During the last 20 years, data centres have experienced a tremendous shift in their architectural design and operational footprint. On-premise server rooms scale-outs. Early installations of scale-out server rooms have largely given way to large-scale colocation sites, cloud-native hyperscale facilities, and growingly decentralised edge deployment environments [5]. This transition has made record-breaking scalability and performance possible, which has led to real-time capabilities of social networking and money transactions, industrial automation, and AI model deployment. Nonetheless, with this exponential growth, further scrutiny has been placed on the energy and environmental implications of the sector.

One of the most energy-intensive elements of the digital infrastructure is data centres. An individual hyperscale data centre may quantitatively use the amount of power a mid-sized town uses as a result of two factors: constant compute requirements and the high degree of environmental control demanded [6]. The sustainability of such facilities has become a serious concern in energy systems planning, particularly against global net-zero emission surges, the European Green Deal, the Energy Star Program, and the United Nations' Sustainable Development Goals. Although technology is improving the sustainability of energy-efficient materials, cooling systems, etc., the increasing popularity of digital services puts more pressure on sustainability initiatives [7].

The most common measurements applied to quantify sustainability in data centre design and operation are standard metrics, like the Power Usage Effectiveness (PUE), a ratio between total facility energy consumption and the power consumed by the IT equipment [8]. Although the PUE provides a valuable basis of internal efficiency, it does not reveal much about the overall impact on the environment or sustainability levels among regions. In addition, it is frequently self-reported and separately computed, relative to which its cross-comparisons are hindered.

Carbon usage effectiveness (CUE), water usage effectiveness (WUE), white space utilisation and gross power provisioning metrics are other significant metrics. White space involves the physical floor space in a data centre dedicated to IT equipment [9]. White space can be efficiently utilised to improve thermal distribution, energy savings, and scalability. In the same vein, gross power capacity is a pointer to the overall energy access capacity of the infrastructure [10]. A combination of these measurement criteria sheds light upon intent to design and the reality of operation.

Although they are relevant, little is operationally known globally with the current data source scarcity, and opportunistically in a structured form. The proposed study will fill that gap by effectively leveraging a dataset comprising actual white space and gross power values across several countries so that efficiency proxies like Power per Centre and White Space per Centre can be derived, which provide a more practical way of looking at energy-service ratios.

The disparity of data availability in favour of North American and Western European data centres, in which large providers, including Google, AWS and Microsoft, publicly disclose energy performance information on some of their hyperscale sites, is one of the most critical problems to be solved in the field. Conversely, colocation facilities in the emerging economies, particularly those related to enterprise and regional ones, are characterised by a lack of energy transparency, most commonly associated with dispersed ownership, an absence of reporting requirements, or insufficient regulatory enforcement [11].

Consequently, the models available in the literature have been mainly theoretical or limited to a specific situation involving high-profile facilities [12]. Not many studies have engaged in comparative studies of data centres by region and based them solely on PUE as the only parameter, and even fewer have compared them to geospatial analysis and visualisation of these differences in data centres. This poses a central blind spot for the world regarding reducing carbon footprints in ICT systems [13]. The solution would integrate multi-region-based and multi-metric data, as well as reproducible analysis studies to draw comparison conclusions.

New frameworks such as carbon-aware scheduling [12] and distributed-energy forecasting for data infrastructures [11] reveal growing interest in coupling energy systems with digital platforms. Studies by [8], [9] highlight that global analyses still concentrate on hyperscale providers in Europe and North America, leaving developing regions empirically under-documented. While prior work largely relies on self-reported PUE ratios, this paper contributes by proposing open-data-driven, normalised indicators that integrate spatial (m²) and electrical (MW) parameters. This approach enables cross-regional benchmarking of infrastructure efficiency where direct carbon or utilisation data are unavailable.

