Data center

Facility used to house computer servers From Wikipedia, the free encyclopedia

A data center is a facility used to house computer systems and associated components, such as telecommunications and storage systems.^[1]^[2] Data centers are critical infrastructure for the storage and processing of information, and they support the global financial system, cloud services, machine learning, and artificial intelligence.^[3]^[4]

Since IT operations are crucial for business continuity, a data center generally includes redundant or backup components and infrastructure for power supply, data communication connections, environmental controls (e.g., cooling, fire suppression), and various security devices. Large data centers operate at an industrial scale, requiring significant energy. Estimated global data center electricity consumption in 2024 was around 415 terawatt hours (TWh), or about 1.5% of global electricity demand.^[5] The IEA projects that data center electricity consumption could double by 2030.^[5] The rapid growth of data center infrastructure has prompted regulatory debates in multiple jurisdictions regarding tax incentives, electricity grid impacts, water consumption, and compatibility with state and national climate commitments^[6]. High demand, driven by artificial intelligence (AI) and machine learning workloads is accelerating the deployment of high-performance servers, leading to greater power density and increased strain on electric grids.^[7]^[5]

Data centers can vary widely in terms of size, power requirements, redundancy, and overall structure. Four common categories used to segment types of data centers are onsite data centers, colocation facilities, hyperscale data centers, and edge data centers.^[8] These categories differ substantially in ownership model, scale, and energy efficiency, with hyperscale and colocation facilities collectively accounting for approximately 74% of U.S. server energy consumption as of 2023^[5], a share that has grown significantly over the past decade as workloads have migrated away from enterprise on-premises infrastructure. In particular, colocation centers often host private peering connections between their customers, internet transit providers, cloud providers,^[9]^[10] meet-me rooms for connecting customers together^[11] Internet exchange points,^[12]^[13] and landing points and terminal equipment for fiber optic submarine communication cables,^[14] which are critical to connecting the internet.^[15]

Classification and types

Data centers are usually classified according to their ownership, scale and operational purposes. Their categories are sharp indicators to reflect the differences in infrastructure designing, redundancy and intended use circumstances.

Enterprise

Enterprise data centers are owned and operated by a single organization for their own internal IT needs, rather than for commercial hosting of other companies' data.^[16]^[17]^[18] They used to be essential infrastructures as in 2014, enterprise data centers accounted for over 60% of the U.S. server energy consumption, but this share fell sharply to approximately 10% by 2023 as workloads migrated to hyperscale and colocation facilities.^[19] Energy efficiency at an enterprise data center tends to be significantly lower than at hyperscale facilities due to its cooling, which can account for over 30% of electricity consumption at enterprise sites, compared to roughly 7% at efficient hyperscale data centers.^[20] This shift away from those enterprise facilities toward cloud and colocation infrastructure is one of the defining structural trends of the previous decade.^[21]

Colocation

This type of data center provides shared physical infrastructure, such as power, cooling, space, etc., to multiple tenants^[16] and their role allows customers to lease space rather than building their own facilities,^[17] which allows businesses to set up their own servers in a shared professional infrastructure rather than maintaining on-premises infrastructure.^[16] In addition, colocation data centers reduces expenditure for organizations while enabling operators to achieve economies of scale in power procurement and cooling efficiency.^[20] Colocation data centers accounted for approximately 22% of worldwide data center capacity as of 2024, making them the second-largest category by capacity after hyperscale facilities.^[21] Major colocation providers include Equinix, and Digital Realty, which operate facilities across multiple continents.

Hyperscale

Hyperscale data centers are large-scale facilities, typically exceeding 100 megawatts of power capacity, designed to support massive, scalable computing workloads for cloud services, AI training, and large-scale data processing.^[18]^[20] As of the end of 2024, there were 1,136 operational hyperscale data centers globally, which a figure that has doubled over five years, with the United States accounting for approximately 54% of total hyperscale capacity.^[21] In particular, the three largest operators, Amazon Web Services, Microsoft Azure, and Google Cloud, collectively account for approximately 59% of all hyperscale data center capacity globally.^[21] As a growing trend, hyperscale data centers represented approximately 41% of worldwide data center capacity in 2024 and are projected to exceed 60% by 2029 as enterprise on-premises infrastructure continues to decline.^[20] Moreover, as artificial intelligence is becoming widely used, those AI-focused hyperscale data centers are being built more rapidly, and they can consume as much electricity as 100,000 households, significantly more than conventional data centers.^[19]

Edge

Edge data centers are smaller facilities, typically ranging from 1 to 10 megawatts,^[22] and they are positioned closer to end users or data sources in order to reduce network latency and support real-time applications^[20] such as autonomous vehicles, industrial automation, and content delivery. Unlike hyperscale data centers, edge data centers are intentionally distributed across many locations rather than consolidated, with deployments occurring in urban areas, cell tower sites, as well as industrial locations.^[20] Over the recent years, the growth of edge computing infrastructure is closely tied to the expansion of 5G wireless networks and the increasing use of Internet of Things (IoT) devices,^[20] which generate data that is more efficiently processed near its source than transmitted to a centralized data center.

