What happened in the 33 seconds the power took to die

12:33:30, Monday 28 April 2025. At that moment, by the clock of the Spanish report, the electricity system of Spain and Portugal went to zero. 33 seconds earlier it was still working. What happened in between I read twice in the non-confidential report of the committee the Spanish government set up, because the first time I understood none of it. It turns out that an electricity grid, when it falls, pushes itself over.

The European grid runs at 50 hertz, and that part is a child’s rule. Generate exactly what is consumed and it stays pinned there. Consume more than is generated and it drops. Generate more and it rises. Voltage is a different matter, and it is the one that counts here. A power plant does not only push electricity into the cables: it also has to hold the voltage up, absorbing when it climbs too high and releasing when it falls. That has to be done every single second. And that lunchtime there were fewer plants doing it than planned.

At 12:03 the grid began to oscillate. An oscillation is exactly what it sounds like, frequency and voltage swinging up and down instead of sitting still. It lasted until 12:07. At 12:19 a second one arrived, the kind that makes a whole region swing against another, and it ended at 12:22. Manoeuvres were carried out to settle it, and it settled. By 12:30 the system looked boring again, which in an electricity grid is the finest adjective available.

An electricity grid is a pile of enormous machines spinning at once, all in the same rhythm. When one region starts to run ahead of or behind another, that shows up as an oscillation, and if it is not settled in time the machines end up pulling on each other. The two from that lunchtime did settle, and the system carried on. What they did not fix was the other thing: that fewer plants than planned were holding the voltage up that day.

At 12:32:57 the first generating installation in the south disconnected. Then another. And another. In twenty seconds, 2,200 megawatts were lost across Granada, Badajoz and Seville.

And they disconnected because too much voltage was reaching them. When the voltage climbs above what a plant can take, it opens itself up so as not to break; that is how it protects itself and it is right that it should. The problem is what comes next. Every plant that opens stops absorbing voltage, so the voltage on the rest climbs a little higher, and that little bit opens the next one. And so on, in a chain, faster each time. The report calls it a chain-reaction overvoltage phenomenon, which is the same idea said the hard way.

With that much generation gone, the frequency started to fall. At 12:33:20 it crossed the first threshold, 49.50 hertz. From there an automatic system designed for exactly this kicks in, cutting the grid into pieces to save what is left. And it fired. It shed 2,037 megawatts of pumping, which are the plants that push water up into a reservoir when electricity is cheap, and that lunchtime they were pumping because the sun had made it cheap, and around 1,400 of consumption, which is what went in the first three of the six stages the system has, which on any other day would have been an enormous amount, and that day was not enough.

At 12:33:21 the alternating-current lines that join us to France tripped, the big ones, and the peninsula was left alone with its frequency falling. At 12:33:22 it dropped below 48 hertz, the last threshold there is. Eight seconds later, at 12:33:30, the Spanish report marks zero voltage. Thirty-one gigawatts of demand were disconnected. In Spain 25.2 and in Portugal 5.9.

The two reports do not give the same time, and neither warns you that the other counts differently. The European one puts the cut of the last cable joining us to France, the direct-current one, at 12:33:24 and closes its account there. The Spanish one carries on to zero, six seconds further out.

What happened afterwards is not in the reports in these words. Trains stopped wherever they were, tunnels included, and between commuter rail and the underground some 35,000 people were taken out of carriages, a good number of them walking along the track by phone torch. With no power the card terminals do not work, so half of Spain went back to cash, and the cash was inside cash machines that did not work either. The petrol stations could not pump. Garages with automatic doors stayed shut with the cars inside. And the old corded landline was one of the few things that kept a dial tone, because the telephone exchange has its own batteries and because that device needs no socket.

And people died. The counts from those days do not entirely agree with one another, which is worth saying before quoting a figure. At least seven deaths in Spain were counted as linked to the blackout, among them three members of one family in Galicia from carbon monoxide inhalation, and one in Portugal, a 77-year-old woman whose ventilator stopped.

While the power was coming back, four explanations were already going round the group chats. A cyberattack. Renewables. A test by the grid operator. A neighbouring country doing it on purpose. All four reached me within an hour and a half, two of them with a cropped screenshot.

The first formal answer arrived on 17 June 2025, 50 days later. The government committee published its non-confidential report, and it says three things. That what happened was that chain-reaction overvoltage phenomenon. That on that day fewer plants than planned had active voltage control, partly because one in the south had broken down the previous afternoon. And that no signs or evidence of a cyberattack were found. That last point is written out in full in a document approved by the National Security Council, and with it the biggest rumour collapsed.

Between June and March there was everything. Two parliamentary inquiries, a round of blame between the system operator and the generating companies, each with its own report and its own press office, and months of people quoting paragraphs they had not read. I downloaded both reports and what surprised me was how little they resemble the way it was all told that afternoon. What both of them contain is a list of things that were supposed to be holding the voltage up and were not doing it well enough.

The final report was published by the European association of transmission operators on 20 March 2026, based on the work of a panel of 49 experts including national regulators and people from the European agency of regulators. Its conclusion is the one that does not fit in a message. That the blackout came out of a combination of many factors acting on one another, and that the centre of it all was voltage control.

On the way, that report demolished the favourite theory of the comment sections, the inertia one, the idea that with more classic plants spinning nothing would have happened. They wrote that even with considerably greater inertia the loss of system synchronism would not have been avoided, which is when the machines stop spinning in the same rhythm and the system breaks apart. The grid was not in good shape, mind you. The recommendations point exactly there. Better control and supervision of whoever has to hold the voltage up, and more interconnection with France, the door through which the peninsula could have asked for help and which closed that Monday in three seconds.

To switch a dead grid back on you need electricity, which is the thing I did not know, because a big power plant needs power in order to start up. So you begin with the ones that can start themselves, the hydroelectric ones, and with help from the neighbours: that day current came in from France and from Morocco. With that you raise small islands, join them to one another, and in the end hook them all back into the same rhythm. That is what was going on while you stared at a phone with no signal.

Power came back over the course of the afternoon and the night, and by early on the 29th almost everything was restored. The report explaining why it went arrived 326 days later, on a Friday in March.