North America’s electric grid watchdog is acting with urgency to guard against widespread power outages stemming from the volatile energy demand of data centers.
The North American Electric Reliability Corp., the grid’s not-for-profit security monitor, is moving to draft new standards for large artificial intelligence computing hubs that could lead to regulation of their operations. A NERC committee is expected to initiate the project Wednesday, seeking final approval by the end of the year.
A recent NERC white paper says the extreme power fluctuations during the training of AI models, when power demand can swing by hundreds of megawatts in an instant, is a “high likelihood, high impact” risk. It’s a threat that could potentially throw carefully calibrated power grids out of balance and lead to uncontrolled, cascading outages.
The report lists “critical reliability gaps” caused by large data centers, calling it “imperative” to address this and other risks.
The step toward regulating computing hubs that are increasing in size and number marks an important new challenge for the AI industry. Google, Amazon, Meta, Microsoft, Anthropic and others are developing gigawatt-scale data center complexes — requiring as much power as a midsize or large American city.
“The emergence of gigawatt-scale loads may create exponential complications for study, planning and operations,” the paper found.
AI’s exponential growth could push total AI peak power requirements to 50 gigawatts by 2030, according to an analysis by the Electric Power Research Institute. Concern that existing power demand and the projections are driving up utility bills for households and other businesses has shaped debate about the AI buildout. Risks to a U.S. power grid already under stress from extreme weather are emerging for the broad array of regulators and standard-setting groups charged with ensuring the lights stay on.
The Edison Electric Institute, representing investor-owned utilities, urged NERC to require data centers to meet the type of engineering standards and protocols governing generators and other major operators on the grid.
“They should be subject to similar regulatory requirements,” EEI said in a statement.
“As very large loads like data centers expand, planners and grid operators need clear visibility into how those loads will connect to and operate on the system,” said Todd Snitchler, CEO of the Electric Power Supply Association.
It happened in Virginia
NERC highlighted a 2024 incident when a transmission outage in Virginia created a sharp voltage spike. Controls for dozens of data centers automatically separated from the grid and switched to backup power to protect sensitive computer chips and their cooling systems.
Grid operators suddenly faced an unexpected, roughly 1,500-megawatt loss in power demand.
The larger grid was able to keep operating. But NERC said the same incident involving a large cluster of supersize data centers would create an unacceptable risk of cascading shutdowns and widespread outages.
The organization issued an alert on these issues last September.
Some of the concerns identified by NERC are well known, including the lack of information from data centers and utilities on how the computing hubs are engineered and operated. That limits grid operators’ understanding of potential risks. The issue is complicated by the speed of data center construction, which often outpaces needed grid upgrades and preparations for the centers’ power requirements.
NERC also cited “a lack of collaboration and coordination” between data center operators, utilities and grid operators. For example, said the NERC working group, some data centers aren’t sharing adequate information about shifts to backup power, which can impact grid operations.
NERC is moving relatively fast to get new standards written before the end of the year. They would then have to be approved by the Federal Energy Regulatory Commission.
Its accelerated timetable “underscores the urgency that the federal regulator and broader industry are treating the reliability risks posed by large load growth,” said Jiecheng “Jeff” Zhao, chief engineer of Elevate Energy Consulting in a recent blog reporting on a NERC technical conference last month. “It will likely lead to mandatory requirements” for data centers, Zhao wrote.
A “critical” risk, NERC says, is the rapid power fluctuations of tens of thousands of computer chips as AI models train to understand and process huge amounts of information. As one training phase ends and the results are compiled, the power demands of a large data center may plunge by 300 megawatts in moments and then ramp back up equally suddenly as training resumes.
NERC noted that data center operators typically do not tell grid operators these sudden shifts are coming or share technical data that grid control room engineers require to safeguard their networks.
Extreme power fluctuations spreading to the grid can activate protective automatic controls that close down power equipment or transmission lines. The result can be a chain reaction of equipment shutdowns, “potentially leading to uncontrolled separation and cascading outages,” the NERC report said.
When a data center ramps down, to the grid it looks like a big power customer just disappeared, said Julia Matevosyan, associate director and chief engineer of the Energy Systems Integration Group (ESIG), a nonprofit research association of grid experts. ESIG recently issued a report documenting data center risks to the grid that agrees with the FERC task force analysis.
“If it’s just one data center, it’s fine. But if you have multiples of them in close proximity, and they all get hit — they all ‘disappear’ — that massively exacerbates the initial loss,” Matevosyan said in an interview.
Ramp up, ramp down
In addition, disturbances could trigger oscillations that spread to nearby power plants on the grid, damaging and — in extreme cases — destroying drive shafts of electric turbines near the data center, Matevosyan said. “There’s no good understanding of what the limit is” when such oscillations hit, she said.
The dangers of sudden grid fluctuations were highlighted in a blog in October by two senior managers at NVIDIA, the dominant AI chip manufacturer. Joint research signed by 50 scientists from NVIDIA, Microsoft, and OpenAI documents the risk, said NVIDIA’s Neeraj Srivastava and Harry Petty.
“These volatile swings — representing hundreds of megawatts ramping up and down in seconds — pose a significant threat to the stability of the utility grid, making grid interconnection a primary bottleneck for AI scaling,” they wrote.
“We’re at a critical inflection point, where the industry can no longer rely on incremental improvements, and a fundamental architectural shift is required … capable of managing the power demands of modern AI,” they added.
The Data Center Coalition, representing the largest AI technology companies and data center developers, has declined to say whether it supported or opposed mandatory reliability standards.
Aaron Tinjum, vice president for energy at the coalition, said in a statement that his organization “is reviewing the recommendations from NERC’s Large Load Working Group, and we will continue to engage with NERC, regional entities, grid operators, and utilities across the country to provide education on the data center industry and ensure solutions support both grid reliability and 21st-century economic growth.”
Tinjum added that grid stability incidents linked to data center operations in Virginia and Texas in recent years have led to increased collaboration and technical information sharing between power companies and data center operators that have helped prevent new incidents.
“It is critical to ensure that the proposed requirements by NERC are based on complete evidence and the technical capabilities and specifications of data center equipment,” Tinjum said.
Dion Harris, NVIDIA’s senior director of high performance computing and AI infrastructure, told E&E News last month that his company, working with Microsoft OpenAI, has found solutions to power fluctuation issues.
New features in NVIDIA’s state-of-the-art Blackwell chip systems smooth out sharp swings in power demand from AI routines, reducing risks to grid stability, he said.
“A number of different solutions are in play,” Harris said. “The key takeaway is that you’ve got enough to make sure there isn’t a stability issue.”
The NERC policy paper said that while voluntary cooperation on grid security is an option, “directly enforceable NERC standards may be needed instead.”