All in all, the distribution of computing workloads across these four categories has shifted dramatically over the past decade, with hyperscale and colocation collectively accounting for approximately 74% of U.S. server energy consumption in 2023, up from less than 40% in 2014; this share is projected to reach 85% by 2028.^[19] However, these classifications are not mutually exclusive from each other and they may overlap depending on the operational structure and service model of the particular data center.

More information Type, Typical scale ...

Common data center types and typical characteristics^[19]^[20]^[21]
Type	Typical scale	Primary purpose	Key characteristic
Enterprise	1–10 MW	Internal IT operations	Owned and operated by a single organization
Colocation	10–100 MW	Shared hosting for multiple clients	Customers lease space, power, and cooling
Hyperscale	100 MW+	Cloud services and AI workloads	Massive scalability; operated by major cloud providers
Edge	1–10 MW	Low-latency local processing	Distributed near end users or data sources

History

NASA mission control computer room c. 1962

Data centers have their roots in the huge computer rooms of the 1940s, typified by ENIAC, one of the earliest examples of a data center.^[23]^{[note 1]} Early computer systems, complex to operate and maintain, required a special environment in which to operate. Many cables were necessary to connect all the components, and methods to accommodate and organize these were devised such as standard racks to mount equipment, raised floors, and cable trays (installed overhead or under the elevated floor). A single mainframe required a great deal of power and had to be cooled to avoid overheating. Security became important – computers were expensive, and were often used for military purposes.^[23]^{[note 2]} Basic design guidelines for controlling access to the computer room were therefore devised.

During the microcomputer industry boom of the 1980s, users started to deploy computers everywhere, in many cases with little or no care about operating requirements. However, as information technology (IT) operations started to grow in complexity, organizations grew aware of the need to control IT resources. The availability of inexpensive networking equipment, coupled with new standards for the network structured cabling, made it possible to use a hierarchical design that put the servers in a specific room inside the company. The use of the term data center, as applied to specially designed computer rooms, started to gain popular recognition about this time.^[23]^{[note 3]}

During the dot-com bubble (1997–2000), there was a boom in data center construction, when companies needed fast Internet connectivity and non-stop operation to establish their Internet presence.^[24] Many companies started building very large facilities, called internet data centers (IDCs),^[25] which provide enhanced capabilities, such as crossover backup: "If a Bell Atlantic line is cut, we can transfer them to ... to minimize the time of outage."^[25]

The global data center market saw steady growth in the 2010s, with a notable acceleration in the latter half of the decade. According to Gartner, worldwide data center infrastructure spending reached $200 billion in 2021, representing a 6% increase from 2020 despite the economic challenges posed by the COVID-19 pandemic.^[26]

The latter part of the 2010s and early 2020s saw a significant shift towards AI and machine learning applications, generating a global boom for more powerful and efficient data center infrastructure. As of March 2021, global data creation was projected to grow to more than 180 zettabytes by 2025, up from 64.2 zettabytes in 2020.^[27]

In 2024, global energy consumption from data centers was 620 TWh, with 80% of that consumption being concentrated in the United States, China, and the United Kingdom.^[4]

In the United States

The United States is currently the foremost leader in data center infrastructure, hosting 5,381 data centers as of March 2024, the highest number of any country worldwide.^[28] According to global consultancy McKinsey & Co., U.S. market demand is expected to double to 35 gigawatts (GW) by 2030, up from 17 GW in 2022.^[29] As of 2023, the U.S. accounts for roughly 40 percent of the global market.^[29] In 2025, it was estimated that the U.S. GDP growth was only 0.1% without the investments in data centers for artificial intelligence.^[30]

A study published by the Electric Power Research Institute (EPRI) in May 2024 estimates U.S. data center power consumption could range from 4.6% to 9.1% of the country's generation by 2030.^[31] As of 2023, about 80% of U.S. data center load was concentrated in 15 states, led by Virginia and Texas.^[31]

Data center design

Racks of telecommunications equipment in part of a data center

Data centers house critical computing resources in a controlled environment and must generally operate with very high availability. Key design elements include providing power for the equipment, temperature and humidity control, cabling, fire safety, and security.^[32]

Information security is also a concern, and for this reason, a data center has to offer a secure environment that minimizes the chances of a security breach.

Obsolescence and modernization

Industry research company International Data Corporation (IDC) puts the average age of a data center at nine years old.^[33] Gartner, another research company, says data centers older than seven years are obsolete.^[34] The growth in data (163 zettabytes by 2025^[35]) is one factor driving the need for data centers to modernize.

