AI data centers have not caused a U.S. blackout yet, but on July 22, 2026 about 3.8 gigawatts of their load dropped off the Northern Virginia grid in seconds, the largest such event in PJM's history, according to Utility Dive. Here is what the reliability record, the regulators and the bills actually show, in plain English.
The Short Version
No U.S. blackout has been attributed to AI data centers, but the near-misses are growing. In July 2026 roughly 3.8 GW of data-center load tripped offline in Northern Virginia after a single line fault, and the grid held. That is the hazard behind NERC's rare Level 3 alert of May 2026: when gigawatt-scale demand vanishes at once, grid frequency swings as if a large power plant had failed.
The second consequence is cost. U.S. data centers used about 4.4% of the nation's electricity in 2023, per Lawrence Berkeley National Laboratory, and in the PJM region the cost of new capacity tied to them is already reaching electricity bills. States are rewriting the rules on who pays, and in Southern Nevada federal assessments still rate near-term blackout risk as normal.
NERC's incident review documents about 1,500 MW of data-center load dropping off the grid during six faults in 82 seconds on July 10, 2024, and on July 22, 2026 roughly 3.8 GW tripped offline in Northern Virginia, per Utility Dive. Neither event caused a blackout.
The hazard is counterintuitive: the danger is demand disappearing, not demand spiking. Data centers protect their servers by shifting to backup power at the first sign of a voltage disturbance, which is sensible for one building and a system-wide shock when dozens of buildings do it in the same second. A loss that large and that fast sends grid frequency, the 60-hertz rhythm every generator must hold, swinging much as it would if a major power plant tripped, and that kind of shock can cascade into further generator trips and, in the worst case, a wider outage.
On May 4, 2026, the North American Electric Reliability Corporation (NERC) issued a rare Level 3 alert, its highest-urgency category, directing the utilities it oversees to report on seven essential actions for these loads. On July 16, 2026, the Federal Energy Regulatory Commission (FERC) went further: in Docket RD26-7-000 it ordered NERC to file mandatory reliability standards for data-center and other computational loads by December 31, 2026.
The longer-term math is just as demanding. NERC's 2025 Long-Term Reliability Assessment forecasts North American summer peak demand rising by more than 224 gigawatts over the next decade. A grid asked to do more, faster, with thinner reserves is the setting in which heat waves, peak-demand days and sudden load swings turn into outages.
No AI-driven blackout has happened. The point of the alert is that the warning signs are serious enough that the people who run the grid now treat reliability as a present-day problem, not a future one.
Not yet. August 3, 2026 was a reporting deadline, not a fix-it deadline: utilities answered 33 structured questions on NERC's seven required actions rather than certifying finished work, according to NERC's alert.
The seven actions cover better modeling, system studies, fault-recording equipment, commissioning tests and communication protocols for data-center loads. NERC's alert document is explicit about what that obligation actually is: "This Level 3 NERC Alert is not the same as a Reliability Standard, and it does not create a mandatory obligation to take the Essential Actions. Your organization will not be subject to penalties for failure to implement the Essential Actions."
The enforceable step comes from FERC's July 16, 2026 order: mandatory standards are due to be filed by December 31, 2026, and in August 2026 NERC posted a draft rule creating a new "Computational Load Owner/Operator" registration category so data-center operators themselves can be held to them. As of October 28, 2026, we found no published NERC summary of the August responses. Until the standards take effect, the May 2026 finding holds: the grid is on a documented path toward fixing this, not at the end of it.
In Southern Nevada, not on current federal assessments: NERC's 2026 Summer Reliability Assessment rates WECC-Southwest, the region covering most of Nevada including Las Vegas and Henderson, as normal risk, stating that "Expected resources meet operating reserve requirements under the assessed scenarios."
The longer horizon is tighter. NERC's 2025 Long-Term Reliability Assessment counts 9,422 MW of large-load additions planned in WECC-Southwest through 2035 and projects the region's reserve margin slipping to 10.7% that year, just under its 11.1% reference level, unless more supply is built. NV Energy has fielded roughly 22,000 MW of large-customer interest against a system peak of about 8,500 MW, according to News 4 Reno's May 2026 coverage of its resource plan, and not all of that will be built. For the outage-by-outage view, see our NV Energy blackout risk guide; for the local build-out, see the Henderson data center breakdown. The worst case is physical damage to the grid itself: large power transformers averaged 128 weeks from order to delivery in 2025, according to Wood Mackenzie. No agency has forecast an 18-month blackout for Nevada, but grid risk is now on Washington's agenda: federal regulators have ordered mandatory reliability standards for data-center loads, and on August 26, 2026 the President declared a national emergency over foreign-made grid equipment; our guide to Executive Order 14421 and long-outage risk covers what is known and what a home can do.
Nevada already has its own version of the data-center supply fight, at Lake Tahoe. In May 2026, Fortune reported that NV Energy would stop supplying Liberty Utilities, which serves about 49,000 customers on the California side of the lake, because it needs the capacity for data centers. NV Energy disputed that account and said "There will be no disruptions in power," per 2News. An interim supply agreement now runs through December 31, 2027, and California regulators approved Liberty's search for a new supplier in July 2026, according to CapRadio. It is a wholesale-contract dispute, not an outage.
