Examining Data Center Energy Consumption and Power Sources

Data centers consume a lot of power — the DOE reported that US data centers consumed 4.4% of total US electricity in 2023. They’re projecting demands of 6.7% - 12% by 2028. Given the increasingly catastrophic effects of climate change, governments, residents, regulatory agencies, and nonprofit organizations are concerned about the impact of data centers on the environment. In light of that, data center operators are taking steps to reduce power consumption and improve data center sustainability.

These efforts come at a time when demand for data center resources is skyrocketing. In 2016, we entered the “Zettabyte Era,” with global Internet traffic reaching a trillion gigabytes per year, or an average of 96 exabytes per month. Traffic now exceeds 33 exabytes (1 billion gigabytes) per day. Each of the world’s 6.4 billion mobile and 1.4 billion fixed Internet connections consumes an average of 4.2 gigabytes per day.

Data center operators must consider these demands when developing power management strategies. It’s also important to understand how power is consumed and what climate-friendly power sources are available.

Data Center Power Usage

Industry experts use mathematical models to estimate data center consumption. One authoritative study estimated that data centers accounted for as much as 1.5 percent of global energy use in 2010, and extrapolations predicted that power consumption would skyrocket. That didn’t happen: Improved energy efficiency of IT equipment, coupled with server virtualization and cloud migration, caused power consumption to remain relatively constant.

However, those experts didn’t anticipate the rise of AI. The graphics processing units (GPUs) and other accelerators needed to support AI workloads require an enormous amount of power. One rack of AI-capable equipment can consume 30kW to 100kW, depending on the number and type of GPUs. A large AI data center has power demands of 100 megawatts or more.

So far, AI has not had a big impact on power usage in most data centers. The average data center rack still consumes about 7 kW at a cost of up to $30,000 a year. However, there is little question that data center power usage will rise as more data centers adopt AI. 

Average Data Center Power Consumption

Overall Energy Usage:

• Small data centers will consume around 1-5 Megawatts (MW) of power.
• Large hyperscale data center consumption varies significantly based on scale, complexity, and application. Consumption can range from 20 MW to 100 MW and beyond. The largest reported consumer is China Telecom’s Inner Mongolia Information Park. That facility consumes at least 150 MW annually.

Per Rack Energy Usage:

• Numbers vary by report, but 7-10 kW per rack is still the estimated average power consumption per rack in a data center. 

Average AI Data Center Power Consumption

Overall Energy Usage:

• AI data centers consume an enormous amount of energy. Estimates for large facilities start at 100 MW per year. That number could be much higher given infinite power availability. Current estimates are capped by power constraints.
• The International Energy Agency estimates that a typical AI data center can use as much power as 100,000 homes. 

Per Rack Energy Usage:

• AI data centers can use 60 kW+ per rack. That’s six times higher than the average of standard data center rack usage.

With power usage increasing by the year, it’s more important than ever to track performance metrics and identify sources of inefficiency. It all starts with understanding how much power the various elements of the IT infrastructure consume.

We’ll break down the components of a data center contributing most to overall energy consumption below:

Servers

Servers account for an estimated 40 percent of data center power use. More powerful servers consume more electricity, but today’s servers are more efficient than previous generations. The key is to optimize the hardware components within a server and use power management features to dynamically adjust power usage based on workload demands. Consolidating workloads on fewer, more powerful servers can also reduce overall power usage.

Storage

Storage devices consume a negligible amount of data center power. The number of storage devices has decreased as capacity has increased, and solid-state drives (SSDs) use far less power than legacy hard disk drives. Increased use of SSDs can further reduce power consumption.

Network

Calculating network power consumption has been complicated by the increasing use of wireless networks. However, networks use less than 5 percent of data center power. Efficiencies can be gained by using fewer, higher-capacity devices.

Infrastructure

Data center infrastructure consumes 54 percent of total power, with the vast majority of that used by cooling and airflow systems. Strategies such as aisle containment and in-row cooling can improve thermal management, enabling higher setpoint temperatures and reducing energy consumption. Increasingly, data centers are also implementing liquid cooling to efficiently manage thermal loads in high-density data centers.

Data center infrastructure covers a lot of ground. Operators need to keep tabs on their power distribution units and uninterruptible power supplies to ensure facilities are running efficiently. 

Miscellaneous Infrastructure

Aside from these four buckets, a lot of miscellaneous infrastructure within the data center uses power. It’s important to keep tabs on everything to maximize power usage efficiency. For example, power distribution systems and UPSs should be investigated as sources of power loss.

Data Center Power Sources

Most data centers use the traditional electric grid for power. The only issue is the electric grid is finite. Utility providers in major markets such as Northern Virginia and Silicon Valley have voiced concerns about available capacity and, in some cases, halted data center development until they can improve infrastructure. Power availability is now the primary factor for data center site selection.

Large hyperscale data centers are turning to alternative power sources to relieve the strain on the grid and reduce greenhouse gas emissions. In fact, corporations like Amazon, Meta, Microsoft, and Google are regularly the largest purchasers of green energy power purchase agreements (PPAs) as they strive to scale, maximize uptime, and minimize greenhouse gas emissions.

Microgrids

Microgrids are small electrical networks that can supplement the main power grid with renewables, hydrogen and other energy sources. They can also operate independently from the grid to improve reliability, reduce costs, and meet sustainability objectives.

Wind and Solar

Wind and solar energy are increasingly used to power data centers, providing clean energy and significantly reducing their carbon footprint. Due to the intermittent nature of these sources, data centers typically combine them with other sources and battery storage to maintain a consistent power supply.

Hydrogen 

Hydrogen fuel cells provide a clean and sustainable power source, with only water as a byproduct. Although hydrogen accounts for about three-fourths of the known universe, most is contained in other compounds. Extracting it requires energy and precious metals. Economies of scale need to be established for hydrogen to become more widely adopted. Still, hydrogen fuel cells and gas turbines hold promise for a greener energy future.

Geothermal

Geothermal energy uses heat from within the earth to create steam to generate electricity. Geothermal plants require significant upfront investments, but they require less land than solar or wind farms and can be placed virtually anywhere. The U.S. is the world leader in geothermal energy production, but one-fourth of Europe could tap this power source.

Nuclear

The concept of nuclear-powered data centers is gaining viability by the day. Small modular reactor (SMR) technology is an emerging source of clean, reliable energy. SMRs use nuclear fission to release heat from uranium pellets to produce up to 300 MW of electricity. They offer an affordable option for use in microgrids and isolated areas with limited resources. However, SMR technology is relatively new, and a regulatory framework is still being developed.

How the Enconnex InfiniRack Can Help

The Enconnex InfiniRack data center cabinet was designed to adapt to the needs of nearly any data center and is ready to handle ever-growing power densities. Its structural design also maximizes airflow and available useable space. Let us customize an InfiniRack for your data center. The options are virtually limitless. Get in touch today.