consensus). This integrated AHP-plus-diagnostic approach yielded insights beyond the baseline AHP results. Notably, the “cabinet resource utilization” indicator showed the highest dispersion, with a substantially larger CV than other criteria. In practice, this means experts disagreed most on the importance of cabinet resource utilization in evaluating green data center performance. Recognizing this divergence is valuable, as it flags that criterion as contentious or inconsistently understood, suggesting a need for further examination or consensus-building. By highlighting such disparities, our approach directly addresses the reviewer’s concern about methodological originality: we retained AHP for its suitability and interpretability, while bolstering it with a dispersion-based consensus analysis to uncover and discuss where expert opinions diverge.

4. Results and Discussions

4.1. Comparison of Evaluation Results

Based on the scoring of indicator weights and the judgment matrix constructed by 19 experts in this study (as described in the model of Section 3.1), the eigenvector, eigenvalue, and indicator weights can be calculated. Expert background information is presented in Appendix A 

Table A1. Expert panel profile.

Number	Organization	Sector	Number	Organization	Sector
1	Shanghai Research Institute of Building Sciences	Research & design institute	11	Jiangtian Data	Internet data center
2	Huawei	Manufacturing	12	Zhongke Xianluo	Technical consulting services
3	National Institute of Metrology, China	Research & design institute	13	Travelsky Borun	Technical consulting services
4	Zhongke Xianluo	Technical consulting services	14	CAS Saineng	Technical consulting services
5	Bank of Ningbo	Finance	15	Wilke Laboratory	Testing & evaluation institution
6	Qinhuai Data (Chindata Group)	Technical consulting services	16	JD.com	Internet
7	National Institute of Metrology, China	Research & design institute	17	Travelsky Borun	Technical consulting services
8	China International Capital Corporation (CICC)	Finance	18	Tencent	Internet
9	Desay Yingchuang	Technical consulting services	19	Bank of Beijing	Finance
10	Shanghai Pudong Development Bank	Finance

. The eigenvectors, weight values ($W$), maximum eigenvalue ($\lambda$)), and values for the four primary indicators—energy utilization, green low-carbon development, scientific layout and intensive construction, and efficient utilization of computing resources—are detailed in

Table 5. AHP hierarchical analysis results of primary indicators.

Item	Eigenvector	Weight Value	Maximum Eigenvalue	CI	CV
Energy-efficient utilization	1.666	41.64%	4.085	0.028	0.362
Green low-carbon development	0.872	21.81%			0.510
Scientific layout and intensive construction	0.566	14.16%			0.538
Efficient utilization of computing resources	0.895	22.39%			0.785

RI = 0.890, CR = 0.032; since CR < 0.1, the consistency check is passed.

. The weight calculation results for the secondary indicators are presented in

Table 6. AHP analysis results of weights for secondary indicators and determination of their dispersion.

Secondary Indicator	Weight Result	Dispersion	CV	Secondary Indicator	Weight Result	Dispersion	CV
Energy efficiency utilization—a11	17.04%	0.10	0.324	Green transformation improvement—a24	4.73%	0.07	0.439
Renewable energy utilization level—a12	7.57%	0.07	0.543	Green public service level—a25	1.75%	0.03	0.569
Comprehensive energy consumption reduction per unit information flow—a13	6.06%	0.08	0.787	Scientific layout level—a31	8.80%	0.16	0.332
Intelligent management and control of energy utilization—a14	6.26%	0.06	0.546	Intensive construction level—a32	5.87%	0.16	0.497
Utilization level of waste heat and cooling—a15	4.81%	0.05	0.620	Cabinet resource utilization level—a41	7.64%	0.20	0.667
Water resource utilization level—a21	5.81%	0.12	0.502	Computing power load utilization level—a42	8.18%	0.11	0.376
Green procurement level—a22	3.68%	0.08	0.632	Network resource utilization level—43	6.59%	0.16	0.652
Green operation and maintenance level—a23	5.21%	0.08	0.405

. All results have passed the consistency check. The results obtained from the calculations are as follows:

This study utilized the coefficient of variation (CV = standard deviation/mean) to assess expert consensus. It linked these diagnostics to AHP global weights, thereby separating system-wide priorities from items requiring sector-specific treatment (Table 5 and Table 6). At the first level, dispersion was highest for A4 (0.786), high for A3/A2 (0.539/0.510), and moderate for A1 (0.363). At the second level, a13 (0.787), a41 (0.667), and a43 (0.652) showed the greatest dispersion, whereas a11 (0.324), a31 (0.333), and a42 (0.376) served as consensus anchors. Divergence chiefly reflected cross-sector business/SLA differences, non-aligned measurement definitions, and regional feasibility gaps. Accordingly, indicators with CV ≥ 0.60 should be re-specified by sector and piloted; those with 0.50–0.60 should enter annual review; and those with ≤0.35 can form unified baselines—while standardizing key definitions, incorporating contextual variables (e.g., grid carbon intensity, climate/hydrology, heat-recovery potential), and adopting a 12–18-month (baseline-monitor-re-weight) cadence with CR < 0.10 and CV thresholds as recalibration triggers. Indicators a13, a41, a43, a22, and a15 show significantly high dispersion (all p < 0.05), whereas a11 and a31 serve as consensus anchors. At the dimension level, A4 displays significantly higher dispersion than A1 (z = 2.04, p = 0.041).

4.2. Discrepancy Analysis Between Evaluation Results and the National Standard

Compared to the national green data center evaluation index weights, the AHP-calculated weight results show significant differences in three dimensions: energy utilization efficiency (PUE), scientific layout and intensive construction, and efficient utilization of computing resources (as shown in Figure 2 and Figure 3). In the current national green data center evaluation index, the weight of PUE is significantly higher than the result calculated using the AHP method. However, for the indicators of scientific layout and intensive construction, and efficient utilization of computing resources, the national green data center evaluation index weights are noticeably lower. Additionally, the survey results indicate that different types of organizations and personnel prioritize different aspects in evaluating a green data center.

Based on the data analysis of the above results, the shortcomings of the current evaluation framework can be summarized as follows:

(1)

Financial institution data centers place relatively low emphasis on energy efficiency:

From the perspective of institutional classification, personnel from financial institutions assign relatively lower importance scores to energy efficiency compared with other types of institutions, while assigning higher importance to the efficient utilization of computing resources and green low-carbon development. Within the secondary indicators, they place greater emphasis on computing load utilization and green operation and maintenance levels. This is primarily because, for financial institutions, transaction security and transaction speed have always been the top priorities. Therefore, the importance of uninterrupted service and high-speed data processing is placed above that of energy efficiency.

(2)

Internet institution personnel assign relatively higher importance to efficient utilization of computing resources:

The survey results indicate that personnel from internet institutions consider computing load utilization and the efficient utilization of computing resources to be relatively important, while rating the importance of scientific layout lower. This difference aligns with the nature of internet companies’ business operations, as they rely more heavily on computing resources for various online services.

(3)

Research institutions place greater importance on scientific layout and intensive construction compared to other institutions:

For research and design institutions, the importance of green low-carbon development is highly recognized, especially among management-level personnel. Staff in research institutions tend to adopt a top-down approach to design and planning, optimizing resource allocation and integrating cutting-edge technologies from a strategic perspective. Consequently, they are more inclined to support and guide data centers in adopting scientific layout and green transformation measures to maximize resource utilization efficiency.

(4)

Technical consulting service institutions prioritize green low-carbon development and cabinet resource utilization:

Technical consulting service institutions assign higher weight values to green low-carbon development indicators. In terms of efficient utilization of computing resources, they also attach relatively high importance to cabinet resource utilization levels. As consulting firms often provide advisory services on green sustainability for the design, construction, and operation of data centers, they are more attentive to aspects related to green low-carbon development.

(5)

Management-level personnel place greater emphasis on the green low-carbon development dimension:

From the perspective of personnel types, management-level personnel attach greater importance to green low-carbon development. Among the secondary indicators, they place a relatively high emphasis on green operation and maintenance levels, as well as green transformation. Within the efficient utilization of computing resources dimension, they pay more attention to computing load utilization levels; in the scientific layout and intensive construction dimension, they emphasize scientific layout levels. It is also evident that management-level personnel tend to promote green transformation of data centers through a top-down approach.

(6)

Significant differences among experts in assessing the importance of cabinet resource utilization:

According to the data dispersion (the degree of variation in weight values assigned by experts), cabinet resource utilization shows the most significant divergence in importance assessment. This indicates that within the industry, there is a lack of consensus among experts regarding the correlation between this factor and the green development of data centers.