Solar Resource USA's own citation tracking logged the question "Will data centers cause blackouts?" in 55 checks across Bing Copilot, Perplexity and Google AI Overviews between June 22 and September 28, 2026; this page was cited in 7 of those answers, and none of them came from Google's AI Overview.
U.S. data centers' electricity use climbed from 58 TWh in 2014 to 176 TWh in 2023, according to Lawrence Berkeley National Laboratory, which projects 325–580 TWh by 2028.
Training and running large AI models requires dense racks of power-hungry chips, and the hyperscale data centers that house them are built and connected in blocks far larger than the commercial loads utilities usually plan for.
The scale is what makes this cycle different. In the PJM region, PJM's board chair has said peak load is projected to grow 32 GW from 2024 to 2030, with all but 2 GW coming from data centers. The grid that has to absorb this was largely built decades ago.
In the PJM region, yes. PJM's July 2026 capacity auction cost about $16.4 billion, cleared at its price cap and still fell 6,831 MW short of target; PJM's independent market monitor attributed $6.3 billion, or 38%, to data centers, per Utility Dive.
PJM, the largest U.S. grid operator, spanning 13 states and Washington, D.C., has now run two $16.4 billion auctions in a row: the December 2025 auction for the 2027/28 delivery year also cost about $16.4 billion, with the market monitor tying roughly $6.5 billion, about 40%, to data-center load. Capacity costs flow into retail bills on a lag that varies by state and utility.
Longer-range estimates are larger. The Natural Resources Defense Council has estimated added costs of about $70 a month for an average household in the PJM region over 2028–2033. The Federal Reserve Bank of Dallas found existing data centers have already raised wholesale power prices 3 to 5% on average nationwide, with about 50% possible in its high-build scenario. In Virginia, Dominion Energy sought its first base-rate increase since 1992, and on November 25, 2025 the State Corporation Commission approved a separate rate class for customers of 25 MW or more, starting in 2027.
The mechanism is simple: new power plants, transmission lines and substations cost tens of billions of dollars, and under traditional rate-making those costs are spread across all customers. That is why the question of who should pay has become a national fight.
Increasingly, the data centers themselves, on paper. By September 2026, 25 states had approved at least one "large-load" tariff and 7 more had one pending, according to the Edison Electric Institute; these tariffs make very large users, not households, carry the cost of the capacity they require.
At the federal level, the White House launched a Ratepayer Protection Pledge in March 2026, a voluntary commitment by major technology companies aimed at keeping the cost of serving their data centers off household bills. By July 23, 2026, the White House said the pledge, now joined by utilities, cooperatives, developers and 23 governors, covered roughly 80 percent of the power delivered to U.S. homes and businesses. In state capitols, more than 300 data-center bills were filed across 30-plus states in the first six weeks of 2026, per MultiState.
Whether these rules actually hold down household bills is still unproven: the pledge is a voluntary commitment, not a rate order, and an approved large-load tariff does not automatically wall off every dollar from residential rates. But the direction of travel is clear. The era in which data centers were courted purely with incentives is giving way to one in which they are expected to pay their own way.
The capacity-auction mechanism described above is a PJM Interconnection story: it applies directly to the 13 states and D.C. inside that grid operator's footprint. Nevada and Arizona sit in different grid regions with their own market structures, so that auction number does not show up on an NV Energy or APS bill the same way. The national capacity crunch still matters here indirectly, as part of why Western utilities are also building fast and filing rate cases, but the numbers are not interchangeable. For the local picture, see our companion guides: Las Vegas Data Centers vs. Your Power Bill, the Henderson breakdown, and NV Energy's 2026 rate case. For Arizona homes, which buy power from APS or SRP rather than NV Energy, our Phoenix solar guide and Scottsdale guide to APS vs. SRP cover the local utility side.
For most households, the AI build-out shows up as two slow-moving risks: electricity that costs more over the coming years, and a grid running with a thinner reserve margin during extreme heat and peak demand. Neither is something one household can fix. For how this national trend meets local reliability, see our Las Vegas grid reliability and power outage guide.
What a homeowner can control is their own exposure. On-site solar reduces how much grid energy you buy at rising prices; a home battery keeps essential circuits running when the grid is down.
Backup power is hard to justify on grid-risk grounds alone where outages are rare and short. The U.S. Energy Information Administration counted about 11 hours without power for the average U.S. customer in 2024, and roughly 2 hours once major events such as storms are excluded. A renter, a household planning to move within a few years, or a home whose roof is shaded most of the day is usually better served by efficiency upgrades and a plan for medical or refrigeration needs than by new equipment. An unbiased solar advisor should say so before quoting anything.
Are AI data centers raising electricity bills in the United States?