This study helps to address the gaps identified in three different ways. First, it uses an open-source dataset encompassing more than 130 data centres serving a variety of geographies with less-represented areas like Latin America, Southeast Asia, and Africa. The fact that white space and gross power measures are included enables the computation of realistic sustainability measures that remain unreported and consolidated [14].

Second, the paper proposes the metrics Power per Center and White Space per Center that capture architectural efficiency and operational density. Comparative measures of the distributional quality of the energy resources against service capacities become possible with these indicators, which can also be used as surrogates in pairs of granular energy use or carbon footprint information [15].

Third, the research uses analytical methodology, including descriptive statistics, correlation analysis, and geospatial visualisation. The methods are combined so that they can demonstrate regularities in infrastructure intensity, gaps in utilisation, and regional imbalances. The results are relevant to the scholarly literature on sustainable energy systems, but the policy implications are threefold and direct: policymakers, utility planners, and infrastructure providers may all be interested in the results. Although much work on sustainable energy systems has been undertaken to assess renewables, transport, and industrial loads, the digital infrastructure industry has been a particularly pertinent frontier to sustainability innovation [16]. The paper provides an empirical and well-timed entry point into that debate, based on practical measurements and scalable methods.

Figure 1 methodology schema introduces a step-by-step design of the workflow evaluation of the sustainability of data centres worldwide. Phase 1 starts by loading a CSV file containing the information about data center characteristics in regions. Phase 2 concerns data engineering, cleaning and calculating efficiency measures such as power per centre and whitespace ratios. Phase 3 performs analytical activities, including Exploratory data analysis, geospatial mapping with a choropleth plot and understanding the regional disparities through correlation and a bar chart. Lastly, Phase 4 will provide usable insights, identify infrastructure inequities, and recommend green IT policies and a possible Global Data Infrastructure Sustainability Index (GDISI). This framework is characterised by clarity, reproducibility and policy-relevance.

Figure 1.

System Architecture

The analysis in this study is based on a publicly available dataset titled “Global Data Centre Energy Footprints” sourced from the Kaggle open data platform. The dataset represents a snapshot of conditions in 2021–2022. Each record corresponds to either a single country or a multi-country region (e.g., Pacific). It was obtained from the Kaggle open-access repository, which aggregates publicly disclosed and crowd-verified infrastructure data. Cross-checks and basic consistency tests were applied to remove duplicates and non-standard units. White space denotes the IT-equipment floor area (m²) available for computing operations, while gross power represents total installed electrical capacity (MW), including cooling and redundancy systems. After cleaning, 135 valid records were retained for analysis.

The preprocessing started by removing a redundant secondary header row and renaming columns to simplify and make them consistent. Coercive parsing was used to convert the non-numeric fields to make them compatible with the analytical functions. Empty rows or rows having null values in important columns like Country, Region, etc., or Available Gross Power columns were excluded to preserve the integrity of comparative analysis. After cleaning all major world regions, the working data set of 135 records was developed.

When comparing was necessary, the Available White Space and Available Gross Power fields were considered the total of the countries or regions. The assumption was that all the power units were in megawatts (MW), and the white space was expressed in the standard data centre floor space units (square meters). This was generally documented in the dataset and checked in internal consistency functions.

Two indicators based on derived measures of efficiency were created to reflect efficiency in two perspectives, namely, spatial and energy provisioning:

1. White Space per Center

   This metric was computed as:

   W avg = W ws n (1)

   It is the average space assigned to a single data centre in any particular country/region. Higher values can correspond to broad infrastructure underutilised, and lower values can represent compact, high-density cores.

2. Power per Center

   This metric was computed as:

   P avg = A gp n (2)

   It can be substituted for the average electrical capacity of provisioning per facility, including cooling, UPS, and redundancy facilities. This indicator shows the scale of the operations and the energy design.

These calculated indicators used in regional comparison were plotted with raw variables to help interpret trends in sustainability, resource intensity, and design efficiency.