Focus on modernization is not new: rapid obsolescence of data center equipment was a concern by at least 2007,^[36] and in 2011 Uptime Institute was concerned about aging equipment.^{[note 4]}

Industry standards

The Telecommunications Industry Association's Telecommunications Infrastructure Standard for Data Centers^[37] specifies the minimum requirements for telecommunications infrastructure of data centers and computer rooms including single tenant enterprise data centers and multi-tenant Internet hosting data centers. The topology proposed in this document is intended to be applicable to any size data center.^[38]

Telcordia GR-3160, NEBS Requirements for Telecommunications Data Center Equipment and Spaces,^[39] provides guidelines for data center spaces within telecommunications networks, and environmental requirements for the equipment intended for installation in those spaces. These criteria were developed jointly by Telcordia and industry representatives. They may be applied to data center spaces housing data processing or Information Technology (IT) equipment. The equipment may be used to:

Operate and manage a carrier's telecommunication network
Provide data center based applications directly to the carrier's customers
Provide hosted applications for a third party to provide services to their customers
Provide a combination of these and similar data center applications

Reliability of electrical power supply

Power supplies, either back up or continuous onsite power consists of one or more uninterruptible power supplies, battery banks, diesel, gas turbine, gas engine generating sets.^[40] Greater primary fuel energy efficiency can be achieved with the use of cogeneration technology, generating electricity, heating and cooling onsite.^[41]

To prevent single points of failure, all elements of the electrical systems, including backup systems, are typically given redundant copies, and critical servers are connected to both the A-side and B-side power feeds.^[42] This arrangement is often made to achieve N+1 redundancy in the systems.^[43] Static transfer switches are sometimes used to ensure instantaneous switchover from one supply to the other in the event of a power failure.^[44]

Low-voltage cable routing

Options for low voltage cable routing might include; Data cabling that is routed through overhead cable trays;^[45] Raised floor cabling, both for security reasons and to avoid the extra cost of cooling systems over the racks; Smaller/less expensive data centers may use anti-static tiles instead for a flooring surface.

Environmental control

Maintaining suitable temperature and humidity levels is critical to preventing equipment damage caused by overheating. Overheating can cause components, usually the silicon or copper of the wires or circuits to melt, causing loose connections and fire hazards. Typical temperature control methods include:

Air conditioning
Indirect cooling, such as the use of outside air,^[46]^[47]^[48] indirect evaporative cooling units, and seawater cooling.

Airflow management is the practice of achieving data center cooling efficiency by preventing the recirculation of hot exhaust air and by reducing bypass airflow. Common approaches include hot-aisle/cold-aisle containment and the deployment of in-row cooling units which position cooling directly between server racks to intercept exhaust heat before it mixes with room air.^[49]

Humidity control not only prevents moisture-related issues: importantly, excess humidity can cause dust to adhere more readily to fan blades and heat sinks, impeding air cooling leading to higher temperatures.^[50]

Aisle containment

Cold aisle containment is done by exposing the rear of equipment racks, while the fronts of the servers are enclosed with doors and covers. This is similar to how large-scale food companies refrigerate and store their products.

Computer cabinets/Server farms are often organized for containment of hot/cold aisles. Proper air duct placement prevents the cold and hot air from mixing. Rows of cabinets are paired to face each other so that the cool and hot air intakes and exhausts do not mix air, which would severely reduce cooling efficiency.

Alternatively, a range of underfloor panels can create efficient cold air pathways directed to the raised-floor vented tiles. Either the cold aisle or the hot aisle can be contained.^[51]

Another option is fitting cabinets with vertical exhaust duct chimneys.^[52] Hot exhaust pipes/vents/ducts can direct the air into a Plenum space above a Dropped ceiling and back to the cooling units or to outside vents. With this configuration, traditional hot/cold aisle configuration is not a requirement.^[53]

Fire protection

Data centers feature fire protection systems, including passive and active design elements, as well as implementation of fire prevention programs in operations. Smoke detectors are usually installed to provide early warning of a fire at its incipient stage.

Although the main room usually does not allow Wet Pipe-based Systems due to the fragile nature of circuit boards, there still exist systems that can be used in the rest of the facility or in cold/hot aisle air circulation systems that are closed systems, such as:^[54]

Sprinkler systems
Misting, using high pressure to create extremely small water droplets, which can be used in sensitive rooms due to the nature of the droplets.

However, there also exist other means to put out fires, especially in Sensitive areas, usually using Gaseous fire suppression, of which Halon gas was the most popular, until the negative effects of producing and using it were discovered.

Security

Physical access is usually restricted. Layered security often starts with fencing, bollards and mantraps.^[55] Video camera surveillance and permanent security guards are almost always present if the data center is large or contains sensitive information. Fingerprint recognition mantraps are starting to be commonplace.

