AI is poised to drive a doubling in data center electricity demand by 2030, studies say
Energize Weekly, June 5, 2024
Data centers, spurred by increased artificial intelligence applications, could double their demand for electricity in the U.S. reaching 9.1 percent of the nation’s electricity load by 2030, according to analysis by the Electric Power Research Institute (EPRI).
“Data centers are one of the fastest growing industries worldwide,” according to the report by EPRI, a nonprofit research group funded by the utility industry.
“Between 2017 and 2021, electricity used by Meta, Amazon, Microsoft, and Google – the main providers of commercially available cloud computing and digital services – more than doubled,” the analysis said.
The EPRI study gave a range of increased data center electricity demand from a low of 4.6 percent to a high of 9.1 percent, compared to 4 percent today.
A Goldman Sachs report projects U.S. data center demand reaching 8 percent of total electricity load in 2030 compared with 3 percent in 2022.
“U.S. utilities will need to invest around $50 billion in new generation capacity just to support data centers alone,” Goldman Sachs said.
The demand of computing service has been growing, but the game changer in terms of electricity load has been the advent of AI, which demands an order of magnitude more computing power.
“On average, a ChatGPT query needs nearly 10 times as much electricity to process as a Google search,” Goldman Sach said. “In that difference lies a coming sea change in how the U.S., Europe, and the world at large will consume power – and how much that will cost.”
Between 2015 and 2019 data center workloads tripled, but electricity demand remained flat at about 200 terawatts-hours a year, thanks to growing efficiencies in computing and data center management. Since 2020, efficiency gains have dwindled, according to Goldman Sachs.
Now AI is going to drive a 160 percent increase in data center power demands by 2030. “By 2028, our analysts expect AI to represent about 19 percent of data center power demand,” Goldman Sachs said.
To accommodate AI and cloud computing, data centers are also getting bigger. “It is not unusual to see new centers being built with capacities from 100 to 1000 megawatts – roughly equivalent to the load from 80,000 to 800,000 homes,” EPRI said.
And while a neighborhood of that scale takes years to plan and build, data centers can be developed and connected to the internet in one to two years, creating an almost instant bump in demand.
The one- to two-year lead for connecting to the grid, with a demand of highly reliable power, often with the goal of getting that power from renewable or other non-emitting generation, “can create local and regional supply challenges,” EPRI said.
Compounding those challenges is the geographic concentration of data centers, with 15 states accounting for approximately 80 percent of the 2023 data center load with, in descending order, Virgina, Texas, California, Illinois and Oregon having the greatest loads.
Other states with clusters of data centers include Arizona, Iowa, Georgia, Washington, Pennsylvania, New York and New Jersey.
“Data centers favor sites where internet connections are strong; where electricity prices, land costs, and disruptive events are low; where skilled labor is available; near population centers and users; and where the centers can develop backup power to ensure power supply (usually natural gas or diesel generators),” EPRI said.