Python with Jupyter Notebooks on Google Colab was used to run the analysis. The following libraries were applied: Pandas and NumPy: working with tabular data, getting descriptive statistics and collecting values. Matplotlib and Seaborn: These will be used to create bar plots, scatter plots, and heat maps, and they should be used in static visualisations. GeoPandas: To map the geospatial data with shapefile and GeoJSON layers.

The geographic analysis was conducted using the Natural Earth GeoJSON dataset, which was processed through GeoPandas. The available gross power was used to shade areas at the country level and show intensity hotspots in all regions. Countries that did not report their data were marked in grey to indicate a lack of reporting.

Although the dataset offers unprecedented access to the infrastructure level variables, several limitations are to be mentioned: There are no real-time operational measures; The data is a point in time and does not consider dynamic utilisation or temporal performance. Lack of PUE or direct carbon emissions: The study treats gross power and white space as surrogates of sustainability, which may not comprehensively reflect environmental performance. Imbalanced regional coverage: Certain regions (e.g., Africa, Southeast Asia) lack entries, which may result in the underestimation of their digital infrastructure intensity. Nevertheless, the dataset can be helpful in sustainability benchmarking and policy framing or where granular energy data is absent.

The descriptive statistics were generated on significant variables as proxies to global data centre infrastructure as the first step of the analysis. These factors are the number of data centres in each country, square meters of available white space and gross power in megawatts. The findings showed a high level of disparity throughout the data. There was only 1 data centre to as many as 2,052 data centres in a country. The mean was about 46.2 centres per country; however, the standard deviation of 184.2 defines a highly skewed distribution toward the right side and is dominated by a few countries, each with tremendous infrastructure deployment requirements.

The same level of differences arose in white space availability. The average amount of white space available per country was estimated at 13.2 m², but once again, this parameter showed a great variety of measures, with some countries having over 800 m² and many having either close to zero or no value. Available gross power reflected the trend. The mean gross power at the country level was about 6.84 MW, and it varied between 0 and the highest value of 407 MW. The interquartile range (0 to 2.5 MW) further affirmed the predominance of the fact that most of the countries are running on low electrical capacity.

All these numbers point to an international situation in which there is an excessive concentration of digital energy resources in a few countries, mainly North American and European, with minimal capacity or reporting in most other countries.

The data were categorised into geographic regions, and the total number of centres and the gross power were calculated. It was also found that most data centres are in the Americas, with 2,470 in the region. It was immediately followed by Europe, which had the number of 2,330 centres, whereas Asia had a smaller number of 752 centres. By contrast, Sub-Saharan Africa, the Pacific, the Middle East, and North Africa (MENA) regions had only 301, 232 and 154 centres, respectively.

Under energy provisioning, the Americas once again ranked first with a total available gross power of 449 MW, while Europe and Asia recorded 337 MW and 72 MW, respectively. These findings denote a high level of digital infrastructure concentrated in more advanced economies, based on regional variation, data sovereignty, the uptake of cloud, and investment in green computing opportunities (Figure 2).

Figure 2.

Total data centres by region. Higher counts in the Americas and Europe reflect mature investment and regulatory frameworks

The scatter plot that provided the gross power and the space of the available white area confirmed a broadly positive correlation between spatial and electrical provisioning. The outline of the significant outliers was not observed, though the United States was evidently in the lead in both axes. There is a linear trend, but the range of the points shows that there are cases of countries that have followed a pattern with space-efficient and high-density server design that enables comparatively high power utilisation in a small area (Figure 3).

Figure 3.

White Space vs Gross Power. Positive correlation indicates coupling of spatial and energy provisioning; deviations imply design inefficiency

The visual representation implied two inferences about the region. In Europe, several countries have moderate white space and significant gross power, which is representative of the high densities of buildings. In other Sub-Saharan African and Latin American countries, however, there were greater physical footholds and somewhat lower levels of power, indicating that the infrastructure is scalable, with electrical extensions possible. This relationship shows how infrastructural design is usually affected by the locally available energy prices, cooling technology, and regulatory schemes.

