When a data center consuming several hundred megawatts of electricity drops off the grid without warning, the effect is roughly equivalent to an aluminum smelter vanishing in an instant. Generators that were matched to that load suddenly have nowhere to send their output. Frequency spikes. Voltage swings. Grid operators scramble to rebalance a system designed around the assumption that large customers do not simply disappear.
That scenario is no longer hypothetical. According to reporting that first surfaced in late May 2026, the North American Electric Reliability Corporation, the federally designated organization responsible for keeping the lights on across the United States, Canada, and parts of Mexico, has activated a Level 3 alert, its most severe classification, over unscheduled data center disconnections that caught grid operators off guard. The alert signals that NERC considers these events a direct, present threat to bulk power system reliability. However, as of early June 2026, NERC has not published a public bulletin or supporting data confirming the alert’s details, and the classification relies on accounts that have not been independently verified against primary NERC documentation.
At the same time, the Federal Energy Regulatory Commission has opened two formal proceedings that treat data centers, especially those running artificial intelligence workloads, as a distinct category of grid risk requiring new rules. Together, the reported operational warning and the documented regulatory response mark a turning point: the institutions that govern North America’s electric grid now regard data centers not as ordinary commercial customers but as a source of system-level instability.
What NERC’s alert system actually means
NERC operates a tiered alert framework that reliability coordinators and transmission operators use to flag emerging threats. Level 1 alerts are informational. Level 2 alerts request that grid entities take specific actions. A Level 3 alert, the ceiling of the system, is reserved for conditions that NERC believes pose an essential risk to the bulk power system and that demand immediate, coordinated response from operators across the continent.
Level 3 alerts are rare. They have historically been associated with severe weather events, fuel supply emergencies, and cyber threats. Issuing one over the behavior of a single category of electricity consumer, data centers, would be without recent precedent and would reflect the sheer scale of the loads involved. A single large AI training facility can draw 300 to 500 megawatts or more, roughly the output of a natural gas peaking plant. When that load vanishes in minutes rather than hours, the grid must absorb the surplus generation almost instantaneously.
Because no public NERC bulletin or supporting reliability data has been released as of June 2026, the specific threshold criteria that triggered the Level 3 designation in this case cannot be independently verified. Readers should treat the alert’s existence as reported but not yet confirmed by primary documentation.
Two FERC proceedings, one message
FERC’s regulatory response is documented in two active dockets. The first, Docket No. RM26-4-000, is a rulemaking titled “Interconnection of Large Loads to the Interstate Transmission System.” The docket’s notice of proposed rulemaking states that the proceeding “addresses the interconnection of large loads, including data centers, to the interstate transmission system” and targets the process by which large consumers connect to the high-voltage network, a process originally built for steel mills and chemical plants, not facilities whose demand can swing by hundreds of megawatts based on whether a training run is active.
The second proceeding addresses a related but distinct problem: data centers that co-locate directly at power plants, drawing electricity from a generator without routing it through the broader transmission system. FERC has ordered action on co-location arrangements specifically tied to AI data centers. The commission’s order states that co-location configurations raise concerns about whether remaining network customers bear reliability and cost consequences when generation capacity is diverted to serve a co-located load. If the data center trips offline, the generator’s full output can surge back onto the grid in ways operators did not anticipate, creating frequency and voltage disturbances that ripple outward.
Taken together, the two dockets send a clear message: existing interconnection standards were not designed for loads this large, this volatile, or this capable of appearing and disappearing on short notice. FERC is not issuing guidance or requesting voluntary cooperation. These are formal commission actions that will produce binding rules.
Why data centers behave differently from traditional large loads
Industrial facilities like refineries and smelters are large electricity consumers, but their demand profiles are relatively stable and predictable. A refinery does not double its power draw on Tuesday and halve it on Wednesday. Data centers, particularly those built for AI model training, operate on a fundamentally different pattern. A large training job can saturate thousands of GPUs for weeks, pushing power consumption to the facility’s maximum capacity. When the job completes, or when a hardware fault forces a restart, consumption can plummet.
The problem is compounded by the speed of the change. Traditional industrial loads ramp down over hours, giving grid operators time to adjust generation dispatch and transmission flows. A data center experiencing a cooling system failure or a network partition can shed load in seconds. From the grid’s perspective, the effect is similar to a large generator tripping offline, except the physics work in reverse: instead of losing supply, the system suddenly has too much.
Grid frequency in North America is maintained at 60 hertz. Deviations of even a fraction of a hertz trigger automatic protective responses, including generator trips and load shedding, that can cascade if not contained quickly. The larger the sudden imbalance between supply and demand, the harder it is to contain. A 400-megawatt data center dropping off the grid without notice creates the kind of imbalance that reliability engineers spend careers trying to prevent.
What is not yet public
Several critical details remain undisclosed. NERC has not published a bulletin identifying the specific regions affected by unscheduled data center disconnections, the number of incidents involved, or the magnitude of the demand swings that triggered the Level 3 designation. The threshold criteria that distinguish a Level 3 alert from lower tiers have not been released in connection with this event.
It is also unclear what caused the disconnections. Equipment failures inside the data centers, grid conditions that forced protective relays to trip, and deliberate operator decisions to shed load during periods of high electricity prices are all plausible explanations, and each carries different implications for how regulators should respond. A facility that trips offline because of an internal fault poses a different reliability challenge than one that strategically curtails consumption to avoid peak pricing, even though both produce the same sudden demand drop from the grid operator’s perspective.
No final compliance deadlines have been announced for either FERC proceeding. Whether interim measures, such as mandatory advance-notice requirements for large-load disconnections, will be imposed before final rules take effect has not been disclosed. Regional transmission organizations and balancing authorities will need time to translate any federal directives into updated tariffs, interconnection agreements, and local operating procedures.
The gap between the severity of the reported alert and the thinness of publicly available operational data is notable but not unusual. Transmission operators and reliability coordinators work from real-time telemetry that is not designed for public disclosure. Formal publication of supporting analysis typically follows months after the triggering events. The regulatory proceedings now underway at FERC suggest that when that data is eventually summarized, it will likely confirm what the alert already implies: large, fast-cycling data center loads have outgrown the rules the grid was built on.
What grid operators and data center developers should prepare for
For utilities negotiating interconnection agreements, the practical signal is concrete. FERC is building a framework that will change how large loads connect to and disconnect from the transmission system. Facilities that cannot demonstrate predictable load behavior, including the ability to provide reasonable advance notice before going offline, are likely to face stricter interconnection terms.
Companies planning new data center construction or expansion should review existing interconnection agreements now and prepare for conditions that may include real-time load reporting to the balancing authority, minimum notice periods for planned outages, and financial penalties for unscheduled disconnections. Developers who treat grid coordination as an afterthought risk delays, higher costs, and regulatory friction that could stall projects already deep in permitting.
How FERC’s new dockets reshape data center interconnection risk
For ordinary electricity customers, the stakes are less abstract than they might appear. Every unplanned large-load disconnection forces grid operators to activate reserves and, in extreme cases, curtail service to other users. As data center capacity continues to expand across North America, driven by demand for AI computing that shows no sign of slowing, the frequency and severity of these events will likely increase unless the rules catch up. FERC and NERC are now racing to close that gap before the next unscheduled disconnection tests a grid that was never designed to absorb it.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