Logging access is required by some data protection regulations; some organizations tightly link this to access control systems. Multiple log entries can occur at the main entrance, entrances to internal rooms, and at equipment cabinets. Access control at cabinets can be integrated with intelligent power distribution units, so that locks are networked through the same appliance.^[56]

Data center transformation

Data center transformation takes a step-by-step approach through integrated projects carried out over time. This differs from a traditional method of data center upgrades that takes a serial and siloed approach.^[57] The typical projects within a data center transformation initiative include standardization/consolidation, virtualization, automation and security.

Data center consolidation consists in reducing the number of data centers^[58]^[59] and avoiding server sprawl (both physical and virtual),^[60] often includes replacing aging data center equipment. Likewise, this process is aided by standardization which makes these systems follow a uniform set of configurations in order to simplify and improve efficiency.^[59] Automating tasks such as provisioning, configuration, patching, release management, and compliance are other ways in which data centers can be upgraded. These changes are needed not just when facing fewer skilled IT workers.^[61] Lastly, security initiatives integrate the protection of virtual systems with existing security of physical infrastructures.^[62]

Raised floor

The first raised floor computer room was made by IBM in 1956 to allow access for wiring.^[63] During the 1970s, raised floors became more common because they allow cool air to circulate more efficiently.^[64] A raised floor standards guide (GR-2930) was developed by Telcordia Technologies, a subsidiary of Ericsson.^[65]

Lights out

The lights-out^[66] data center, also known as a darkened or a dark data center, is a data center that, ideally, has all but eliminated the need for direct access by personnel, except under extraordinary circumstances. Because of the lack of need for staff to enter the data center, it can be operated without lighting. All of the devices are accessed and managed by remote systems, with automation programs used to perform unattended operations. In addition to the energy savings, reduction in staffing costs and the ability to locate the site further from population centers, implementing a lights-out data center reduces the threat of malicious attacks upon the infrastructure.^[67]^[68]

Noise levels

Generally speaking, local authorities prefer noise levels at data centers to be "10 dB below the existing night-time background noise level at the nearest residence."^[69]

OSHA regulations require monitoring of noise levels inside data centers if noise exceeds 85 decibels.^[70] The average noise level in server areas of a data center may reach as high as 92–96 dB(A).^[71]

Residents living near data centers have described the sound as "a high-pitched whirring noise 24/7", saying "It's like being on a tarmac with an airplane engine running constantly ... Except that the airplane keeps idling and never leaves."^[72]^[73]^[74]^[75]

External sources of noise include HVAC equipment and energy generators.^[76]^[77]

A typical server rack, commonly seen in colocation

Design criteria and trade-offs

Availability expectations: The costs of avoiding downtime should not exceed the cost of the downtime itself^[78]
Site selection: Location factors include proximity to power grids, telecommunications infrastructure, networking services, transportation lines and emergency services. Other considerations should include flight paths, neighboring power drains, geological risks, and climate (associated with cooling costs).^[79]
- Often, power availability is the hardest to change.

High availability

Various metrics exist for measuring the data-availability that results from data-center availability beyond 95% uptime, with the top of the scale counting how many nines can be placed after 99%.^[80]

Modularity and flexibility

Modularity and flexibility are key elements in allowing for a data center to grow and change over time. Data center modules are pre-engineered, standardized building blocks that can be easily configured and moved as needed.^[81]

A modular data center may consist of data center equipment contained within shipping containers or similar portable containers.^[82] Components of the data center can be prefabricated and standardized which facilitates moving if needed.^[83]

Dynamic infrastructure

Dynamic infrastructure^[84] provides the ability to intelligently, automatically and securely move workloads within a data center^[85] anytime, anywhere, for migrations, provisioning,^[86] to enhance performance, or building co-location facilities. It also facilitates performing routine maintenance on either physical or virtual systems all while minimizing interruption. A related concept is Composable Infrastructure, which allows for the dynamic reconfiguration of the available resources to suit needs, only when needed.^[87]

Side benefits include

reducing cost
facilitating business continuity and high availability
enabling cloud and grid computing.^[88]

Software/data backup

Non-mutually exclusive options for data backup are:

Onsite
Offsite

Onsite is traditional,^[89] and one of its major advantages is immediate availability.

