A broken cable, a planned repair and a controller fooled by a live wire knocked banks, courts and deliveries offline for hours.
The morning London's cloud went dark
On 20 October, part of Google's cloud in London lost power. A cloud is a large set of rented computers in big buildings, and many UK services run on it. For hours, many people had trouble with their banking apps. The video system for remote court hearings went down. Delivery networks were hit too.
On 4 November, Google explained what went wrong. An underground cable outside the building broke. Batteries took over, as planned, but only for minutes. The diesel generators that should have come next never started. Why would a backup generator simply stay off?
The switch the controller could not see
The building has two power lines from the grid, called feed A and feed B. Either one can power the whole building on its own. On 14 October, a regular generator test found problems, so staff opened the switch on feed B to repair it. For a while, the building ran on feed A only.
Then the cable on feed A failed. The batteries ran out after about seven minutes. At that point a small control computer should have noticed the voltage drop and started the generators. But its sensors looked at the street-side wire of feed B, which still had power, not at the building's own supply. So the computer saw healthy power and never sent the start signal. That is why the generators stayed off.
There was a main controller and a backup. Both ran the same logic and read the same sensors, so both made the same mistake. Staff tried to order the switch-over remotely, and the controller blocked it for the same reason. In the end, they moved the building to a separate standby grid connection.
“I think the broken cable was bad luck, but a backup that cannot tell a closed switch from a working line is a design gap.”
Why one dark zone hurt the others
Google's London region has three zones. A zone is a separate group of data centers, built so that one can fail while the others keep going. Only one zone lost power. So why did the trouble spread?
Google says the work from the dark zone rushed to the other two at once, like shoppers all moving to the next checkout when one closes. The two zones could not take the sudden crowd, so queues grew and requests failed. Separately, some management systems, the programs that direct the machines, kept waiting for answers from machines that had no power. Customers in healthy zones saw errors too.
There was also nowhere else to go. London is Google's only cloud region in the UK, so customers without a multi-region plan had no second UK site to move to. Revolut, Lloyds, HSBC and Barclays all had problems. The Criminal Bar Association warned that trials and hearings "can not proceed".
Google says the main impact was over within 6 hours and 8 minutes. Some customers waited longer. That evening, only six of the twenty-seven services on Revolut's UK status page were working. Revolut says customers' money and data stayed safe.
Watchdogs, a rehearsal and a worldwide check
Under the UK's Critical Third Parties regime, regulators directly oversee a few outside suppliers that banks depend on. Since July 2026, Google's European cloud company has been one of four. On the day of the outage, Dame Meg Hillier, who chairs the Treasury Committee, wrote to the Bank of England and the Financial Conduct Authority, the UK's financial watchdog. She asked how many firms and customers were hit, how long the trouble lasted and what losses customers suffered. She wants full answers by 20 January 2027. Earlier that month, the Bank of England and 38 large banks and market operators had rehearsed a worldwide cloud failure.
Google has reset and tested the controllers. It plans a full review of the controller code and says it will make sure such reviews are done when a site is built. It froze planned maintenance at sites built the same way, and it plans to check controller settings at similar sites around the world by 15 November.
What stays with me is how ordinary each step was. A regular test, a planned repair, a cable that failed at a bad moment. Each one looks harmless on its own. I would like to know how many other buildings run on the same quiet assumption, and I hope the results of that worldwide check are made public.
Nov 10, 2026
