Dive Brief:
- Electricity demand from data centers could more than double over the next decade, threatening U.S. grid reliability and increasing operational risks for other businesses, the Conference Board said in a May 30 report.
- Due to limited grid and transformer capacity and long permitting and interconnection timelines for new generation and transmission projects, expected data center load growth could slow the pace of the United States’ transition to a net-zero electricity system, the report said.
- “More positively, the broad adoption of AI tools might unlock greater efficiencies in the way electricity is used, as well as generated, stored and transported,” wrote authors Alexander Heil, senior economist at the Conference Board’s Economy, Strategy and Finance Center, and Ivan Pollard, leader for the Conference Board’s Marketing and Communications Center.
Dive Insight:
Global energy consumption from data centers, including those devoted to AI and cryptocurrency mining, could jump from 460 TWh in 2022 to more than 1,000 TWh in 2026, according to an International Energy Agency projection cited by Heil and Pollard.
In the United States, data centers’ electricity consumption could double to more than 9% of total U.S. generation by 2030, the Electric Power Research Institute said in May.
Some models expect even more dramatic growth over the next five to six years, said Josh Claman, CEO of Accelsius, a provider of computer chip cooling technology.
High-end predictions for data center energy consumption “sound like hype, but we can’t discount them yet,” Claman said.
Accelsius and competitors like LiquidStack and Marathon Digital Holdings develop direct-to-chip cooling systems that use non-water refrigerants that transition from liquid to vapor to maximize cooling efficiency. Data center operators are quickly realizing that this type of system offers the longest technological runway for AI and high-performance computing applications, Claman said.
Cooling accounts for roughly 40% of data centers’ total energy consumption today, but two-phase liquid chip-cooling systems like Accelsius’s can reduce server racks’ power consumption by 50% or more, and the industry as a whole will become more efficient as more data centers transition from air- to water-based cooling, Claman said.
Still, demand for AI could outpace those efficiency improvements and sustain the recent rapid growth in data centers’ absolute electricity consumption, he said. Though improvements in silicon-based chip technology or the emergence of reliable quantum computers could dramatically reduce the energy demands of high-performance computing, current research on those fronts is likely a decade or more from commercialization, meaning utilities and grid operators can’t expect them to mitigate near-term load growth, Claman added.
In the meantime, load growth attributable to data centers could slow progress toward utility, state and federal net-zero goals, the Conference Board report said.
Smaller utilities that can’t readily procure additional renewable power or quickly deploy renewable generation may be hardest-hit. Oregon’s 16,000-customer Umatilla Electric Cooperative saw its carbon emissions quadruple between 2018 and 2021 following an influx of Amazon data centers into its territory, the report noted.
Utilities with inadequate renewable capacity could also lose high-load customers with ambitious near-term net-zero goals, including technology companies and data center operators, the Conference Board said.
Such customers could “relocate…to another location where [their] environmental needs could be met,” the report said.
Utilities and data center operators can mitigate the threat to climate goals by locating facilities “next to distributed power resources such as solar, wind or even nuclear [plants to] allow low-carbon energy to be utilized directly,” the report said.
Public Service Enterprise Group is in talks with data center operators to sell power generated by two New Jersey nuclear plants, Utility Dive reported in May.
AI-driven load growth could be a boon for the U.S. nuclear industry in general, according to the Conference Board. “Any dramatic increase in electricity demand from data centers and AI might further prioritize the use of nuclear energy as an emissions-free power source,” the report said.
The report downplayed the potential for AI to drive dramatic breakthroughs in grid efficiency. Although more sophisticated AI tools could enable more efficient electricity generation, storage, delivery and use, “how much AI can offset its own increase in load demand remains an open question,” Heil and Pollard wrote.
Instead, public and private actors could offset AI-driven power demand growth through more conventional solutions like energy efficiency, peak shifting, expedited permitting and interconnection for new energy resources and broader adoption of small nuclear reactors, they said.