Offsite backup storage

Data backup techniques include having an encrypted copy of the data offsite. Methods used for transporting data are:^[90]

Having the customer write the data to a physical medium, such as magnetic tape, and then transporting the tape elsewhere.^[91]
Directly transferring the data to another site during the backup, using appropriate links.
Uploading the data "into the cloud".^[92]

Energy use

Energy consumption is a central issue for data centers. Power draw ranges from a few kilowatts (kW) for small server racks to several tens of megawatts (MW) for large facilities. Modern hyperscale data centers can exhibit power densities exceeding 100 times those of conventional office buildings, primarily due to the high concentration of servers and cooling systems required to manage continuous digital workloads.^[94] For higher power density facilities, electricity costs are a dominant operating expense and account for over 10% of the total cost of ownership (TCO) of a data center.^[95]

As of 2024, data centers in the United States are primarily powered by natural gas, which supplies 40% of their electricity (with renewable energy at 24%, nuclear at about 20% and coal at about 15%).^[96] The Associated Press reported that electricity for AI data centers in the United States would likely come from natural gas or oil, as companies prefer using currently available power plants, which primarily use fossil fuels. Fossil energy is also often cheaper in locations where data centers are developed, and experts believe that energy demands from generative AI and data centers would be difficult to fulfill with renewable energy alone. Some companies such as Google, Amazon and Meta have expressed interest in nuclear power for their data centers.^[97] Other data centers, including xAI's Colossus, OpenAI's Stargate, and Meta's Prometheus use their own off-grid natural gas plants.^[98]^[99] Electric vehicle and lithium-ion batteries have also been used for powering data centers, including for Colossus.^[100]^[101]

Power utility companies make upgrades to their infrastructure to handle demands of new data centers, and the price for these changes typically falls on consumers:^[96]^[97] smaller businesses or individual households.^[96]

In December 2025, the Federal Energy Regulatory Commission published a unanimous order allowing data centers in the United States to have a direct connection with power plants.^[102] United States Secretary of Energy Chris Wright expressed support for un-retiring coal plants to power AI data centers.^[103] Trump paused leasing for offshore wind projects, a decision that Gizmodo criticized due to their potential to provide power to AI data centers.^[104] Electricity demands from AI data centers have slowed or reverssed the retirement of peaking power plants in the United States.^[105]

Greenhouse gas emissions

In 2024, data centers are estimated to account for about 1.5% of global electricity consumption (approximately 415 TWh) and around 1% of greenhouse gas emissions according to U.S. Environmental Protection Agency (EPA).^[106] However, the rapid expansion is causing projections to rise sharply. Due to the accelerated demand from AI, data center's global electricity consumption is projected to more than double to around 945 TWh by 2030 in the IEA's base-case scenario, which represents just under 3% of 2030 total global electricity consumption.^[5] This growing electricity demand, much of which is still generated by fossil fuels, increases the potential environmental impact.^[107] They also said that lifecycle emissions should be considered, that is including embodied emissions, such as in buildings.^[108]

Global data center carbon dioxide emissions are projected to rise from an estimated 220 million tonnes in 2024 to 300–320 million tonnes by 2035.^[109] Google and Microsoft now each consume more power than some fairly big countries, surpassing the consumption of more than 100 countries.^[110] As a result, there is increasing industry pressure for decarbonization. Companies are pursuing direct clean energy agreements, such as Tencent who has pledged to be carbon neutral by 2030,^[111] and Microsoft's 2024 agreement to re-open the Three Mile Island nuclear power plant to provide 100% of the electric power for its AI data centers for 20 years.^[112]

Economic analysis of energy use

Energy efficiency and overhead

Main article: Power usage effectiveness

The most commonly used energy efficiency metric for data centers is power usage effectiveness (PUE), calculated as the ratio of total power entering the data center divided by the power used by IT equipment.

\mathrm {PUE} ={{\mbox{Total Facility Power}} \over {\mbox{IT Equipment Power}}}

=1+{{\mbox{Non IT Facility Energy}} \over {\mbox{IT Equipment Energy}}}

PUE measures the percentage of power used by overhead devices (cooling, lighting, etc.). The average U.S. data center has a PUE of 2.0,^[113] meaning two watts of total power (overhead + IT equipment) for every watt delivered to IT equipment. State-of-the-art data centers are estimated to have a PUE of roughly 1.2.^[114] Google publishes quarterly efficiency metrics from its data centers in operation.^[115] PUEs of as low as 1.01 have been achieved with two phase immersion cooling.^[116]

The U.S. Environmental Protection Agency has an Energy Star rating for standalone or large data centers. To qualify for the ecolabel, a data center must be within the top quartile in energy efficiency of all reported facilities.^[117] The Energy Efficiency Improvement Act of 2015 (United States) requires federal facilities—including data centers—to operate more efficiently. California's Title 24 (2014) of the California Code of Regulations mandates that every newly constructed data center must have some form of airflow containment in place to optimize energy efficiency.

