U.S. Power Outages in 2025: Why a Better Average Still Hides Local Risk

The comparison that changes the story

EIA’s Electric Power Annual table 11.1 lists 662.6 minutes for 2024 and 269.2 for 2025 in its IEEE-method, all-events SAIDI column. That is an approximately 59.4% decline, calculated as (662.6 − 269.2) ÷ 662.6. The same method’s duration excluding major-event days falls only about 4.8%. The comparison therefore depends strongly on whether exceptional days remain in the measure. [1]

Averages do not describe every customer

EIA defines SAIDI as average annual interruption minutes, SAIFI as average interruption frequency, and CAIDI as average restoration duration per interruption. These describe different denominators. A count of affected accounts at 4 p.m. is none of those annual indicators. EIA publishes utility-level reliability information separately from national and state aggregates. [2]

Consider an illustrative system with 100 customers. If ten customers each lose power for ten hours, total customer interruption time is 100 hours and average duration is one hour per customer. If all 100 instead lose power for one hour, the average duration is identical. The service experiences are plainly different. This arithmetic example is not a measured utility result.

A second illustrative comparison shows why frequency should accompany duration. Ten one-hour interruptions and one ten-hour interruption produce the same total duration for an affected account, yet imply very different experiences of repeated disruption. An annual average can also conceal customers with no interruptions and customers with several long ones. A useful provider evaluation therefore asks for the distribution and consistency of experience, rather than translating a national average into a promise about the next interruption. Neither example changes EIA’s reported results; each identifies information an aggregate leaves unresolved.

Why weather belongs in the comparison

EIA’s retrospective on 2024 attributes about 80% of that year’s interruption hours to major events and identifies Beryl, Helene, and Milton as important contributors. It reports approximately eleven outage hours per customer that year. That historical explanation supplies context for an unusually high starting point; it does not establish why every utility changed in 2025. [3]

Our interpretation is that the national improvement should trigger more specific questions: did a particular provider improve on ordinary days, avoid a major storm, change its reporting coverage, or experience some combination? A two-year aggregate cannot isolate those mechanisms. Calling the entire decline an infrastructure success would require evidence beyond the national table.

How to evaluate your local reliability

EIA allows IEEE and other reporting methods in its reliability dataset. A defensible local comparison therefore holds the provider, reporting method, customer coverage, and major-event treatment as consistent as the records permit. Comparing an all-events number with a number excluding major events can produce a misleading ranking even when both are correctly reported. [4]

For a household, the useful evidence package combines the provider’s annual results with dated event information and its own interruption experience. For a regional researcher, it adds the distribution of duration across customers and locations. The national average remains valuable as a broad benchmark, while those additional observations explain how a specific community actually experienced the year.