A single AI data center shutting down without warning could yank 1,000 megawatts off the power grid in an instant, enough electricity to serve roughly 750,000 homes. The North American Electric Reliability Corporation, the organization responsible for keeping the lights on across the continent, has flagged that scenario at its highest level of concern, warning that the resulting voltage and frequency swings could cascade through transmission lines and trigger blackouts in communities miles from the facility itself.
The warning lands at a moment when the nation’s power infrastructure is already straining under a surge of data center construction driven by companies like Microsoft, Amazon, and Meta, all racing to build the computing muscle behind artificial intelligence. Federal regulators at the Federal Energy Regulatory Commission are now working against a self-imposed deadline to rewrite the rules governing how these massive facilities connect to the grid, with final action expected by mid-2026.
Why regulators are treating this as a structural threat
Two formal actions by FERC frame the federal response. First, the Commission’s Chairman released a statement launching a review of risks tied to co-locating large electric loads, including AI data centers, at existing power plants. The review targets both grid reliability and the question of who pays when a single customer consumes electricity at a scale once reserved for entire towns.
Second, the Commission announced it will act on a large-load interconnection docket by June 2026. That deadline carries weight because interconnection queues, the formal process through which new projects get permission to plug into the grid, are already badly backlogged. PJM Interconnection, the grid operator covering 13 states from Virginia to Illinois, reported in its 2024 queue data that pending requests had ballooned to record levels, with data centers accounting for a growing share.
The two actions are linked by a specific engineering problem. Co-location allows a data center to draw power directly from an adjacent generator before that electricity ever reaches the broader transmission network. For the operator, this can slash costs by avoiding charges for shared grid infrastructure. For everyone else, it creates a hidden vulnerability.
If the data center suddenly disconnects, the generator’s full output floods back onto transmission lines that were never sized to handle the surge. If the generator trips offline instead, the data center’s enormous demand instantly shifts to the surrounding grid, forcing neighboring utilities to fill the gap in milliseconds. Either failure mode can send voltage and frequency disturbances rippling across a wide area, potentially triggering protective relays that knock additional equipment offline in a chain reaction.
The scale of what’s coming
To understand why regulators are alarmed, consider the numbers. U.S. data center power demand stood at roughly 17 gigawatts in 2022, according to estimates from the International Energy Agency. Industry projections from McKinsey and Goldman Sachs suggest that figure could more than double by 2030, driven almost entirely by AI workloads that require far more electricity per server rack than traditional cloud computing.
A single hyperscale AI campus can consume 300 to 1,000 megawatts at full build-out. For comparison, a large nuclear reactor produces about 1,000 megawatts, and a mid-sized American city of 200,000 people might use 400 to 600 megawatts on a hot summer afternoon. These facilities are not growing incrementally the way steel mills or aluminum smelters did over decades. Some can reach full power consumption within two to three years of breaking ground.
That speed is the core of the problem. Traditional grid planning assumes load growth is gradual enough for utilities to build transmission capacity in step with demand. AI data centers upend that assumption, and their developers often seek to accelerate the process further by co-locating with existing generators rather than waiting years in the standard interconnection queue.
The geographic concentration makes the risk sharper. Northern Virginia’s “Data Center Alley” in Loudoun County already hosts the largest cluster of data centers on Earth, and Dominion Energy has publicly discussed the strain on local transmission infrastructure. Central Texas, where ERCOT manages an isolated grid with limited ability to import power from neighboring regions, has seen a wave of new data center proposals. The greater Phoenix area and central Ohio are also emerging hotspots.
What has not been confirmed
Transparency matters here, and several important details remain unresolved as of June 2026.
The specific technical criteria behind NERC’s highest-level designation have not been published in full. It is not yet clear whether the alert applies uniformly across all regions or targets corridors where data center construction is most concentrated. NERC’s reliability assessments, including its 2024 Long-Term Reliability Assessment, have flagged rapid load growth from data centers as a top-tier risk, but the detailed modeling behind the 1,000-megawatt scenario threshold has not been released publicly.
No grid operator has publicly confirmed recording a sudden load drop of that magnitude from a single data center. Grid operators routinely plan for the loss of their single largest generator, a contingency known as “N-1” in reliability planning. But the scenario NERC describes involves a massive load vanishing rather than a supply source tripping, and the automatic systems designed to rebalance the grid may not respond identically in both cases. Until operators release event reports or disturbance analyses, the 1,000-megawatt figure represents a modeled risk, not a documented incident.
FERC has also not disclosed what form its final interconnection rule will take. The options range widely. Mandatory “ride-through” requirements would force data centers to stay connected during grid disturbances and absorb power fluctuations rather than disconnecting to protect their own equipment. Stricter study requirements, on the other hand, would mainly slow the approval of new co-location arrangements without changing the behavior of facilities already operating. The engineering and financial implications of those two paths are very different, and the Commission has not signaled which direction it is leaning.
No specific co-location agreements or cost-allocation disputes have been made public as part of the current review. The financial incentives driving co-location are strong: a data center that buys power directly from an adjacent generator can avoid paying its share of transmission infrastructure, effectively shifting those costs onto residential and small-business ratepayers. But without disclosed contracts, the scale of that cost shift remains an informed estimate rather than a documented figure.
What this means for the people paying the electric bill
For the tens of millions of Americans whose electricity travels through the same transmission corridors now being claimed by hyperscale computing campuses, two things are now concrete.
First, the regulatory clock is running. FERC has committed to issuing rules by mid-2026 that will govern how very large electric loads connect to the grid, how their risks are managed, and how costs are divided. Utilities, grid operators, and data center developers are already adjusting their planning in anticipation. Projects are being delayed or redesigned to avoid running afoul of rules still being drafted.
Second, the worst-case scenario that NERC has modeled, a single AI campus dropping roughly a gigawatt of demand in an instant and sending blackout-triggering disturbances across the surrounding grid, remains a possibility rather than a recorded failure. Whether it stays in the realm of remote contingency or becomes a lived experience depends heavily on what FERC produces in the coming months and whether grid operators update their protective systems to account for a type of disruption the power system was never designed to handle.
The stakes are not abstract. Every new hyperscale facility that connects to the grid without updated safeguards narrows the margin between a stable power system and one that is vulnerable to a single point of failure. The rules written in 2026 will determine how thin that margin is allowed to get.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
