Understanding US Power Outages

Modern civilization relies on electric power for almost everything, and even small disruptions to electric service are incredibly disruptive. Because of this, we demand a high level of reliability in electrical service. In 2023, the average US electricity customer was without power for only 366 minutes over the course of the year, equivalent to a service uptime of more than 99.9%. Other countries do even better: in 2021 the average German customer was without power for just over 12 minutes, an uptime of greater than 99.997%.

Because reliable electrical service is so important, I wanted to better understand trends in US power outages. I partnered with Poweroutage.us, a service that provides live and historical power outage data for the US going back to 2017, on a county-by-county and utility-by-utility level. Their API not only gives live outage counts, but the "shape" of outages reported by the utilities. Poweroutage.us tracks approximately 94% of US electrical customers, letting us take a close look at power outage trends in the US.

This dataset will let us understand broad trends in US power outages, but let’s first start at the bottom, and look at power outages in a couple of counties. Below are the last seven years of power outages for Cobb County, Georgia (where I live), and Los Angeles County, California (the most populous county in the US). The graphs show power outage minutes per week, where a power outage minute is one customer being without power for one minute.

These graphs show two very important facts about power outages.

The first is that power outages are heavily driven by a small number of extreme events. Very bad power outages will occur relatively infrequently, but can be hundreds of times worse than the typical week. In the week of October 26th 2020, Cobb County had a huge spike in power outages due to Hurricane Zeta: the 156 million outage minutes that week were about 300 times higher than average. Similarly, Los Angeles had a huge spike in power outages in January of this year due to the Eaton Fire. Outage minutes in LA the week of the fire were more than 80 times higher than the average week over the previous year.

The second important fact is that these extreme events are highly regional in nature. Hurricane Zeta had no impact on electrical service in California, and the Eaton fire had no effect on electrical service in Cobb County. There’s a great deal of geographic variation in power outages more generally: even two counties in the same state might have drastically different power outage frequencies. Hurricane Zeta caused enormous power outages in Cobb County, GA, but had virtually no impact on Richmond County less than 200 miles away. The worst power outages in Richmond County were caused by Hurricane Helene, which had little impact on Cobb County.

It’s important to keep these facts in mind when analyzing power outages, because they can easily make summary statistics misleading. In 2020, Cobb County had an average of 3.5 million power outage minutes per week, but this masks the fact that there was one week with 156 million outage minutes, and 51 weeks with ~550,000 outage minutes each. Averages for the country as a whole, or even individual states, will conceal a great deal of regional variation, and averages over long time periods will conceal brief but severe spikes in the number of outages.

Keeping this caveat in mind, let’s look at nationwide power trends. Here are average US power outage minutes per customer over time. For comparison, I’ve graphed them against the EIA values for the same measure.¹

We can see that the Poweroutage.us and EIA values track each other very closely. We can perhaps see a general trend upward in outage minutes per customer (though it’s hard to tell), with 2024 being particularly bad. Outage minutes per customer in 2024 were more than 50% higher than in 2023. This uptick in outages was driven by the twin impacts of Hurricane Helene and Hurricane Milton, which caused such severe outages in the Southeast that they drove up the nationwide average. Here's the nationwide average but with Florida, Georgia, South Carolina and North Carolina omitted.

Excluding the Southeast states hit by Helene and Milton reduces average US outage minutes per customer by roughly a third.

Another important trend: power outages tend to be more severe in rural areas than in highly populated urban areas. Here are nationwide outage minutes for the 100 most populous counties in the US (which combined have around 40% of the US population), both including and excluding FL, GA, SC and NC.

Outside of the Southeast, power outage minutes per customer in the 100 most populous counties was less than half the value in the country as a whole in 2024 (as well as 2022).

You can see the impact of major storms even more clearly by looking at a county-by-county map of power outages. Here’s outage minutes per customer for each US county in 2024.²

You can clearly see the path of Hurricane Helene across the Southeast, as well as the impact of large thunderstorms in southeast Texas in May. And here’s a map of outages in 2021.​​

In contrast to 2024, we see few severe outages in FL, GA, SC, or NC. Instead, the major power outages in 2021 were in Texas (due to Winter Storm Uri), Louisiana (due to Hurricane Ida), California (due to wildfires), and Michigan (due to severe storms).

This county-by-county look gives us a very granular look at regional variation in power outages, but it makes it hard to understand trends over time (it’s hard to compare week-by-week or even year-by-year numbers for 3000 different counties). To better understand this, we can look at outage minutes by state over time.

There is lots of variation from state to state (in any given year the worst state is 20-50 times as bad as the best state) and from year to year. I counted 17 times when outage minutes per customer varied by a factor of 4 or more from one year to the next, and in several cases it varied by more than 10 times.

We can also look at power outages broken down by metro areas, which gives us a reasonably granular look at different regions while still letting us see trends over time. Below are power outage trends in 12 of the largest metro areas in the US, which collectively have a population of around 100 million (about 30% of the country).

You can clearly see how a small number of extreme events drive power outages (the graphs above are all flat-ish lines with the occasional huge spike), and how these are relatively uncorrelated from region to region. (You can also see that Miami hasn’t seen a major hurricane in the last 7 years.) A quick explanation of some of the causes of the above power outages are:

• March 2018, New York, DC, Philadelphia and Boston: Winter Storm Riley.

• June 2019, Dallas: Large thunderstorms.

• October 2019, Bay Area: Pre-emptive power shutoff to avoid wildfires.

• August 2020, Chicago: Large derecho.

• August 2020, New York: Transmission system failure.

• October 2020, Atlanta: Hurricane Zeta.