The European Union also has a similar initiative: EU Code of Conduct for Data Centres.^[118]

Efficiency improvements and renewable energy integration are helping offset some emissions, but fossil fuels remain a major electricity source for data center operations worldwide.^[119]

In 2011, server racks in data centers were designed for more than 25 kW and the typical server was estimated to waste about 30% of the electricity it consumed. The energy demand for information storage systems is also rising. A high-availability data center is estimated to have a 1 MW demand and consume $20,000,000 in electricity over its lifetime, with cooling representing 35% to 45% of the data center's total cost of ownership. Calculations show that in two years, the cost of powering and cooling a server could be equal to the cost of purchasing the server hardware.^[120] Research in 2018 has shown that a substantial amount of energy could still be conserved by optimizing IT refresh rates and increasing server utilization.^[121] Research for optimizing task scheduling is also underway, with researchers looking to implement energy-efficient scheduling algorithms that could reduce energy consumption by anywhere between 6% to 44%.^[122]

In 2011, Facebook, Rackspace and others founded the Open Compute Project (OCP) to develop and publish open standards for greener data center computing technologies. As part of the project, Facebook published the designs of its server, which it had built for its first dedicated data center in Prineville. Making servers taller left space for more effective heat sinks and enabled the use of fans that moved more air with less energy. By not buying commercial off-the-shelf servers, energy consumption due to unnecessary expansion slots on the motherboard and unneeded components, such as a graphics card, was also saved.^[123] In 2016, Google joined the project and published the designs of its 48V DC shallow data center rack. This design had long been part of Google data centers. By eliminating the multiple transformers usually deployed in data centers, Google had achieved a 30% increase in energy efficiency.^[124] In 2017, sales for data center hardware built to OCP designs topped $1.2 billion and are expected to reach $6 billion by 2021.^[123]

Power and cooling analysis

Power is the largest recurring cost to the user of a data center.^[125] Cooling at or below 70 °F (21 °C) wastes money and energy.^[125] Furthermore, overcooling equipment in environments with a high relative humidity can expose equipment to a high amount of moisture that facilitates the growth of salt deposits on conductive filaments in the circuitry.^[126]

A power and cooling analysis, also referred to as a thermal assessment, measures the relative temperatures in specific areas as well as the capacity of the cooling systems to handle specific ambient temperatures.^[127] A power and cooling analysis can help to identify hot spots, over-cooled areas that can handle greater power use density, the breakpoint of equipment loading, the effectiveness of a raised-floor strategy, and optimal equipment positioning (such as AC units) to balance temperatures across the data center. Power cooling density is a measure of how much square footage the center can cool at maximum capacity.^[128] The cooling of data centers is the second largest power consumer after servers. The cooling energy varies from 10% of the total energy consumption in the most efficient data centers and goes up to 45% in standard air-cooled data centers.

Energy efficiency analysis

An energy efficiency analysis measures the energy use of data center IT and facilities equipment. A typical energy efficiency analysis measures factors such as a data center's Power Use Effectiveness (PUE) against industry standards, identifies mechanical and electrical sources of inefficiency, and identifies air-management metrics.^[129] However, the limitation of most current metrics and approaches is that they do not include IT in the analysis. Case studies have shown that by addressing energy efficiency holistically in a data center, major efficiencies can be achieved that are not possible otherwise.^[130]

Computational Fluid Dynamics (CFD) analysis

This type of analysis uses sophisticated tools and techniques to understand the unique thermal conditions present in each data center—predicting the temperature, airflow, and pressure behavior of a data center to assess performance and energy consumption, using numerical modeling.^[131] By predicting the effects of these environmental conditions, CFD analysis of a data center can be used to predict the impact of high-density racks mixed with low-density racks^[132] and the onward impact on cooling resources, poor infrastructure management practices, and AC failure or AC shutdown for scheduled maintenance.

Thermal zone mapping

Thermal zone mapping uses sensors and computer modeling to create a three-dimensional image of the hot and cool zones in a data center.^[133]

This information can help to identify optimal positioning of data center equipment. For example, critical servers might be placed in a cool zone that is serviced by redundant AC units.

Green data centers

Data centers use a lot of power, consumed by two main usages: The power required to run the actual equipment and then the power required to cool the equipment. Power efficiency reduces the first category.

Cooling cost reduction through natural means includes location decisions: When the focus is avoiding good fiber connectivity, power grid connections, and people concentrations to manage the equipment, a data center can be miles away from the users. Mass data centers like Google or Facebook do not need to be near population centers. Arctic locations that can use outside air, which provides cooling, are becoming more popular.^[134]

Renewable electricity sources are another plus. Thus countries with favorable conditions, such as Canada,^[135] Finland,^[136] Sweden,^[137] Norway,^[138] and Switzerland^[139] are trying to attract cloud computing data centers.

