Back
From AI to Facebook to Google Maps, the nation’s demand for computing power is growing, with households in the U.S. now averaging a whopping 21 devices — think smartphones, TVs, and thermostats — all connected to the internet.
That was one of many statistics lobbed at North Carolina utility regulators last week as they gathered to grapple with the coming onslaught of data centers, the immense buildings filled with hardware that make our around-the-clock connectivity possible but could strain the state’s electric grid, raise utility bills, and increase pollution.
Over the course of a two-day discussion on how to avoid these downsides, one simple solution came up again and again: Data centers could commit to limiting their electricity consumption slightly for a handful of periods during the year, formalizing the practice of modulating energy use that’s already standard across the industry.
“One of the issues that the commission is particularly interested in is load flexibility,” Karen Kemerait, the commissioner presiding over the technical conference, said to more than one presenter last week, before pressing them on the concept.
In response to Kemerait, experts from Google and other tech giants, along with North Carolina’s predominant utility, Duke Energy, all voiced degrees of support for the notion.
Yet how and whether regulators move to actualize load flexibility remains unclear. The Utilities Commission isn’t required to take action following its Oct. 14 and 15 meeting. And unlike other reforms repeatedly mentioned, such as a special tariff for data centers, the policy doesn’t easily translate to a rate case or other dockets before the panel.
That’s part of why Tyler Norris, a former solar developer and a thought leader on load flexibility who presented last week, hopes it will become a choice for data centers if nothing else.
“At minimum, why not have a voluntary service option that enables a large load to connect faster in exchange for bounded flexibility?” Norris told Canary Media. “In every conversation I’ve been in, I’ve heard no objection to the idea. Obviously, it’s at the discretion of the commission — whether they want to encourage it.”
Data centers aren’t the only new large customers driving ever-growing electricity demand forecasts in North Carolina, which Duke used to justify a massive new fleet of gas plants in its most recent proposed long-term plan. But the centers are the most voracious consumers by far, accounting for over 85% of the energy demand in the economic development pipeline, the utility said last week.
Not all of these facilities in the pipeline will come to fruition: It’s not uncommon for tech companies to request grid connections in multiple locations before deciding where they’ll actually build. But many will materialize, posing thorny issues for the utility and its regulators.
What if Duke can’t build generation quickly enough to serve the energy-hungry centers? Can the company do so while still zeroing out its carbon pollution, as required by state law? How can regulators assure that tech giants, not residential customers, pay for new power plants and associated upgrades to the grid?
Load flexibility could provide an elegant answer to these vexing questions.
The idea is rooted in a counterintuitive reality: Data centers don’t run at maximum tilt 100% of the time — they routinely adjust processing power even as we can post videos to Instagram or EMS responders can transmit lifesaving patient data in the middle of the night.
That’s true for a number of reasons, Norris wrote on his Power and Policy site, including the fact that computer chips could overheat if stretched to their maximum theoretical processing speeds 24/7 and also that data centers plan for redundancy.
“Many facilities are overbuilt to ensure uptime, with servers periodically taken offline for routine maintenance, software upgrades, or hardware replacements,” Norris explained in the August post.
Information on data centers’ exact electricity use is scant, and it appears to vary based on type, but research suggests the facilities’ peak consumption is about 80% of what they could pull from the grid.
Yet utility planners typically assume otherwise, categorizing data centers as “firm loads” that need “firm capacity,” such as an on-demand power plant with an ample supply of fuel, plus an extra reserve margin — in Duke’s case, 22% — in the event of emergency.
In the simplest terms, while Duke might build 122 megawatts of generation to serve a data center that can draw a maximum of 100 megawatts of electricity, the center may never use more than 80 megawatts.
But if prospective data centers were transparent about their electricity-utilization plans and committed to them on paper, utilities could adjust how they anticipate new power capacity — averting the construction of massive amounts of fossil-fuel infrastructure as well as expensive grid improvements.
In September, analytics groups GridLab and Telos Energy published a report finding that Nevada’s biggest utility could delay the need for hundreds of megawatts of new power plants if data centers committed to modest flexibility terms that allow “uptimes” of 99.5%.
Similarly, Norris, a Ph.D. student at Duke University — which has no connection to the utility — is the lead author of a February paper showing that if data centers shaved just 0.5% off their use over the course of the year, 4.1 gigawatts of power capacity in Duke’s territory in the Carolinas could be avoided.
The figure “isn’t everything in terms of their load forecast,” Norris told regulators last week, “but it is arguably a meaningful share.”
While enlisting data centers to curtail their own energy use is still more theory than practice, that’s slowly starting to change. Pacific Gas and Electric in California, for instance, has piloted flexible service agreements that could get data centers online more quickly.
In August, Google announced voluntary flexibility agreements with Indiana Michigan Power and the Tennessee Valley Authority. The following month, the tech giant revealed a similar arrangement with Entergy in Arkansas.
The company vaunted those agreements, along with its plans to self-generate carbon-free electricity, at last week’s meeting. “Google is leaning in,” Rachel Wilson, a representative for the company, told commissioners.
The Data Center Coalition is an alliance of Google, Microsoft, Meta, Amazon, and dozens of other companies that own, operate, or lease data-center capacity. The coalition listed “voluntary demand response and load flexibility” as a recommendation to regulators last week, so long as data centers could get something in return — such as a quicker connection to the grid.
“There has to be some reciprocal value for data centers,” said Lucas Fykes, director of energy policy for the coalition.
Still, the AI race, a lack of transparency about data-center electricity use, and a genuine inability of anyone in this space to predict the future could complicate efforts around load flexibility.
“Not even the most sophisticated data center owner-operators” know what their load will look like in “a rapidly shifting competitive landscape,” Norris wrote in August. “Amid such uncertainty, their preference is generally to maintain maximal optionality.”
Indeed, though Duke expressed openness to load flexibility last week, the company advised caution for the long term.
“Looking at these [load-flexibility agreements] as temporary is important,” said Mike Quinto, the company’s director of planning analytics. “A well-designed voluntary program, that’s great. It’s not something we think should be mandated on a long-term basis.”
And while utilities are well-practiced in demand response for large industrial customers, Public Staff, the state-sanctioned customer advocate, voiced worry last week about scaling the same concept to data centers.
“We haven’t seen these magnitudes trying to interconnect and … potentially drop off the system,” said Dustin Metz, director of the agency’s energy division. “From an academic standpoint, if we can shave off some of those peaks, then that could potentially reduce some of the generation assets that we need to build out,” he said. But enforcement would be essential, and North Carolina is still new to data-center growth. “We’re a little bit of a living lab,” he said.
.svg)