4.3. Design Recommendations for a Green Data Center Evaluation Framework

Based on the above analysis, different types of institutions show distinct preferences regarding the relative importance of evaluation indicators. To enhance the specificity and operability of the framework, sector-differentiated recommendations are proposed on top of the unified national scheme:

(1)

Financial Institutions

Financial institutions prioritize transaction security and business continuity, while attaching comparatively less importance to energy efficiency indicators such as PUE. Accordingly, the evaluation framework should: adjust the weight allocation by reducing the dominance of single efficiency indicators and strengthening indicators linked to financial risk management, such as carbon intensity and renewable energy utilization; refine indicator definitions by clarifying their “green finance relevance,” and promote standardized disclosure formats for energy and environmental data, making them directly applicable to credit and investment decisions; develop financial-sector-specific evaluation templates that integrate green indicators into credit approval and investment assessment processes.

(2)

Internet enterprises

Internet companies emphasize computational efficiency and energy use within data centers, but pay relatively less attention to supply chain emissions and e-waste management. Therefore, it is recommended that the framework: introduce tailored sub-indicators, such as “energy consumption per unit of computing power” and “e-waste recycling rate,” to capture industry-specific practices; promote standardized GHG accounting and reporting protocols across the ICT sector, requiring firms to disclose a broader scope of emissions; provide industry-oriented performance scorecards to guide internet companies toward balanced improvements in computing efficiency, renewable energy use, and green supply chain practices.

(3)

Consulting and research institutions

Consulting and research organizations often stress comprehensiveness and methodological rigor, but may neglect practical applicability at the execution level. To improve operational effectiveness, the framework should: engage these institutions in drafting detailed implementation guidelines that specify calculation boundaries, data sources, and applicability conditions for each indicator; establish cross-industry green data-sharing platforms, coordinated by research institutions, to unify data formats and ensure consistency; support the development of standardized evaluation tools or software that translate national indicators into user-friendly scoring systems, coupled with training and consulting services to promote adoption.

(4)

Integrated perspective

Beyond sector-specific tailoring, the evaluation framework should also emphasize overarching principles: Ensure a balance between energy efficiency, environmental impact, and social value; introduce a dynamic calibration mechanism to update weight assignments every 12–18 months in line with technological advances and industry practices; incorporate long-term sustainability objectives into annual evaluations, thus forming a closed-loop system from short-term improvements to medium- and long-term green development goals.

Through these differentiated yet integrated recommendations, the evaluation framework can maintain alignment with national standards while accommodating sectoral diversity, thereby improving both its practical applicability and adaptive capacity across industries.

5. Conclusions

By applying the Analytic Hierarchy Process (AHP), this study elicited sector-specific preferences for green data-center indicators across four domains—finance, internet, research/design, and technical consulting—so that sector-level composite weights that reflect how each industry prioritizes criteria could be derived. Because these weights indicate where China’s national framework aligns and where discrepancies remain, the analysis shows that uniform application without industry-specific calibration would likely introduce systematic bias and misguide decisions. The principal findings are as follows:

Compared with the weights in the national green data-center evaluation framework, energy efficiency was assigned a lower relative importance in the financial sector. In contrast, a greater weight was assigned to the efficient utilization of computing resources and to green, low-carbon development. In the internet sector, the efficient utilization of computing resources was given the most significant emphasis among all industries. In research and design institutions, greater importance was attached to scientific layout and intensive construction. Technical consulting firms focus more on cabinet resource utilization.

It was essential for a comprehensive evaluation system of green data centers that indicator selection and weight assignment were further refined, which enabled more precise sector-specific assessment. Based on this study, three design measures were recommended: a business-oriented approach that tailored indicators and weights to each industry’s operational needs; inclusion of an integrated-benefits dimension that considered energy cost, environmental impact, and social benefit jointly; a dynamic calibration mechanism at the goal level that updated and optimized the system as technology and social conditions evolved.

Because green data-center evaluation was complex and multidimensional, multiple real-world determinants had to be jointly considered. It was recommended that subsequent research test how regional heterogeneity shaped needs and priorities; clarify high-disagreement indicators—such as cabinet-resource utilization—through larger, longitudinal datasets and mechanism-oriented analysis; assess and integrate the effects of emerging technologies into the framework; and, by balancing energy efficiency, environmental impact, and social benefits, develop a more comprehensive, context-adaptive system that accommodated diverse institutional settings.