• February 2021, Dallas and Houston: Winter Storm Uri.

• October 2021, Boston: Large nor’easter.

• May 2024, Dallas and Houston: Large thunderstorms.

• January 2025, Los Angeles: Eaton Fire.

We can also see this by looking at the fraction of power outage minutes that comes from just the five worst weeks in each metro:

In 6 of the 12 major metros we looked at, from 2018 through 2024, the five worst weeks of power outages (just over 1.3% of weeks) were responsible for more than 50% of total outage minutes.

Knowing that extreme events are such large drivers of power outages, one obvious question is “are these extreme events getting more frequent”? Because they occur relatively rarely, and are highly regional in nature, this can be hard to tease out: Cobb County, Richmond County, and LA County have each seen just one extreme event apiece over the last 7 years. One way to see if extreme events are increasing is to look at nationwide trends in large-scale outages. Here’s a state-by-state look at number of severe outages over time, where “severe outage” is a week where outage minutes per customer exceeds 50:

Lots of variation, but the trend is definitely upward (though not uniformly in every state). Large-scale power outages are getting worse over time.

Power outages outside extreme events

While it’s interesting to note the extent power outages are driven by a small number of extreme events, and that these events are getting more frequent, this can only tell us so much about overall electrical system reliability. While we want to know how likely we are to lose power in a major storm, we also want to know how reliable we can expect service to be on a typical day outside these extreme events. Because the extreme events make it hard to spot trends in more typical day-to-day service, we can get a sense of typical reliability by looking at the data with extreme events removed. Here’s a county-by-county and state-by-state map of outage-minutes per customer in 2024, with the three worst weeks for each county removed.

When we’ve removed outliers, the worst areas in the country are Arkansas, Mississippi and east Texas, western Texas, West Virginia, Maine, and northern/eastern California. The best areas in the country are the Midwest and southern New England. If we look at 2023 and 2022, we see a roughly similar pattern, though now the northern parts of the Midwest look worse.

If we look at outage minutes per customer with outliers removed on a state-by-state level over time, we don’t see much of a trend. While there’s variation from state to state, it’s not trending in any particular direction.

We see something similar if we look at metro areas. Here are the same 12 metro areas we looked at previously, with the five worst months for each removed.

Some metros do seem to be getting worse over time — Dallas, Philadelphia, Atlanta, and the Bay Area — but for most metros the trend is relatively flat.

This view also lets us see which metros have the best baseline reliability. Phoenix, Miami, and Los Angeles all do well here, with very low typical outage minutes, while Philadelphia, Houston, and the Bay Area do much worse.³

We can also use metro data to look at seasonal trends in power outages. The graph below shows the average number of outage minutes for each month for each of the 12 metros considered, with the five worst months for each metro removed.⁴

We can see that power outages peak in the summer and (in many metros) the winter, and decline in the fall and the spring. It’s not obvious to me the extent that this is driven by seasonal weather (such as summer thunderstorms and winter snowstorms), and to what extent it's driven by seasonal energy use (air conditioning and heat pumps driving up summer and winter electricity consumption).

We can also see the wide variation in baseline electricity reliability from region to region. Chicago and Phoenix have fewer than 100 average outage minutes per customer per year, while Houston and the Bay Area have more than twice that.

Conclusion

To sum up:

• Power outages in the US are heavily driven by extreme events. A bad week for a city, county, or state can be hundreds of times worse than the typical week, in terms of the number of customers affected and the duration of outage.

• Power outages are also highly regional. The hurricanes, wildfires, winter storms, and other extreme events that cause large-scale power outages are typically localized to specific parts of the country, and often to specific portions of a state.

• Large-scale power outages caused by extreme events are getting more frequent over time.

• Outside of extreme events, different areas of the country exhibit very different “baseline” levels of reliability. Certain areas of the country (east and west Texas, West Virginia, Maine) show consistently high rates of power outages. Rural areas have significantly worse power reliability than dense, urban areas. Outage frequency can vary by a factor of two or more between major cities.

• Baseline power reliability in the US as a whole doesn’t seem to be changing much, but there is variation from region to region. Some areas are getting less reliable over time, others staying relatively flat.

• Power outages peak in the summer and, in some places, the winter.

Thanks to Poweroutage.us for making the data available for this essay. All errors are my own.

1. A note on methods: this is total minutes of power outage, divided by my estimated number of accounts, which I constructed by using the total number of US households plus an additional factor to take into account commercial and industrial accounts. Number of customers for states and counties were similar estimated by multiplying county population by the ratio of total US accounts to total US population.

2. A note on methods. One challenge when analyzing this data is that utilities don’t always indicate when a power outage ends. A naive analysis of the data would occasionally show extremely long power outages (6 months or more) because the utility didn’t indicate when one power outage ended and another started. I was able to filter out the most egregious cases, but in some cases it’s not obvious whether something is a very long (2+ day) outage or whether data is simply missing, so there’s likely still an overestimation of outage minutes in some cases. I suspect this is particularly prevalent in some rural counties with small utilities that aren’t necessarily incredibly diligent about updating outage status. So some of the random rural counties with very high numbers of outage minutes may be overestimates.

3. I had hoped to do an analysis of individual utilities, but this proved difficult for a few reasons. The biggest was that in many areas of the country a single utility is dominant, so it’s hard to separate regional effects from a utility’s performance. Only in a very small number of counties would it be possible to compare different utilities. It was also hard to get good estimates of the number of customers for a utility. I could get them for a utility overall, but not for individual regions like counties.

4. This will be biased somewhat by the fact that some metros are getting worse over time, but it’s still a useful view of the data.