Data-center growth is triggering a new, escalating type of grid disturbance — but a broad rise in outages for ordinary local businesses remains a documented risk, not yet a measured trend

Summary

Short answer: not yet at scale for ordinary customers, but the underlying risk is real, worsening fast, and now serious enough that grid regulators have escalated to mandatory action — while broad-based business-interruption impact on non-data-center businesses remains a documented risk rather than a measured trend. National outage-duration statistics (EIA SAIDI) show no clean upward trend attributable to data centers: year-to-year swings since 2013 track major storm years (e.g., 2020's record ~483 minutes/customer was a bad-hurricane year; 2022 was back down to 336 minutes), and when major events are excluded, average interruption duration has held essentially flat around two hours per customer per year nationwide since EIA started tracking in 2013. But in the specific regions where data-center growth is heaviest — Dominion Energy's Northern Virginia zone within PJM, and ERCOT in Texas — grid operators have documented a new, escalating, and mechanically distinct failure mode: large numbers of data centers simultaneously disconnecting from the grid during routine transmission faults because their internal voltage/frequency protection is oversensitive. Three such events have hit PJM's Dominion zone: ~1,500 MW in July 2024, ~1,500 MW in February 2025, and a record ~3,800 MW event on July 22, 2026 — the largest in PJM's history, per PJM's own August 2026 presentation. None has yet caused a sustained blackout for ordinary customers (grid operators recovered control within minutes each time), but the trend is escalating fast enough that NERC issued a rare, highest-severity Level 3 alert in May 2026 mandating seven specific actions from every transmission planner, operator, and balancing authority in North America. NERC's own January 2026 assessment now rates 13 of 23 North American grid regions — including PJM, ERCOT, and MISO — at elevated or high risk of energy shortfalls within five years, driven mostly by data-center demand growth. On business-interruption exposure: insurers treat power supply as the single largest driver of BI losses at data centers themselves (45% of data-center outages, per the Uptime Institute, cited in Swiss Re's 2026 sigma report), and Swiss Re documents insurers already pricing regional "accumulation risk" — the danger that one grid event could trigger correlated claims across many co-located data centers and their programs at once. What is not yet documented anywhere found in this research is a quantified rise in BI claims, premiums, or actual outage exposure for ordinary local businesses (not data centers) specifically because of nearby data-center growth; the connecting evidence is rising electricity costs and explicit rolling-blackout warnings from utility and grid officials in these regions, not a measured change in those businesses' own outage statistics.

What the AskAmerica corpus does and doesn't cover

askengine's catalog was searched directly (search_catalog, find_recipe) for outage, reliability, SAIDI/SAIFI, and business-interruption measures. Confirmed gap: no table in this corpus carries power-outage frequency, duration, or reliability-index data — the term "outage" matched zero columns or tables across all ~26 schemas. The closest tables are energy.eia_power_plants, energy.eia_electricity_generation, energy.eia_electricity_prices, and energy.eia_utility_annual, all of which cover generation, capacity, and retail prices, not interruption events. This is a genuine sourcing gap rather than a query-shape problem, so the entire outage/reliability/insurance analysis below is built from primary and near-primary external sources — EIA's own published reliability releases, NERC's own assessment, alert, and incident-review documents, PJM's own August 2026 event presentation (via Utility Dive's direct reporting), and Swiss Re Institute's own sigma research (via Risk & Insurance) — fetched directly rather than from this connector's warehouse.

1. Has outage frequency or duration actually gone up?

Nationally: no clean trend tied to data centers. EIA has collected SAIDI (System Average Interruption Duration Index, hours of non-momentary interruption per customer per year) and SAIFI (frequency of interruptions) from utilities since 2013. U.S. customers experienced the longest average interruption duration on record in 2020 (~483 minutes, roughly eight hours, including major events). By 2022 the national average had fallen to 336 minutes (5.6 hours), a decline EIA attributes directly to fewer major weather disasters that year versus 2021 (18 billion-dollar weather disasters in 2022, including Hurricane Ian, Hurricane Nicole, and Winter Storm Elliott). Critically, EIA's own analysis states that "since 2013, the average duration of electricity interruptions each year has remained consistently around two hours after excluding major events" — i.e., the swings in the headline national number are a weather-events story, not a baseline-grid-degradation story, and EIA's releases do not attribute the trend to data-center load growth. States with the longest interruptions in the most recent data (Florida, West Virginia, Maine, Vermont, New Hampshire) are hurricane- and forest-exposed states, not the data-center-heavy states (Virginia, Texas).

In the specific regions with the heaviest data-center concentration, a new and escalating disturbance pattern has emerged that the standard SAIDI/SAIFI index doesn't directly capture, because in every documented case so far it did not reach retail customers as a sustained outage. PJM's own presentation to its Operating Committee (Aug. 6, 2026) documents three successive events in Dominion Energy's Northern Virginia zone — home to the world's largest data-center cluster: roughly 1,500 MW of data-center load tripped off in 2024, another ~1,500 MW in 2025, and a record 3,800 MW on July 22, 2026, the largest such event in PJM's history, triggered by a normally-cleared fault on a single 230-kV line. PJM's Operating Committee chair stated bluntly that "these data centers are too sensitive to the kind of voltage that they like, and we feel they are disconnecting too early." PJM recovered system balance within nine minutes (versus NERC's 30-minute standard) each time, and no sustained customer blackout resulted, though the July 2024 event caused roughly 10 minutes of voltage flicker felt from Chicago to Boston to Miami. A parallel pattern has occurred in ERCOT, where cryptocurrency-mining and other large flexible loads caused 25 separate load-loss events (100-400 MW each) since November 2023.