The correlation matrix produced using three key indicators: the total data centres, available white space, and available gross power, indicated the strong positive relations. Pearson correlations between each pair of variables were above 0.99, implying an organisational relationship between the number of facilities, the total area they occupied, and power demand. This ensures the internal consistency of the dataset, but at the same time, the correlation also indicates that volume-driven metrics are overused in sustainability measures (Figure 4).

Figure 4.

Correlation Matrix. Strong associations among facility count, floor space, and power capacity (r > 0.99) confirm internal data consistency

Notably, there were outlier economies in these strong associations. In nations with smaller data centre territory, the correlations were uneven across smaller economies, and the relationship between space and energy was more jagged. This observation supports the argument of having more local benchmarks and metrics derived that normalise the infrastructure indicators at the level of a specific facility.

A heatmap of gross power availability globally showed how concentrated data centre energy provisioning is geographically. Dominating an uncanny degree of blistering colour levels, North America and Western Europe illustrate a huge gross power value. Besides having the broadest scope of hyperscale and colocation facilities, the regions enjoy the advantageous grid infrastructure and energy transition to renewable energy (Figure 5).

Figure 5.

Available Gross Power by Country. Map showing concentration of energy-intensive digital capacity; grey areas denote missing data

By comparison, large geographic regions, such as most of Africa, Central Asia and the Pacific, reliably recorded low amounts of power or did not disclose it fully. Such disparity in spatial distribution highlights the features of the infrastructural and data transparency disparity of the emerging markets. Even some countries that do not have power data might have small-scale or under-reported facilities that the mainstream data does not cover. It bears both regional carbon accounting and digital equality implications.

To address the shortcomings of raw sums, two indicators derived by efficiency were ascertained: white space per data centre and gross power per data centre. These indicators rest the totals of infrastructure provisions on the counts of facilities and help us see how architecture, in general, and energy supply to a given facility, on a particular scale.

The analysis revealed that Europe performs better on the two indicators, averaging 0.19 square meters of white space and 0.13 MW of gross power per centre. This shows the region's popularity of well-tuned hyperscale facilities and energy-sensitive rules. Asia came next with a similar white space per centre (0.18 m²), yet relatively few gross power per centre (0.12 MW), potentially caused by the lack of space and hybrid deployments (Figure 6). When interpreted jointly, Power per Centre and White Space per Centre reveal the Degree of architectural optimisation. High power with low space indicates dense hyperscale facilities; conversely, high space but low power suggests under-utilised infrastructure. Policymakers can distinguish between regions constrained by energy access and those constrained by design efficiency.

Figure 6.

Efficiency Metrics per Region. Europe’s higher ratios denote optimised, large-scale facilities; the Pacific’s lower ratios indicate fragmented capacity

Americas, which had the most significant total number of centres and power, had a low power per centre average of only 0.03 MW. This alludes to the prevalence of miniature or archaic information centres in Latin America and the south of the United States. Sub-Saharan Africa and MENA had moderate efficiency scores, indicating emerging yet deployable deployments. The region in the Pacific had the lowest ratios, indicating infrastructural fragmentation and poor provision of energy.

The histogram of the global skew in digital power distribution provided a distribution of the available gross power. Most countries were well concentrated in the 0 to 10 MW range, and those above 100 MW were just a few. Such a skewed distribution further underpins that a few economies create most of the global ecosystem's digital infrastructure footprint (Figure 7).

Figure 7.

Distribution of Available Gross Power (MW). Skewed distribution reveals dominance of a few high-capacity economies and global inequality in digital energy resources

The long-tailed pattern shows that policies that aspire to enhance the sustainability of data centres must be separated according to country size and infrastructure maturity. Already developed economies should adopt carbon mitigation and energy reuse, and emerging ones should adopt modular renewable-ready infrastructure development.