One zone, six hours: what the Google Cloud outage in London shows
On 20 October a power fault in one Google Cloud zone in London disrupted banks for hours, Revolut in Switzerland too. What failed and what to check now.
On Tuesday 20 October 2026, one of the three zones of Google Cloud's London region lost power. For several hours, banks, couriers and online shops in Britain were partly or fully unreachable, and bank customers on the European mainland felt it too, Revolut's customers in Switzerland among them. The cause, according to Google, was a fault in the equipment that is supposed to make a power cut harmless.
What happened
At 05:53 UTC, the grid power feeding the data centre behind the zone europe-west2-b was disrupted. The batteries took the load, as they are meant to. The generators did not. According to Google's preliminary incident report, both the primary and the backup controllers that start the generators and switch the load to them "failed to initiate". At 06:00 UTC, 08:00 in Switzerland, the batteries were empty, and the hosts, network switches and local storage in that zone went dark.
Google counts the disruption as six hours and ten minutes, until 12:10 UTC, when more than 99 % of its Compute Engine virtual machines were healthy again. It posted its closure notice at 14:03 UTC and published the preliminary report on 22 October. A final report with preventive measures is still to come.
The power cut hit one zone. The effects reached the whole region. Google's report says that failover traffic "saturated available compute slots and processing capacity in other availability zones", and that checks in its control plane which cross zone boundaries spread the impact further. Regional Persistent Disk and Cloud SQL are named among the services with failover and degradation problems. During the incident, Google's only workaround was to move workloads to another zone or region "if feasible".
The list of those affected was long. Reuters reported disruption at Revolut, Lloyds and HSBC. The courts service in Britain reported problems with the video platform it uses for some remote hearings. In Switzerland, Revolut, which has 1.3 million customers here, reported problems with card payments, transfers and cash withdrawals, according to Netzwoche. At 17:00 Swiss time, an hour after Google had declared the incident over, Revolut's status page still showed 27 of its services in Switzerland as partly unavailable. Hostinger and MongoDB Atlas also reported problems linked to the outage.
The regulators moved within a day. The European Banking Authority contacted Google Cloud to assess the impact on EU financial firms. Google is one of the 19 providers that the European Supervisory Authorities designated as critical ICT third-party providers under DORA in November 2025. The EBA added a sentence worth reading twice: the designation "does not transfer responsibility for incident management from the provider or relieve financial entities of their own regulatory obligations". Google answers for its incident. The bank still answers for its service, whoever runs the servers.
What it means for a company or an administration
The first lesson is in Google's own documentation: "Zones should be considered a single failure domain within a region." On 20 October the zone was a data centre whose generators did not start. Anything that runs in one zone only, a single virtual machine, a disk, a database without a replica elsewhere, shares the fate of that building's generators. That is not special to Google. Most data centres rely on the same chain of batteries, generators and automatic controls, and that chain can fail anywhere.
The second lesson is less comfortable. Spreading a service over several zones was not enough on 20 October. When one zone fell away, the work moved to the other two, and they ran short of capacity. A plan that assumes spare room somewhere else only works if that room is reserved and the switch has been tried under load.
The third is about time. Google's figure was six hours and ten minutes. For Revolut's customers in Switzerland it was longer. Your outage lasts until your own service is back, not until your provider's status page turns green. Google's own recovery took staged restarts, cleared backlogs and checks of health and storage, and every customer has its own version of that list.
The fourth is about the chain. People and companies who never signed a contract with Google were hit anyway, because their bank or their hosting company runs there. The question is whether you know which cloud your own suppliers depend on, and in how many zones.
Where Elchi Studios stands
Our services, and the company websites we run, are on our own cluster: eight servers at five providers in four countries, Switzerland, Germany, the Netherlands and France. I chose that spread so that the loss of one building or one company does not take the whole service with it.
- Two entrances, where a browser's connection arrives: Hetzner in Falkenstein and UpCloud in Amsterdam. When one fails, the other takes the traffic.
- Two application nodes, both serving at the same time: Infomaniak in Geneva and Scaleway in Amsterdam.
- Three database nodes, in Frankfurt, Nuremberg and Paris. A change counts as written only once all three hold it.
The part with the least spread is stored files. EMX keeps messages and attachments encrypted in Geneva, and profile pictures and logos sit in object storage at Infomaniak, with a nightly copy at Scaleway. A nightly copy protects against losing files. It does not keep them reachable while the original location is down.
Two more things, in the same spirit as the EBA's sentence. First, we have not yet switched off a whole location under load and published the result. Until we have, the design above is a plan and not a measurement, and the result will be reported in this journal whatever it is. Second, the failure of a provider we chose is our failure. For hosted client projects we promise 99.99 % a year, and outages of the hosting providers and DNS services we picked count against it, as we explained when we lowered that promise from 99.999 %. Four independent monitors, running in different places, watch our services, and status.elchi.dev publishes what they measure.
What to check this week
- List what runs in one zone only. Virtual machines, disks, databases without a replica, a cache, a queue. For each, write down what happens to your service if that zone disappears for six hours.
- Check that the surviving zones can carry the load. If your failover relies on capacity being free elsewhere, reserve it or test it. On 20 October the capacity in the other London zones ran short.
- Look for dependencies that cross zones. A control plane, a configuration store, a secrets service or a licence check that sits in one place can stop the rest from starting.
- Ask your suppliers where they run. Your payment provider, your hosting company, your database service: which provider, how many zones, how many regions, and how they will tell you when something breaks.
- Measure from outside. A monitor outside your own infrastructure tells you when your customers are affected. Your provider's status page tells you when the provider is affected. On 20 October those were different times.
- Read the availability clause in your contracts. Check the figure, how it is measured, and whether the failures of your provider's own suppliers count or are excluded.
- Plan for a whole region. Google itself advises a disaster recovery plan for the loss of a region. A backup in another region only counts once a restore from it has been tried.
Sources
- Preliminary Incident Report: multiple services in europe-west2 (London), Google Cloud Service Health, read
- Geography and regions, Google Cloud, read
- Google Cloud says a 'power event' caused today's outage, The Register, read
- Europe's financial regulators query Google over cloud outage that hit banks, Reuters, via Global Banking & Finance Review, read
- European banks, websites hit as Google Cloud data centre goes offline for hours, ABC News, read
- Störung bei Google Cloud beeinträchtigt zahlreiche Dienste in der Schweiz und Europa, Netzwoche, read
- Grossausfall bei Google Cloud, Inside IT, read
- EU regulators designate critical tech providers under DORA, Pinsent Masons Out-Law, read
- The October 20, 2026 Google Cloud europe-west2 Outage, IncidentHub, read
Written by Samuel Krauss, Founder. Filed under hosting, availability, cloud, operations.