In the PJM region, measurably yes. PJM's July 2026 capacity auction cost about $16.4 billion, and its independent market monitor attributed $6.3 billion, or 38%, to data-center demand; the December 2025 auction had a similar cost and share. The Federal Reserve Bank of Dallas estimates existing data centers have already raised wholesale power prices 3 to 5% on average nationwide, and the NRDC has estimated added costs of about $70 a month for an average PJM-region household over 2028–2033. Outside PJM, the effect runs through each utility's own rate cases.
How much electricity do AI data centers use?
U.S. data centers used roughly 176 terawatt-hours in 2023, about 4.4 percent of national electricity, and Lawrence Berkeley National Laboratory projects that could reach 325–580 TWh by 2028. In the PJM region, PJM's board chair has said peak demand is projected to grow about 32 gigawatts between 2024 and 2030, with all but roughly 2 gigawatts coming from data centers.
Can AI data centers cause blackouts?
They can contribute to one, and regulators treat the risk as real, but no U.S. blackout has been attributed to data-center load. The danger is sudden load loss: on July 22, 2026, about 3.8 GW of data-center load tripped offline in Northern Virginia after a single line fault, the largest such event in PJM's history, and the grid held. Losses that size swing grid frequency much like a large generator tripping, which in a worst case can cascade. NERC issued a rare Level 3 alert on May 4, 2026, and FERC has ordered mandatory standards to be filed by December 31, 2026.
Will data centers cause blackouts in Las Vegas or Nevada?
Not on current federal assessments. NERC's 2026 Summer Reliability Assessment rates WECC-Southwest, the region covering most of Nevada including Las Vegas and Henderson, as normal risk. The longer view is tighter: NERC counts 9,422 MW of large-load additions planned in WECC-Southwest through 2035 and projects the region's reserve margin falling to 10.7% that year, just under its 11.1% reference level, unless more supply is built.
Is Lake Tahoe losing power because of Nevada data centers?
No outage has happened or been scheduled. In May 2026, Fortune reported that NV Energy would stop supplying Liberty Utilities, which serves about 49,000 customers on the California side of Lake Tahoe, because it needs the capacity for data centers; NV Energy disputed that account and said there would be no disruptions in power. An interim supply agreement runs through December 31, 2027, and California regulators approved Liberty's search for a new supplier in July 2026.
Should I buy backup power because of data centers?
Not on data-center risk alone. No U.S. blackout has been attributed to data-center load, and the U.S. Energy Information Administration counted about 11 hours without power for the average customer in 2024, roughly 2 hours once major events such as storms are excluded. Backup power is worth pricing when outages are already frequent where you live, when someone at home depends on powered medical equipment, or when you plan to stay in the home long enough for the equipment to earn its cost. Renters and households planning to move soon are usually better served by a plan for refrigeration and medical needs.
Why did NERC issue a Level 3 alert about data centers?
NERC issued the Level 3 "Essential Actions" alert on May 4, 2026 because very large data-center loads were tripping offline at the same time in response to minor grid disturbances. When a gigawatt-scale block of demand vanishes in seconds, system frequency can spike and set off a cascade of further equipment and generator trips. The alert set out seven essential actions, and the utilities it oversees had to report on them by August 3, 2026.
Did NERC's August 2026 data center deadline actually fix the grid?
Not yet, and NERC's own alert says so directly. August 3, 2026 was a reporting deadline, not a fix-it deadline: registered utilities had to answer 33 structured questions on how far along they were on NERC's seven required actions, not certify the work was done. NERC's alert itself says it is not a Reliability Standard and carries no penalties for failing to take the actions. The enforceable step is FERC's July 16, 2026 order, which requires NERC to file mandatory standards by December 31, 2026.
Who pays for the grid upgrades that data centers require?
This is the central policy fight. Historically, the cost of new generation and transmission has been spread across all customers, which means households can subsidize industrial users. By September 2026, 25 states had approved at least one large-load tariff designed to make very large users cover the cost of serving them, and 7 more had one pending, according to the Edison Electric Institute. At the federal level, the White House Ratepayer Protection Pledge, launched in March 2026, is a voluntary commitment by major technology companies aimed at keeping data-center power costs off household bills; by July 23, 2026 the White House said it covered roughly 80 percent of the power delivered to U.S. homes and businesses. The pledge is voluntary, not a rate order.
Which U.S. states are most affected by data center electricity demand?
Northern Virginia, often called Data Center Alley, is the clearest example. Its main utility, Dominion Energy, sought its first base-rate increase since 1992, and on November 25, 2025 Virginia regulators approved a separate rate class for customers of 25 MW or more. By September 2026, 25 states had approved at least one large-load tariff for hyperscale users, per the Edison Electric Institute. Western states such as Nevada and Arizona are also seeing rapid build-out, though the PJM auction mechanism does not apply outside PJM's footprint.
We're an independent solar advisor — not an installer and not a utility. If you want a straight, numbers-first read on how rising rates and grid risk affect your home, and whether solar or battery storage actually pencils out for you, we'll walk through it with no pitch and no data resale.
Talk It ThroughSolar Resource USA · America's unbiased solar advisor · Free consultation