The Impact of AI on Data Center Energy Efficiency

Data centers were already struggling with the level of electricity demand. What needs to change with the advent of AI?

Data centers are crucial for modern businesses, but their energy consumption has risen sharply. They need substantial power, which drives up costs and global energy use. With growing AI demand, data center energy consumption is projected to increase by over 20 percent each year. This highlights the importance of sustainable practices and efficient energy management to enhance efficiency while cutting costs and environmental impact.

Which industry produces more emissions? Airlines or data centers?

When businesses need to manage data-dependent processes, they rely on data centers. These centers are crucial for data-driven companies, containing servers, storage and infrastructure. They handle billions of data transactions daily, resulting in significant energy costs.

Data centers use as much power as large industries, often going unnoticed. For context, in 2024, airlines will account for 2 percent of global carbon emissions [2], while data centers already exceed this at about 3 percent [3]. Predictions suggest that by the decade's end, extensive AI use could push data center energy consumption to 9 percent of global energy [4].

AI cranks up the wattage

Along with the excitement around the benefits of AI, there is much chatter re the energy demands that come with it. Processing AI workloads consumes more energy for many different reasons, some of which include:

• Data volume: For AI models, particularly deep learning algorithms, to train effectively, they need large datasets. The larger the dataset, the more computational power is required to handle it.
• Complex algorithms: AI models need significant computational power due to their intricate nature.
• 24/7 operation: AI systems frequently operate around the clock to manage real-time data and deliver immediate responses, which increases energy usage.
• Cutting-edge hardware: AI depends on high-performance hardware like GPUs (Graphics Processing Units), which consume significant amounts of energy. These components are crucial for handling extensive datasets and complex algorithms.

Therefore, data centers incorporating AI use significantly more energy. In the United States alone, data center electricity consumption is projected to triple from 2022 to 2030 [5]. AI usage in data centers consumes significantly more energy than in the past two decades. For example, one Google query consumes about 0.3 watt-hours per request, while one ChatGPT request consumes ten times more energy at 2.9 watt-hours per request [4].

Data centers consume more than just energy

All data centers require large-scale cooling systems. Servers and routers consume tremendous amounts of energy to manage data, generating significant heat. High-capacity cooling is essential to dissipate this heat and keep the equipment from overheating. These cooling systems can be nearly as energy-intensive as the IT equipment they support.

According to an article by the International Journal of Refrigeration, up to 45% of energy used in a data center can go towards cooling [6]. Data centers require the use of water to optimally cool their overheating servers with air condition systems, chillers, and/or humidifiers. Cooling servers through water usage is more efficient than air conditioning entire rooms, and a tandem combination of water and air cooling is the optimal choice for maximum data center cooling energy savings.

Data centers' high water usage can strain local resources. In some places, like The Dalles, Oregon, Google's data centers use over 25 percent of the city's water—a fact revealed only after a legal battle [7]. With growing AI demands, water consumption is set to rise as more cooling is needed. Many data centers don't disclose their actual electric and water usage, but The Dalles shows a notable recent increase.

Data centers require significant square footage for servers, often covering 100,000 square feet, comparable to a Manhattan city block. The largest US data center, The Switch Tahoe Reno, spans 7.2 million square feet, surpassing the Pentagon's size [8]. The global number of data centers has surged, from under 8,000 in January 2021 to 10,978 by November 2023, and it is expected to continue rising due to AI demands, requiring even more land [9].

Rising data center costs and carbon emissions: Impact on affordability?

Operating and maintaining data centers has become more costly. Between 2014 and 2020, electricity prices in US cities rose only slightly [12]. However, this trend reversed in 2021, with a significant increase in the cost per kilowatt-hour, as shown in the graph below. Companies unprepared for increased energy consumption and rising costs will face tough financial choices.

Capital expenditure now can result in valuable operational expense savings and improved ESG scores in the future

Reducing energy use in data centers benefits both businesses and investors. Given their massive energy and water consumption, utility expenses can be high. Implementing efficient cooling systems, using renewable energy, and optimizing data centers can help cut these costs.

Companies aiming to cut operational costs can explore financial strategies within data centers. Upgrading emissions efficiency can involve significant capital investments, such as installing liquid cooling systems. Although these projects require substantial initial outlays, they can ultimately reduce ongoing operational expenses.

In addition to electricity savings, businesses can benefit from increased investor performance and stakeholder happiness when "going green." Organizations often set goals for "Net Zero Carbon Emissions by 2050" [11] or decreased emissions output by a specific date to show investors their particular business is committed to sustainability. This is heavily regulated, as the six biggest banks in the United States were required to show the Federal Reserve their risks associated with achieving net zero by 2050 and climate change preparations [12]. Many do not actively publish emissions numbers, but regulations are getting stricter; a new California law requires large corporations to reveal carbon emissions by 2026 [13]. In the EU, even stronger standards are being implemented to regulate companies' ESG calculations [14]. With data centers accounting for significant emissions for companies, any improvements will tremendously help corporations meet their increasingly regulated ESG numbers.

Areas of research and exploration for data center energy efficiency

A novel idea introduced by WindCORES, a German renewable energy company, is to create data centers inside wind turbines [15]. This company stacked server racks up to 150 meters high inside wind turbines in a western Germany wind park, successfully creating working data centers and effectively rendering them nearly carbon neutral. Other cutting-edge ideas, such as underwater data centers [16], are being researched to find new ways to combat the energy-consuming nature of data centers

AI itself can also help drive energy efficiency

While AI advancement is spearheading the energy requirement boom, that very same advancement can offer tools to create a more efficient data center. WWT has been testing innovative AI solutions to address growing computing loads for years. Tools with inbuilt AI models trained to monitor and learn server power usage and CPU utilization patterns show promising possibilities for meeting server workloads more efficiently. Optimization of Power Utilization Effectiveness and other key metrics can help streamline data center power consumption. Advanced tools are being used to identify solution areas for energy efficiency within data centers and cooling efficiency.

Conclusion

The rise of AI has brought about a substantial increase in data center energy consumption, posing significant challenges for sustainability. However, innovative solutions and efficient energy management practices can help mitigate these challenges. By adopting cutting-edge technologies and optimizing energy usage, data centers can reduce their environmental impact and contribute to a more sustainable future. It is crucial for businesses to invest in these solutions to ensure the long-term viability and efficiency of their data centers.

Take steps to understand the impact of proposed changes beforehand. Given the increasing focus on data center sustainability and the increased energy consumption that AI will drive, data center operators must now focus heavily on their energy consumption forecasts. 

WWT can assist operators in comprehending the comprehensive effects of any suggested modifications by offering complete carbon footprint assessments. This enables an evaluation of the possible impacts on net zero objectives or other commitments due to proposed changes. WWT conducts an in-depth analysis of current sustainability commitments, examines existing data on power usage, and contrasts this with the proposed hardware's consumption and emissions.

Works cited