LOUDOUN COUNTY, Va. — High up on a series of transmission towers that connect to a vast complex of digital data centers in Northern Virginia, power line workers are performing one of the energy industry’s most frightening jobs.
Specially trained technicians are disconnecting the live, uninsulated high-voltage wires – crackling with a dangerous 230,000 volts of electricity – and replacing them with advanced cables that will increase power delivery by 50 percent.
“We wouldn’t go down this route unless we absolutely had to,” said Matthew Gardner, vice president of transmission for Dominion Electric, the state’s largest utility, which owns and operates the transmission lines.
Dominion says they must.
Here, outside Dulles International Airport, the largest cluster of data centers in the United States is powered by Dominion’s power lines, and electricity use is soaring due to a surge in artificial intelligence-generated traffic and the expectation of more.
Dominion opted for an unusual “hotline” replacement: It couldn’t simply shut down existing lines that feed the growing data center hub’s 24/7 power needs. “It’s absolutely essential that we complete these projects to accommodate future growth,” Gardner said in an interview.
It took Dominion 115 years to reach its current power supply levels. “Rapid growth, driven by data centers, is expected to double the system load over the next 15 years,” Gardner said.
Here and in other data center clusters across the U.S., an electric grid industry struggling with a historic transition from fossil fuel generation to renewable energy is suddenly facing unprecedented demand projections from the next phase of America’s digital economy without an overall plan.
At present, the US electric power transmission industry is far behind in achieving President Joe Biden’s goal of zero-carbon generation needed to mitigate the risk of catastrophic climate events. Given the scale of federal spending on clean energy under the Biden administration, the 2024 presidential election will be a referendum on climate policy. Recent studies have shown that AI-driven energy demand poses another barrier to reducing carbon emissions when utilities rely primarily on gas and coal-fired generation to meet demand.
Arshad Mansoor, CEO of the Electric Power Research Institute (EPRI), said utility companies, grid planners, data center operators and their customers should work together to develop clean energy.
In a report to Energy Secretary Jennifer Granholm’s Energy Advisory Council, EPRI called on the nation to “transition to zero-carbon sources of power for data center operations and low-carbon technologies for backup power.”
Data centres and climate change?
In order to meet electricity demand while still fighting global warming, the plans and policies for the U.S. power grid have undergone a sudden shift.
As an example of digital technology’s voracious appetite for power, EPRI noted that currently, conducting a simple search on a laptop browser consumes about one-third of a watt of power, but as advanced AI techniques and high-performance computing train software algorithms to create answers from massive data files, electricity demand is soaring.
The EPRI report concluded that currently a complex generative AI question via ChatGPT requires 3 watts of power, 10 times the amount of a simple search, adding that a Google search with similar generative AI capabilities could require 9 watts.
Depending on how quickly AI takes hold, data center electricity demand could soar from about 4% of U.S. power grid output to more than 9% in just six years, according to EPRI.
“No one fully understands how generative AI will affect every aspect of society,” Mansoor said in an interview. “So anything we say now is likely to be wrong in six months,” he added.
San Francisco energy research firm Energy & Environmental Economics (E3) has compiled 13 analyses under various assumptions about how much power output will need to increase to accommodate data center expansion.
It’s predicted that between 20,000 and 100,000 MW of new capacity will be added by 2030. Many unknowns remain, including how “smart” generative AI will be, how much efficiency can be built into AI chips, and whether strained power grids can actually provide the power data centers need.
“There’s still a lot of uncertainty, especially around the highest end projections,” said Kushal Patel, a senior partner at E3 who co-authored the report. “The future probably won’t provide any greater certainty,” he added.
Moody’s Investors Service warned in a recent research report that consumers could end up paying higher costs if some utilities end up building too much infrastructure to serve data centers.
The 100,000 MW figure is roughly equivalent to doubling the generating capacity of all 94 U.S. nuclear reactors in six years.
While the impact is vast, the potential benefits from AI applications are fueling growth as a race for leadership among hyperscale internet and cloud infrastructure companies, led by Amazon Web Services, Microsoft, Google, Meta, Apple and TikTok, are fueling growth.
“I hesitate to use that word, but it’s almost an arms race,” he said.
“They’re saying, ‘We don’t know how much we need, but we know we need power,'” EPRI’s Mansour said. “‘We need to build as much power as we can.'”
The cost of new electricity
E3 used the healthcare industry as an example to illustrate the potential scale of artificial intelligence’s power demands.
The company estimates that there are about 32,000 radiologists in the U.S., with an average annual salary of $354,000. If AI could learn to accurately analyze X-rays, ultrasounds and MRI images and replace just 10% of radiologists, the replacement revenues could be more than $1 billion a year, according to E3 researchers.
This would help cover the costs of energy-hungry supercomputers.
The long time it takes to connect wind and solar power to the grid and the two political parties are at odds over federal transmission policy mean gas-fired power is the quickest source of new power for data centers, E3 said.
North Dakota officials, in discussions with big tech companies, see the potential for booming data centers as an opportunity to tap some of the natural gas produced in the region’s abundant oil fields. Josh Teigen, director of the North Dakota Department of Commerce, said state officials are exploring ways to develop more gas-fired generating capacity to power data centers and sell excess electricity to the grid.
Grid operators and utilities have postponed retirements of some fossil fuel plants, citing threats to electricity reliability. “These withdrawals clash with customers’ environmental goals and investments and jeopardize state and utility emissions reduction targets,” E3 analysts said.
For example, delaying the retirement of a typical 1,000MW coal-fired power plant by one year would release 3.8 million tonnes of CO2 into the atmosphere, E3 calculates, which would offset the carbon savings of a 3,000MW large-scale solar power installation.
What can you do?
As an immediate measure, Gardner said utilities need to make greater use of advanced transmission cables and “grid-hardening technologies” such as dynamic line-rating sensors that tell grid operators when more power can be moved without overheating the cables.
EPRI analysts suggest that data center operators could synchronize AI operations to coincide with the days and nights when renewable power is most plentiful, or large AI customers could help fund pilots of advanced carbon-free power sources that run 24/7.
At a meeting on July 30, the Secretary of Energy’s Advisory Committee urged utilities, regulators, and data center users and builders to work together to accelerate the development of wind, solar, and batteries. Long-term technologies include small modular reactors (SMRs), long-term grid batteries, hydrogen hubs, fuel cells, and underground storage of carbon emissions from fossil fuel power plants.
Andy Bockman, senior power grid strategist at Idaho National Laboratory, highlighted small reactors.
“Coal-fired and gas-fired power plants can be replaced one-to-one with similarly sized SMR facilities,” Boffman said. “To complement this, we need to expand wind, solar and storage on microgrids as quickly as possible.”
“As an industry, we’re going to have to consider self-generation until the grid catches up,” said Lawton Shanks of Aligned Data Centres.
EPRI’s Mansour said hyperscale operators have a lot of financial muscle and could partner with utilities to build pilot plants at data centers, potentially eliminating some of the financial risk of backing emerging technologies like hydrogen and carbon capture.
“We can have 10 SMRs and 10 carbon capture and storage projects operational by 2030,” Mansour said.
He also said 10 carbon capture and storage facilities could be installed at natural gas and coal-fired power plants.
Will big data center users invest? “They’re talking,” Mansour said.