Singapore lifted a three-year ban on new data centers in April 2022. A major data center hub for the Asia-Pacific region,^[140] Singapore lifted its moratorium on new data center projects in 2022, granting 4 new projects, but rejecting more than 16 data center applications from over 20 new data centers applications received. Singapore's new data centers shall meet very strict green technology criteria including "Water Usage Effectiveness (WUE) of 2.0/MWh, Power Usage Effectiveness (PUE) of less than 1.3, and have a "Platinum certification under Singapore's BCA-IMDA Green Mark for New Data Centre" criteria that clearly addressed decarbonization and use of hydrogen cells or solar panels.^[141]^[142]^[143]^[144]

Energy reuse

It is very difficult to reuse the heat which comes from air-cooled data centers. For this reason, data center infrastructures are more often equipped with heat pumps.^[145]

Social and environmental impacts

Water use

The rapid expansion of AI data centers has raised significant concerns over their water consumption, particularly in drought-prone regions. According to the International Energy Agency (IEA), a single 100-megawatt data center can use up to 2,000,000 litres (530,000 US gal) of water per day—equivalent to the daily consumption of 6,500 households.^[146]^[147] Its water usage can be divided into three categories, on-site (direct usage from data centers), off-site (indirect usage from electricity), and supply-chain (water usage from manufacturing processes).^[148]

On-site water use refers to the direct water consumed by the data center for the cooling of its equipment.^[148] Water is used specifically for space humidification (adds moisture to the air), evaporative cooling systems (air is cooled before entering server rooms),^[146] and cooling towers (water is used to remove heat from the facility).^[149]

Off-site water use is the indirect water usage from the electricity generated in data centers. It is estimated that 56% of U.S. data centers' electricity comes from fossil fuels in thermal power stations, which use water to generate power via steam.^[150]

Lastly, data centers consume water through the process of AI chip and server manufacturing. These chips, specifically, consume a vast amount of ultrapure water for fabrication and cooling of semiconductor plants.^[148] While this scope of water usage is not as significant as the on-site and off-site water usage it is still a contributing factor.

Impact of water use

Since 2022, more than two-thirds of new data centers have been built in water-stressed areas, including Texas, Arizona, Saudi Arabia, and India, where freshwater scarcity is already a critical issue. The global water footprint of data centers is estimated at 560 billion litres (150×10^⁹ US gal) annually, a figure projected to double by 2030 due to increasing AI demand.^[151]^[152]

In regions like Aragon, Spain, Amazon's planned data centers are licensed to withdraw 755,720 cubic metres (612.67 acre⋅ft) of water per year, sparking conflicts with farmers who rely on the same dwindling supplies. Similar tensions have arisen in Chile, the Netherlands, and Uruguay, where communities protest the diversion of water for tech infrastructure.^[151]^[153]

Tech companies, including Microsoft, Google, and Amazon, have pledged to become "water positive" by 2030, aiming to replenish more water than they consume. However, critics argue that such commitments often rely on water offsetting, which does not address acute local shortages.^[151]^[153]

At least 59 additional data centers are planned for water-stressed U.S. regions by 2028 and AI's projected global water demand is projected to reach 6.6 billion cubic metres (1,700×10^⁹ US gal) by 2027. Arizona State University water policy expert Kathryn Sorensen questioned the data center build out, asking: "Is the increase in tax revenue and the relatively paltry number of jobs worth the water?"^[154]^[155]^[152]

Electronic waste

Data centers generate significant electronic waste (e-waste) due to the frequent replacement of hardware such as servers, GPUs, CPUs, memory, and storage devices, often every 2–5 years to meet demands for digital transformation and artificial intelligence.^[156] Globally, e-waste totaled 62 million metric tons in 2022, with generative AI projected to contribute 1.2 to 5 million metric tons annually by 2030, including valuable metals like copper and gold alongside hazardous substances such as lead and mercury.^[156] In the United States, data centers contribute to an annual loss of $10 billion in discarded e-waste value, including $4 billion in precious metals. Only 22% of global e-waste is formally collected and recycled, exacerbating environmental pollution and health risks in informal processing sites, often in developing countries where exported waste is handled.^[156]^[157] From a political ecology perspective, data center e-waste highlights power imbalances in resource management and environmental justice, as tech corporations benefit from tax incentives while externalizing costs to marginalized communities through toxic exports and landfill burdens.^[158] This ties into debates on the commons, where shared resources like metals are depleted without equitable governance. Mitigation strategies include modular hardware designs, secure data erasure for reuse, and extended producer responsibility policies to reduce waste by up to 86% in optimized scenarios.^[156]

Impact on electricity prices

The International Energy Agency expects that the AI boom could double global demand for electricity from data centers between 2022 and 2026.^[159]^[160]^[161] According to one 2025 energy model, the United States could see an increase of 8% on energy prices nationally by 2030.^[159] This has led to increased electricity prices in some regions,^[162] particularly in regions with lots of data centers like Santa Clara, California,^[163] and parts of upstate New York.^[164] Data centers have also generated concerns in Northern Virginia about whether residents will have to foot the bill for future power lines.^[165] It has also made it harder to develop housing in London.^[166] A Bank of America Institute report in July 2024 found that the increase in demand for electricity due in part to AI has been pushing electricity prices higher and is a significant contributor to electricity inflation.^[167]^[168]^[169] A Harvard Law School report in March 2025 found that because utilities are increasingly in competition to attract data center contracts from big tech companies, they are likely hiding subsidies to those trillion-dollar companies in power prices by raising costs for American consumers.^[170]

