Data centers are the backbone of our digital world, but their energy consumption is massive. On average, data centers use between 100 to 200 kWh (kilowatt hour) per hour, which is 72,000 to 144,000 kWh per month. To put this into perspective, the average American home consumes 10,500 kWh of energy per year. This means that one month of data center energy can power roughly 7 to 13 homes for an entire year.
The sheer scale of energy consumption by data centers has significant financial and environmental impacts. As demands surge, driven by factors like AI and cloud computing, the number and size of data centers are expanding worldwide, prompting concerns about their carbon footprint.
This article delves into the environmental impact of data center power consumption and explores both existing and proposed methods for making these giant facilities more energy-efficient.
The exact energy consumption of data centers may vary by size, location, and the number of servers it houses. According to the International Energy Agency (IEA), worldwide data center energy consumption was 260 to 360 TWh (terra watt hour) in 2022. This accounted for 1 to 1.15% of global energy consumption.
The United States, home to most hyperscale data centers, has over 5,300 such facilities, making it the leading country with the most data centers. In 2019, the US data center load was 19 GW, and it is projected to grow to 21 GW this year, with estimates suggesting it could reach 35 GW by the end of the decade. This indicates that US data centers currently account for over 4% of the total US electricity demand, a figure that could rise to 9% by 2030.
While the United States Data Center Energy Usage Report doesn’t necessarily classify the consumption by resource, i.e. clean vs non-clean energy usage, the numbers highlight the rapid growth and energy demands of data centers nationwide.
The rise of cloud computing has played a significant role in this increase. As more businesses, big and small, migrate their data to large facilities, the number of power-hungry servers and racks grows, driving up energy consumption.
Looking ahead, the power consumption of data centers is set to soar. The IEA Energy 2024 report predicts that by 2026, data center energy consumption could double, reaching an astonishing 1,000 TWh. A report from Newmark estimates that US data center energy consumption will rise to 35 GW by 2030, while McKinsey puts the estimate at 50 GW. This increase is largely driven by the growing adaptation of Artificial Intelligence (AI) and cryptocurrency mining.
Artificial Intelligence (AI) is revolutionizing industries, but it comes at a hefty energy price tag. AI workloads demand a lot of computing power, with hundreds of GPUs in a single server performing thousands of computations. Machine learning is an energy-intensive technological process.
The sharp rise in generative AI development since 2022 has increased data center power requirements. Training an AI model, for example, can produce a staggering 600 000 lbs of CO2 emissions - compare that to the average person’s annual emissions of 11 000 lbs.
As AI technologies become more widespread, the environmental toll of their energy consumption raises significant concerns. Utility companies in the United States are already bracing for surges in the power needs of data centers.
Like a home or office, data centers require power for a variety of purposes. Servers, storage systems, lighting, and especially cooling systems all require power. Most of the energy consumption of a data center is used by cooling systems, sometimes due to inefficiencies, like operating at much cooler temperatures than required. The American Society of Heating, Refrigerating, and Air-Conditioning Engineers (ASHRAE) recommends a range of 64.4 to 80.6 degrees Fahrenheit to optimize energy use.
Inefficient cooling technologies, like computer room air conditioning, are particularly energy-intensive. Adopting more efficient cooling systems could significantly reduce overall energy consumption.
Carbon emissions are a pressing global concern, with data centers contributing to the problem through their energy consumption. Although the devices in these centers don’t directly produce carbon dioxide, the electricity they consume does, especially when produced from fossil fuels.
According to the IEA, data centers were responsible for 330 Mt CO2 emissions, which is approximately 1% of global energy-related emissions. A data center's carbon footprint depends directly on its energy usage and source. Facilities powered by renewable energy sources, like wind or solar, would have a much smaller footprint.
The good news is that many companies are making strides towards sustainability. Hyperscalers, like Equinix, Scala, and Google have invested in clean energy for data centers to reduce their carbon footprint. However, as the emission figures show, most of the industry still relies on fossil fuels - and with the rise of AI, energy demands from data centers are set to increase, potentially leading to higher carbon emissions if renewable energy is not adopted more widely.
Improving efficiency is the key to reducing energy consumption and emissions in data centers. While switching to clean energy is the ideal solution, it’s not always feasible due to the location and/or cost. However, there are several practical measures that data centers can implement to enhance their energy efficiency:
Cooling is one of the largest energy consumers in data centers. By adopting more efficient cooling methods, like liquid cooling or natural/free-air cooling, data centers can significantly reduce their energy use. For instance, liquid cooling can dissipate energy much more efficiently than traditional cooling systems, while free-air cooling leverages natural air to maintain optimal temperatures in colder climates. Another viable option is temperature optimization, which entails keeping temperatures between 64.4 and 80.6 degrees Fahrenheit, and has been successfully implemented in Singapore with the world’s first standard for operating data centers in tropical climates.
Efficient use of hardware and software resources can also improve efficiency and reduce energy consumption. Shutting off extra, or unused, servers consolidating smaller servers, and increasing virtualization are all strategies that can improve efficiency. For example, running multiple virtual machines on a single physical server will optimize resource use and improve server utilization rates, leading to maximized efficiency and reduced energy consumption.
As more tech companies move towards sustainability, hardware manufacturers are making energy efficiency a design priority. Data centers should prioritize purchasing Energy Star-rated devices when refreshing any kind of equipment, which, while seemingly minor on an individual basis, can result in a favorable power usage effectiveness (PUE) ratio and collectively lead to major energy savings.
Making your data center more energy efficient is an investment that has several positive outcomes:
Even if regulations don’t mandate it, going green is a smart move for data centers. Starting with energy efficiency not only benefits the environment but also makes good economic sense. By exploring innovative cooling solutions, optimizing resource use, and investing in energy-efficient equipment, data centers can significantly reduce their environmental impact!
Learn more about sustainable practices for data centers!
Data centers generate a significant amount of heat, typically from servers, storage, and networking equipment. Although exact numbers can vary, we can use energy consumption to estimate heat. For example, a data center consuming 1 MW of power will generate approximately 1 MW of heat.
To calculate power consumption in a data center, you can use the formula:
Power Consumption (kW) = Total Power Draw of IT Equipment (kW) + Total Power Draw of Supporting Infrastructure (kW).
This includes the power used by servers, storage devices, networking equipment, cooling systems, and other supporting infrastructure like power distribution units and lighting.
China Telecom’s data center facility in Hohot, Inner Mongolia is considered the top-energy consuming data center in the world with an annual consumption of 150 MW (Mega Watt).
Data centers consume a lot of energy because of the high power demands of running and cooling thousands of servers and associated equipment around-the-clock. Servers require significant electrical power to perform computations, store data, and manage network traffic. On top of that, the heat generated by these operations requires cooling systems, which also consume a substantial amount of energy.
The need for maintaining uptime also contributes to energy usage, as redundant systems and backup power supplies are needed.