The power grid jolts every minute: now we know why, and I simulated it

At the start of every minute, the power grid jolts. Its frequency, the value that in Chile sits around 50 Hz and gives away any mismatch between what is generated and what is consumed, dips for an instant right as the clock hits second zero. It happens in grids on several continents, and for years nobody could explain it well.

A team at Forschungszentrum Jülich, with the University of Cologne and KIT, published the cause in Nature Energy, and the press release came out on 30 October 2026. It caught my eye because the culprit is so ordinary, and because I could not find a single story about it in Spanish. So I am bringing it over, and instead of repeating the press release I wrote a small simulation to see, with my own numbers, how much the coincidence weighs.

Also on YouTube as a Short (54 s).

What was reported

These are the facts from the study and its press release, each confirmed by at least two sources (see Sources):

• Grid frequency has a dip that repeats exactly every 60 seconds. It is not a European quirk: the study analyzed recordings from twelve grids on five continents.
• The work comes from Forschungszentrum Jülich with the University of Cologne and KIT, published in Nature Energy in October 2026 under the title “One-minute frequency disturbances from unintended synchronization in cyber-physical power systems.”
• They measured devices at 80 locations of Jülich’s Living Lab Energy Campus.
• In Ireland, where data centers with large batteries account for about a fifth of electricity demand, the amplitude of the one-minute signal quadrupled between 2016 and 2021.
• On each spike, hydroelectric turbines slow down for an instant and speed up again, which adds wear.

The cause: shared punctuality

Uninterruptible power supplies (UPS) are batteries that keep a data center, a hospital or an office running through an outage. To stay ready, they periodically check their state of charge, and that check draws a little extra power for a brief moment.

On its own, one battery’s check is negligible. The problem the Jülich team found is the schedule: many of these devices run it at the start of each minute, and since they all take their time from precisely synchronized clocks, they run it in the same second. Thousands of small loads land together and turn into a step the grid does feel. The paper’s title calls it unintended synchronization: nobody designed it that way, but the combined effect is real.

The press release frames the batteries as the main culprit, not the only cause: the authors do not rule out that other digital devices taking their time from the same clock add their share, and they observed similar patterns in things as varied as a networked oven and small computer clusters.

All at second 0

summed load

Each at a random second

summed load

My demonstration: 2,000 batteries in a script

To see the size of the effect I wrote a dependency-free Python script that does the simplest possible thing. There are 2,000 batteries, each draws 1 unit for 1 second once a minute, and I add up the load of all of them second by second over 10 minutes. I run it in two cases: all of them check at second 0, or each one picks a random second between 0 and 59 at the start and keeps it. The seed is fixed, so the result comes out the same on any machine.

python3 lab/red-electrica-cada-minuto/sim.py

Average load comes out identical in both cases, 33.33 units per second, because the batteries use the same energy. What changes is how it is spread:

• Synchronized: the peak is 2,000 units at second 0 and the other 59 seconds sit at zero.
• At random: the highest second of the minute reaches 47 units and the lowest 22.
• The synchronized peak is 42.6 times the random peak.

The synchronized case is trivial by construction (2,000 batteries in the same second make 2,000), so the figure the simulation adds is the other one: spreading checks at random does not leave the load perfectly flat, it leaves seconds ranging from 22 to 47 units, and the peak still drops more than 40 times. That is why the authors’ fix is so cheap: it needs no new batteries and no lower consumption, only a different moment for the check.

Measured here (simulation, 2,000 batteries)

Peak, all at second 02,000 units/s

Peak, each at a random second47 units/s

×42.6times higher synchronized

From the study (not measured here)

×4Ireland: growth of the one-minute signal, 2016 to 2021

The simulation does not model grid frequency, turbines or how operators respond; it only shows why lining up in time turns a negligible load into a spike. The fourfold growth in Ireland is from the study; I did not measure it.

Why it matters

A momentary frequency dip does not switch off anyone’s home, and the study describes no outages from this cause. What it describes is a cost that piles up: hydro turbines braking and accelerating sixty times an hour, every day. And it is a cost that tracks the expansion of data centers. In Ireland, where such centers with large batteries account for about a fifth of demand, the signal quadrupled between 2016 and 2021, though the timing alone does not isolate how much of that growth comes strictly from them.

It also leaves a design lesson that applies far from electricity. When thousands of independent devices share the same clock and the same rule (“do it when the minute starts”), they end up acting as one. Scheduled jobs on servers that all hit a database at midnight do the same, and the remedy is the same too: add a random delay. On the grid, that delay is the paper’s concrete proposal.

Sources

• Nature Energy · One-minute frequency disturbances from unintended synchronization in cyber-physical power systems (en, paper)
• Forschungszentrum Jülich · Mystery of 60-second fluctuations in power grids solved (en, press release, primary source)
• scinexx · Ursache für regelmäßige Leistungsspitzen im Stromnetz entdeckt (de)
• WinFuture · report on the Jülich study and the 60-second fluctuation (de)
• Solarserver · Forscherinnen lösen Rätsel um 60-Sekunden-Schwankung in Stromnetzen (de)

Frequently asked questions

What is the power grid's once-a-minute jolt?

It is a brief dip in power grid frequency that repeats exactly every 60 seconds, at the start of each minute. According to Forschungszentrum Jülich it shows up in grids on several continents, not only in Europe, and tracking down its origin took years.

What causes it?

The study published in Nature Energy in October 2026 points to large uninterruptible power supply (UPS) systems, the backup batteries of places such as data centers, as the main culprit. They check their state of charge at the start of every minute, synchronized by the clock, and those brief loads add up in the same instant. The press release frames it as the cause that appears to dominate, without ruling out that other clock-synchronized devices also contribute.

How did they find out?

The Jülich team, with the University of Cologne and KIT, measured devices at 80 locations of Jülich's Living Lab Energy Campus and found the same pattern wherever large backup batteries ran their check at the start of each minute.

Is it dangerous for the power supply?

The study does not describe blackouts caused by the effect. What it points to is wear: on each spike, hydroelectric turbines slow down for an instant and speed up again. The signal also grows where data centers expand: in Ireland, where they account for about a fifth of demand, its amplitude quadrupled between 2016 and 2021, though the timing alone does not isolate how much of that growth comes strictly from them.

What is the fix?

The authors propose that each battery run its check at a random moment of the minute instead of all at second zero. Total consumption stays the same; what goes away is the peak. In my simulation with 2,000 batteries, the per-second peak drops from 2,000 to 47 units, 42.6 times lower.

What exactly does this post's simulation measure?

Only how much load coincides in each second: 2,000 batteries drawing 1 unit for 1 second once a minute, either all at second 0 or each at a random second, over 10 minutes. It does not model grid frequency or turbines. The script lives in the blog's repository (lab/red-electrica-cada-minuto/sim.py) and is deterministic.

Written on 1 November 2026. What was measured is the simulation lab/red-electrica-cada-minuto/sim.py (Python 3, standard library only, seed 42, 2,000 batteries, 10 minutes): a peak of 2,000 units per second synchronized versus 47 at random, a ratio of 42.6, with the same average of 33.33 (file sim.json). The 60-second periodicity, the 80 measurement points, the fourfold growth of the signal in Ireland between 2016 and 2021 and the turbine wear come from the paper and the Jülich press release, not from a measurement of my own.