Bitcoin used up 2% of U.S. electricity in 2023.^[165]

In political ecology

Data centers have increasingly been analyzed through the lens of political ecology, which explores the intersections of power, politics, and environmental change in technological infrastructure.^[171] Scholars argue that these facilities reshape urban and rural landscapes by locating in marginalized or post-industrial areas, often repurposing abandoned sites while promising economic revitalization through job creation and tax revenues.^[171] For instance, hyperscale data centers in rural towns like Prineville, Oregon, or Luleå, Sweden, challenge traditional core-periphery dynamics but can exacerbate global inequalities in digital access and resource distribution.^[171] In the context of digital transformation, data centers enable the expansion of cloud computing, artificial intelligence, and machine learning, but at significant ecological and social costs.^[172] Critics from science and technology studies (STS) highlight how corporate sustainability pledges, such as carbon neutrality targets by companies like Microsoft and Google, often rely on offsets and renewable subsidies that privatize benefits while socializing environmental harms.^[172] This ties into debates on the commons, where data centers' intensive use of public resources (like electricity grids and water supplies) represents an enclosure, with tech firms receiving tax breaks and priority access at the expense of local communities.^[172]

Opposition to data centers

Bipartisan^[173] community resistance to data center development has grown, with residents voicing concerns about water scarcity, rising utility bills, noise pollution, and land sprawl.^[174]^[175] Data Center Watch reported that multiple projects collectively worth $64 billion had been stopped or delayed between May 2024 to March 2025.^[176] An additional $98 billion worth of projects were blocked or delayed between March and June of 2025.^[175]

Protests in the Netherlands led to a temporary national ban on new mega-centers in 2022.^[174]^[175]

Critics have pointed out that jobs created by data centers tend to be temporary or few in number.^[177]^[178]^[179] Residents have been concerned about air, water and noise pollution,^[180] as well as property devaluation,^[181] traffic,^[182] and the risk of fires.^[183] Other environmental concerns involving AI data centers include e-waste^[184] and construction materials that emit greenhouse gases such as concrete and cement.^[185]

Network infrastructure

An operation engineer overseeing a network operations control room of a data center (2006)

Communications in data centers today are most often based on networks running the Internet protocol suite. Data centers contain a set of routers and switches that transport traffic between the servers and to the outside world^[186] which are connected according to the data center network architecture. Redundancy of the internet connection is often provided by using two or more upstream service providers (see Multihoming).

Some of the servers at the data center are used for running the basic internet and intranet services needed by internal users in the organization, e.g., e-mail servers, proxy servers, and DNS servers.

Network security elements are also usually deployed: firewalls, VPN gateways, intrusion detection systems, and so on. Also common are monitoring systems for the network and some of the applications. Additional off-site monitoring systems are also typical, in case of a failure of communications inside the data center.

Modular data center

For quick deployment or IT disaster recovery, several large hardware vendors have developed mobile/modular solutions that can be installed and made operational in a very short amount of time.

Micro data center

Micro data centers (MDCs) are access-level data centers which are smaller in size than traditional data centers but provide the same features.^[187] They are typically located near the data source to reduce communication delays, as their small size allows several MDCs to be spread out over a wide area.^[188]^[189] MDCs are well suited to user-facing, front end applications.^[190] They are commonly used in edge computing and other areas where low latency data processing is needed.^[191]

Data centers in space

Design of a sun-synchronous orbit data center, they would orbit above the dawn / dusk transition of the planet.

Sun-synchronous orbit animation of AI supercomputing satellites

Data centers in space is a proposed idea to place a data center in outer space in low Earth orbit. The theoretical advantages are that of space-based solar power, in addition to aiding in weather forecasting and weather prediction computation from weather satellites,^[192] and the ability to freely scale up.^[193]

Challenges include temperature fluctuations, cosmic rays, and micrometeorites.^[192]

Notes

[1]
Old large computer rooms that housed machines like the U.S. Army's ENIAC, which were developed pre-1960 (1945), are now referred to as data centers.
[2]
Until the early 1960s, it was primarily the government that used computers, which were large mainframes housed in rooms that today we call data centers.
[3]
In the 1990s, network-connected minicomputers (servers) running without input or display devices were housed in the old computer rooms. These new "data centers" or "server rooms" were built within company walls, co-located with low-cost networking equipment.
[4]
In May 2011, data center research organization Uptime Institute reported that 36 percent of the large companies it surveyed expect to exhaust IT capacity within the next 18 months. James Niccolai. "Data Centers Turn to Outsourcing to Meet Capacity Needs". CIO magazine. Archived from the original on 2011-11-15. Retrieved 2011-09-09.