The unstoppable growth of energy consumption in data centers
Data center energy consumption has continued to grow due to the increasing demand for information technologies by businesses and individuals. According to the International Energy Agency (IEA), data centers consumed approximately 415 TWh in 2024, representing about 1.51 trillion metric tons (T3T) of global electricity consumption. Projections indicate that this figure could double to 945 TWh by 2030, driven primarily by the widespread adoption of artificial intelligence.
Current efforts to reduce energy consumption focus on the operation of IT equipment and cooling systems, and include expanding the operating temperature ranges for IT equipment and increasing the cooling system's set points. Cooling, for example, accounts for between 301 and 501 TW of the total energy consumed in a data center.
Where does the energy go in a data center?
Most estimates suggest that between 501 and 551 TP3T of energy is used for supporting infrastructure—chillers, transformers, UPS, lighting, and security systems. The Green Grid suggests that only 301 TP3T of energy could actually be converted into useful computing work, depending on how many infrastructure elements (chillers, transformers, CRAC, etc.) are present in the facility.
The goal is clear: to create a data center that consumes the least amount of energy possible for any given operation or activity, but to do so without sacrificing security, protection, or availability of IT operations.
One of the most effective ways to measure and improve this efficiency is through the indicator PUE (Power Usage Effectiveness).
What is PUE and why is it essential?
PUE is the globally recognized standard for comparing the energy efficiency of data centers. It is calculated as the ratio between the total energy consumed by the facility and the energy used exclusively by IT equipment.
An ideal PUE of 1.0 would mean that all energy is used directly by IT systems, with no losses in supporting infrastructure. In practice, this is unattainable, but it sets the benchmark.
What is the current average PUE?
According to the Uptime Institute's 2024 global survey, the industry average PUE is at 1.56, This value has remained relatively stable for the past five consecutive years. This stagnation, however, masks significant progress in newer and larger installations: many recent builds consistently achieve a PUE of 1.3 or even lower. Leading hyperscalers like Google report a PUE of 1.09, while Microsoft is targeting a design PUE of 1.12.
The economic impact is considerable: for a 10 MW data center, reducing the PUE from 1.6 to 1.3 can save more than 26 million kWh annually, equivalent to approximately $1.3 million per year in electricity costs.
Design criteria for energy efficiency
The following are the fundamental criteria that should be considered when designing —or redesigning— a data center with a focus on energy efficiency.
Site selection
The location of the data center is one of the most crucial decisions for its long-term efficiency. Factors such as the following should be evaluated:
-
Climate: Regions with low temperatures allow for greater use of free cooling, drastically reducing the need for mechanical cooling. Intel demonstrated that, with air economizers and server inlet temperatures set to 35°C, the free cooling It was possible during all hours of the year except 39, achieving a PUE of just 1.07.
-
Proximity to water: necessary for water cooling systems and cooling towers.
-
Air quality: It affects the viability of direct air economizers and filtration requirements.
-
Energy sources: access to renewable energy, stability of the electricity grid and energy costs in the region.
Building construction and layout
The building's architectural design directly influences operational efficiency. Key aspects include:
-
Building materials: Proper thermal insulation of floors, roofs, and walls reduces cooling effort. The use of reflective facades or green roofs is recommended.
-
Multipurpose/multi-tenant vs. purpose-built: A data center designed exclusively for this purpose allows for the optimization of every aspect of the infrastructure.
-
Modular data center: Containerized and scalable designs allow for on-demand growth without over-dimensioning the initial infrastructure, adjusting capacity to actual needs.
-
Layout of operational and support spaces: Intelligent distribution minimizes the travel distances of air, water, and energy.
-
Use of alternative energy: Integration of sources such as solar, wind, or thermal storage. While it doesn't directly improve PUE, it significantly reduces the carbon footprint and aligns with ESG strategies.
-
High-performance construction methods: Certifications such as LEED, ISO 50001 and energy efficiency metrics validate best construction practices.
Selection of energy-efficient IT equipment
Choosing energy-efficient servers, storage devices, and network equipment is a fundamental pillar. Some best practices include:
-
Prioritize servers with higher processing capacity and lower consumption, preferably with ENERGY STAR certification.
-
Use power supplies with the 80 PLUS Titanium standard, which achieve efficiencies higher than the 92%. At $0.12 USD/kWh, savings can range from $2,000 to $6,000 per rack per year.
-
Consolidate loads through virtualization and containers, which reduces both direct electricity consumption and the need for refrigeration.
Selection of efficient infrastructure devices
Electrical distribution components have a direct impact on energy losses in the data center:
-
Transformers high efficiency with lower heat losses.