Regulators have now escalated this from a monitored curiosity to a mandatory compliance issue. After a Level 2 warning in 2025 drew what NERC called insufficient responses ("entities generally did not have sufficient processes, procedures, or methods to address emerging computational loads"), NERC issued a rare Level 3 "Essential Actions" alert — its highest severity level, previously used only for inverter-based-resource risks — in May 2026, requiring transmission planners, operators, and balancing authorities across North America to collect detailed modeling data from computational loads, revise reliability-study triggers, and install dynamic fault-recording devices, with compliance responses due Aug. 3, 2026. Separately, NERC's January 2026 Long-Term Reliability Assessment rates 13 of 23 North American assessment areas — including PJM, ERCOT, MISO, and SERC-Central — at elevated or high resource-adequacy risk over the next five years, citing data-center-driven demand as the largest single contributor to a projected 224 GW rise in summer peak demand (a rate NERC says is the highest it has tracked since 1995). Virginia's own Joint Legislative Audit and Review Commission (JLARC) reached the same regional conclusion independently, and Central Virginia Electric Cooperative's CEO told Cardinal News he personally expects rolling blackouts in the PJM footprint within three to five years. Net effect: the frequency of grid-stress events mechanically tied to data centers has clearly and measurably increased and is still escalating (1,500 MW → 1,500 MW → 3,800 MW across three successive years), but this has not yet translated into a measured rise in sustained outage frequency or duration for ordinary customers — it is a forward risk indicator that grid regulators are now treating as urgent, not a realized outage-statistics trend.

2. Does that translate into more business-interruption exposure for other local businesses?

For data centers themselves, yes, clearly and with quantified figures. Swiss Re Institute's 2026 sigma report (via Risk & Insurance) states that power supply is the largest single driver of business-interruption risk at data centers, responsible for 45% of data-center outages per the Uptime Institute, and that global insurance premiums tied to data centers are projected to more than double from $10.6 billion to $24.2 billion by 2030. Swiss Re explicitly flags "accumulation risk": developers cluster multiple data centers within roughly 20-mile radiuses (e.g., Abilene, TX), meaning a single regional grid or catastrophe event can trigger correlated claims across many insured facilities and separate insurance programs (building, equipment, and power-plant coverage are often placed separately, making total exposure hard to track) simultaneously. Swiss Re's engineering lead is quoted saying the industry must move "beyond passive risk transfer to actively enabling resilience" given how concentrated, energy-dependent, and systemically critical these facilities have become.

For ordinary local businesses that are not themselves data centers, no source found in this research quantifies a rise in BI claims, premiums, or realized outages specifically attributable to nearby data-center load growth. The documented connecting mechanisms are: (a) rising electricity costs that raise operating costs for every business in the region — Virginia's JLARC study found that data centers currently cover their own cost of service, but that future demand growth (projected to require Virginia to use 15% more energy even before new data-center demand, doubling within a decade) is likely to raise costs for other customers, and PJM capacity prices have risen sharply amid data-center-driven demand growth; (b) explicit, named rolling-blackout warnings from utility and grid officials serving these same regions — Central Virginia Electric Cooperative's CEO stated plainly "I believe we will have rotating blackouts before this gets resolved" within three to five years in PJM, a warning that applies to every customer on that grid, not just data centers; and (c) regulatory and legislative responses treating the spillover as a real, expected risk — JLARC recommended legislation clarifying that Virginia utilities may delay (though not deny) new large-customer service specifically to protect existing customers' reliability, and Georgia adopted a 2025 rule requiring large-load customers to co-fund grid upgrades, both of which are policy actions premised on an expected, not-yet-fully-realized, spillover risk to other ratepayers and businesses.

Bottom line and what would resolve this more precisely

These are two separate, honestly-labeled findings, not one: (1) a new, mechanically well-documented, and rapidly escalating grid-disturbance pattern tied directly to data-center concentration in PJM/Dominion (Northern Virginia) and ERCOT (Texas) — three successive Virginia events growing from 1,500 MW to a record 3,800 MW in three years, serious enough that NERC escalated to its highest-severity Level 3 mandatory alert in May 2026, with 13 of 23 North American grid regions now rated at elevated or high resource-adequacy risk; and (2) no published measurement yet that ordinary local businesses in those regions are experiencing more or longer outages, or higher business-interruption losses, specifically because of that data-center growth. What exists instead is a chain of documented, named risk: rising electricity costs, explicit rolling-blackout warnings from the same utility executives serving those businesses, and regulatory bodies (JLARC, Georgia PSC) treating cost/reliability spillover as expected enough to legislate against. A rigorous causal test would require utility-territory-level SAIDI/SAIFI data (EIA Form 861 detail, not yet in this corpus, would need to be pulled at the individual-utility rather than national level) compared before and after major data-center interconnection in specific counties like Loudoun, VA, against a comparable non-data-center-growth region — that comparison was outside the scope of what could be completed here but is the next step that would convert this from a documented risk into a measured trend.