The North American Electric Reliability Corporation (NERC) issued its most urgent grid reliability warning in response to a new threat: artificial intelligence data centres causing massive, near-instantaneous power fluctuations that leave grid operators with virtually no time to respond.
NERC’s Level 3 Essential Action Alert, released with a 3 August deadline for utility responses, marks a turning point in how the power industry addresses the explosive growth of AI infrastructure.
The alert outlined seven mandatory actions for registered entities to implement immediately, citing “customer-initiated large load reductions and significant oscillations that occur in seconds, leaving little or no room for real-time responses, threatening bulk power system (BPS) reliability.”
The warning comes as individual AI campuses begin consuming electricity on the scale of small cities, while their computational workloads create power demand that can swing from zero to 100% in milliseconds.
Try Premium for just $1
- Full premium access for the first month at only $1
- Converts to an annual rate after 30 days unless cancelled
- Cancel anytime during the trial period
Premium Benefits
- Expert industry analysis and interviews
- Digital access to PV Tech Power journal
- Exclusive event discounts
Or get the full Premium subscription right away
Or continue reading this article for free
The scale of the problem
“We’re not talking about small fluctuations anymore—we’ve seen events where 1,000MW or more dropped off the grid in a matter of seconds,” says Thomas Sisto, co-founder and CEO of flow battery company XL Batteries. “Now think about where this is headed: individual AI campuses drawing several gigawatts, which is the power demand of a small city. When a load that size ramps up or down within milliseconds, the grid simply wasn’t built to absorb that.”
The speed of these fluctuations represents a departure from traditional industrial loads. Where conventional power consumers followed predictable patterns that utilities could forecast using decades of historical data, AI data centres present an entirely different challenge.
“Data centre loads or AI-based data centre loads are very highly dynamic,” explains Dr. Hooman Ghaffarzadeh, director of power systems engineering at Wärtsilä Energy Storage. “They can go from zero to 100% in just a few milliseconds, and this is not just a one-time event. These changes can happen every 10 milliseconds, which is very aggressive.”
The physical reality involves hundreds of thousands of graphics processing units (GPUs) ramping up or down nearly simultaneously as AI training workloads shift. “That can create a very large change in electricity demand very quickly, and the equipment between the data centre and the grid has to absorb that change,” Sisto explains.
The modeling challenge
The deadline required utilities and grid operators to report on their capabilities to characterise and manage these unprecedented load behaviours—a task complicated by the lack of reliable data about how AI workloads actually behave.
Dr. Ghaffarzadeh’s team at Wärtsilä has been in contact with several hyperscale data centre developers over the past 12 to 18 months. “It was very common for developers to not have any forecasted load profile for data centres, and some of them might have some emulated load profile, but not the actual load profile that can be used for studies,” he notes.
Traditional loads were “highly predictable,” Dr. Ghaffarzadeh explains. “We could use historical data for the past 20 years to come up with forecast data for the next 24 hours or week or even next couple of months. But for data centre applications, this is totally different.”
Traditional grid monitoring happens at the substation or transmission level, often with seconds or minutes of lag—far too slow to catch millisecond-scale disturbances. “Utilities today may be able to see broader system conditions, but they often lack a real-time picture of voltage fluctuations and other power quality disturbances,” says Matthew Williams, Founder and CEO of IONATE, which manufactures intelligent transformer technology.
Regulatory changes ahead
Industry experts anticipate that NERC’s initial steps will evolve into formal requirements similar to those developed for renewable energy interconnections over the past decade.
“I believe the immediate change will be more standardised or mandatory packages for dynamic studies, and this process will start from a very early stage, from design of the data centres all the way to commissioning,” Dr. Ghaffarzadeh predicts. He expects the US will adopt approaches similar to Australia’s comprehensive compliance framework.