-
UPS Systems with efficiency of 96% or higher, operating in eco or double conversion mode.
-
Power distribution units (PDUs) and smart power strips with rack or phase monitoring.
Deployment decisions: open racks vs. closed cabinets
The choice between open racks and closed IT cabinets affects airflow management and cooling efficiency. Closed cabinets with containment systems facilitate the separation of hot and cold air, improving climate control efficiency.
Selection of efficient refrigeration equipment
Cooling is the largest energy consumer in a data center's infrastructure. Design options include:
-
Overhead, perimeter, or in-row mounted air management systems (in-row): Each configuration has advantages depending on the power density and the rack layout.
-
Minimize the distance between the cooling source and the equipment: Reducing the distance air, water, or energy traveled between the source and the point of use decreases losses and increases system efficiency.
-
Liquid cooling systems: For high-density environments such as AI/HPC, direct-to-chip liquid cooling is becoming the new standard.
Design of mechanical systems to optimize the operating environment
The mechanical design of the data center must optimize temperature and humidity conditions according to the guidelines of ASHRAE TC 9.9, which recommends server inlet temperatures between 18°C and 27°C for Class A equipment, and between 18°C and 22°C for high-density Class H1 systems (such as AI accelerators).
Raised floors vs. slab
The upper floors (raised floor) allow for controlled distribution of cold air from below, while above-ground designs with overhead or in-row cooling eliminate plenum inefficiencies and can be more effective in high-density configurations.
Equipment layout: hot aisle / cold aisle
The alternating hot and cold aisle configuration is a standard practice that improves airflow efficiency by preventing the mixing of cold air with warm return air.
Aisle containment systems
Physical containment of aisles—whether hot or cold—is one of the most effective measures for improving PUE. These systems prevent the recirculation of hot air toward IT equipment and allow cooling systems to operate more efficiently.
Economizing operation: air or water side
Economizers allow the so-called “free cooling”, Taking advantage of external environmental conditions to cool the data center without relying exclusively on mechanical cooling. There are two main methods:
-
Air economizer (airside): It replaces the hot return air with cooler, filtered, and treated outside air. It can reduce annual cooling consumption by more than 60%.
-
Water saver (waterside): It uses cooling towers to produce cold water without activating the chiller, achieving maximum savings during periods of low outside temperature.
Integration of intelligent management systems
Measurement, reports and controls
You can't improve what you don't measure. Implementing temperature and power consumption sensors by zone or rack, along with alarm systems and predictive analytics, is essential to maintaining a stable PUE and preventing deviations.
DCIM: Infrastructure Management Software
The use of tools Data Center Infrastructure Management (DCIM) allows you to monitor, track, and manage both IT assets and infrastructure components—racks, cooling units, power distribution—from a centralized platform. It is recommended that:
-
Minimize the use of proprietary or single-task-specific software protocols.
-
Opt for platforms with a common language that can interact with other systems.
-
Integrate DCIM with a building information management system (BIM, Building Information Management), which allows you to create a digital twin (digital twin) of the facility for impact analysis, capacity planning and change management.
Studies indicate that automating monitoring and control can reduce infrastructure consumption by up to 10% on a continuous basis.
Complementary efficiency practices
Cable management and airflow
Power and data cabling should not be run in areas intended for airflow, such as under a raised floor. Cables can create obstructions that disrupt air circulation, generating hot spots and reducing the efficiency of the cooling system.
Efficient lighting
While lighting represents a small fraction of total energy consumption, every improvement counts. It is recommended to install high-efficiency LED lighting with motion and occupancy sensor controls, so that light is only on when and where needed.
Towards a more efficient data center: a comprehensive vision
Designing an energy-efficient data center depends not on a single measure, but on the strategic integration of multiple criteria—from site selection to management software—working in a coordinated manner. In a context where data center energy demand could represent almost half of the growth in electricity demand in the United States by 2030, every improvement in PUE and every kilowatt saved has a significant economic and environmental impact.
The key lies in adopting a holistic approach: combining intelligent site selection, advanced cooling technologies, efficient IT equipment, containment systems, energy economizers, and monitoring tools such as DCIM and BIM. This results in a data center that not only consumes less energy but also transforms the largest possible proportion of that energy into useful computing work—without compromising availability or operational security.
Key references consulted: ASHRAE TC 9.9 Thermal Guidelines, Uptime Institute Global Data Center Survey 2024, IEA “Energy and AI” Report 2025, US Department of Energy Best Practices Guide.(Uptime Institute)