In the midst of the ongoing energy transition, data centers have emerged as key players in the realm of electricity consumption, driving demand and influencing the trajectory of power generation and distribution. According to an analysis conducted by the Electric Power Research Institute (EPRI), data centers could potentially consume between 4.6% and 9.1% of the total electricity in the United States by the year 2030.
The EPRI’s white paper titled “Powering Intelligence: Analyzing Artificial Intelligence and Data Center Energy Consumption” delineates four distinct scenarios outlining the growth of electricity consumption by data centers from 2023 to 2030. These data centers operate round the clock, necessitating significant power to sustain their operations and infrastructure. The power consumption within data centers is typically divided between IT equipment and auxiliary resources such as cooling and power conditioning systems.
In 2023, global data centers collectively consumed 7.4 GW of power, marking a notable 55% increase from the 4.9 GW consumed in the previous year as reported by Cushman & Wakefield. This surge underscores the substantial environmental impact associated with the escalating energy demands of data centers. A separate report from the Federal Energy Regulatory Commission (FERC) projects that the U.S. data center load could ascend to nearly 21 GW in the current year, up from 19 GW in 2023, with an anticipated climb to 35 GW by the end of the decade.
Presently, data centers account for over 4% of the total electricity load in the U.S., a figure that could potentially surge to 9% by 2030. This growth is primarily attributed to the escalating computing power required by artificial intelligence (AI) applications. The demands of AI queries surpass those of traditional internet searches by approximately 10 times, with the creation of AI-generated content like music, photos, and videos necessitating even more power. For instance, a typical Google search utilizes about 0.3 watt-hours (Wh), whereas a query using ChatGPT, the chatbot developed by OpenAI, consumes around 2.9 Wh as reported by EPRI.
The International Energy Agency projects that the electricity consumption of U.S. data centers will escalate from 200 TWh in 2022 to approximately 260 TWh by 2026, constituting 6% of the total power demand. Boston Consulting Group further estimates that this figure could potentially reach 7.5% by 2030. The EPRI report delves into four scenarios analyzing the future energy consumption of data centers, ranging from low-growth to high-growth trajectories.
In a low-growth scenario, data centers’ electricity consumption is expected to grow by 3.7% annually, reaching 4.6% by 2030, primarily driven by limited AI tool adoption and efficiency gains. Meanwhile, moderate and high-growth scenarios foresee annual growth rates of 5% to 15%, with data centers potentially consuming 6.8% to 9.1% of the nation’s electricity by 2030, predicated on rapid AI adoption and efficiency improvements.
EPRI underlines three key strategies to manage this burgeoning growth, including enhancing data center efficiency and flexibility, fostering better coordination between data center developers and electricity providers, and developing robust modeling tools to facilitate long-term grid investments. These measures aim to support technological advancements while upholding grid reliability and minimizing impacts on consumers.
As of March, the Electric Power Research Institute reported a total of 10,655 data centers worldwide, with approximately half situated in the United States. In 2023, about 80% of the U.S. data center load was concentrated in 15 states, with Virginia and Texas leading the pack. The challenges posed by data centers’ demands for reliable power, requests for non-emitting generation sources, and short lead times for connection pose potential strains on local and regional electricity supplies.
Virginia, for instance, witnessed data centers consuming roughly a quarter of the state’s electricity in 2023, the highest proportion in the U.S., followed by North Dakota, Iowa, Nebraska, and Oregon, each exceeding 11%. EPRI projects that in a high-growth scenario, Virginia’s data center load share could potentially rise to nearly 50%, with other states potentially approaching 20% or more. Hyperscalers and colocation centers dominate the U.S. data center landscape, with significant growth anticipated in regions such as Dallas-Fort Worth, Silicon Valley, Chicago, New York Tri-State, and Atlanta, potentially leading to a surge in power demands by 50% or more.
Across the United States, utilities are increasingly recognizing the opportunities and challenges posed by the development of data centers, particularly concerning load growth and generation investments. Analysts foresee that discussions on U.S. utility earnings in the first quarter of 2024 will be influenced by the burgeoning power demands of data centers. EPRI’s findings underscore the imperative need for strategic planning and infrastructure development to accommodate the escalating energy demands of data centers while ensuring grid reliability and aligning with environmental objectives.
David Porter, the Vice President of Electrification and Sustainable Energy Strategy at EPRI, emphasized the necessity for enhanced collaboration between major data center owners, developers, utilities, government entities, and other stakeholders to effectively meet the power needs of AI applications while ensuring the provision of reliable and affordable power to all consumers.