Sisto sees the Level 3 Alert as a clear signal. “A Level 3 alert is NERC signalling real events and challenges happening right now, and it’s going to intensify as more gigawatt-scale campuses come online,” he says. “Historically, that kind of signal eventually turns into interconnection requirements and ride-through standards, the same way we saw happen with large-scale renewables over the last decade.”
Addressing millisecond-scale grid disturbances requires technology that can respond at comparable speeds.
“Control speed and precision matters because today’s power systems can cascade into failure very fast,” Williams explains. He claims that IONATE’s Hybrid Intelligent Transformer is designed to detect, analyse, and respond to power disturbances in less than a millisecond. “If an event occurs, the system can instantaneously rebalance conditions and help bridge the gap before longer-duration solutions, such as BESS or on-site generation, come online.”
While fast-responding power electronics can handle sub-second fluctuations, experts emphasise that BESS plays an equally critical role in managing longer disturbances and preventing sudden disconnections.
“There isn’t one piece of equipment that fixes it,” Sisto says. “Fast-responding power electronics and controls can manage the shortest fluctuations, while batteries can provide a buffer and keep the facility stable through longer disturbances.”
Dr. Ghaffarzadeh’s research at Wärtsilä has demonstrated that BESS serve multiple functions beyond traditional backup power. “We show that using battery systems, we can reliably help to put voltage and frequency within the stable range or the acceptable thresholds, and we can reliably also reduce the torsional stresses on the engines,” he explains.
“A battery that’s designed primarily to sit idle until an outage has a very different job than one that’s being used every day to manage load, smooth demand and support the facility through grid events,” Sisto notes. “As these campuses scale, storage increasingly needs to be designed around that second use case.”
Williams frames the need for multiple technologies in terms of response timeframes. “Real-time transformer intelligence addresses the fastest disturbances occurring in milliseconds that affect power quality and voltage stability. Storage is extremely valuable for balancing supply and demand over seconds, minutes, and hours. Together they form part of a layered approach to reliability.”
Disconnection
One contentious question emerging from NERC’s alert is whether data centres should be allowed to disconnect from the grid during disturbances or required to provide grid support.
Sisto advocates for a middle path. “What I want to see is disconnecting become the last resort instead of the first response,” he says. “Storage offers a third path. It acts as a shock absorber between the data centre and the grid—instead of a gigawatt swing hitting the transmission system directly, the storage system takes the impact and the facility rides through it.”
Williams emphasises that the trend is toward greater participation from large energy users. “Greater visibility into and capability to manage voltage, power quality, and power factor can help large energy users protect their own operations while also ensuring they aren’t introducing issues back onto the grid—going from simple consumers to better ‘grid citizens.'”
Next steps
Despite growing awareness, experts suggest the industry still has significant ground to cover.
“It’s severe, and it’s getting more severe every quarter as these campuses scale,” Sisto says. “The industry is being more proactive than it was a year ago, but there’s still a gap between awareness and action.”
The timeline mismatch between infrastructure development and demand growth adds urgency. “Transmission upgrades take years to plan, permit, and build—sometimes closer to a decade,” Sisto notes. “Data centre demand isn’t going to pause for that timeline. Whatever standards eventually land, we need solutions that work now, and storage is one of the few levers that can be deployed on a timeline that actually matches the pace of demand growth.”
Experts recommend that utilities and data centre operators begin implementing solutions immediately rather than waiting for mandates. “Whatever NERC ultimately requires, the operators who build this in now will already be ahead of it, and ahead of competitors still treating this as tomorrow’s problem,” Sisto says.
Battery Asset Management Summit USA 2026 will be held 15-16 September in Garden Grove, California, hosted by Energy-Storage.news publisher Solar Media (part of the Informa Group). The agenda emphasises addressing the roles of AI, cybersecurity, and second-life applications, broken down into two tracks: Technical Asset Management and Commercial Asset Management. This year, the conference is also co-located with Solar & Storage Finance Summit USA. Visit the official site for more details.
