Google Cloud Outage: A Power Event in One London Zone Disrupted UK Banks and Courts

The Google Cloud outage that disrupted UK banks and the courts service on October 20, 2026 began with a power event in a single zone, europe-west2-b, inside Google’s London region. Google’s incident record lists 38 affected products and runs from 07:08 to 15:00 BST, just under eight hours.

The Google Cloud Outage Started With a Power Event in Zone europe-west2-b

Google’s service health dashboard dates the start of the incident to 23:08 Pacific time on October 19, which was 07:08 in London on October 20. For roughly the first five hours of the incident, Google described only elevated errors and intermittent network failures across the europe-west2 region and said engineers were investigating. At 04:12 Pacific time, or 12:12 BST, Google attributed the problem to a power event that had caused an outage in the europe-west2-b zone and said the power event was mitigated. A Google Cloud spokesperson later told the BBC the disruption happened at a third-party facility in the region. Google has not yet said what failed in the power supply.

Google Listed 38 Affected Products, From Compute Engine to Spanner

The incident record names 38 Google Cloud products, which covers most of the layers a typical application is built on. The list includes Compute Engine, Kubernetes Engine, Cloud Run and App Engine for compute; Persistent Disk, Cloud Storage and Filestore for storage; Cloud SQL, AlloyDB, Spanner, Bigtable, Firestore and several Memorystore variants for data; and Cloud Load Balancing, Cloud NAT and Virtual Private Cloud for networking. Google’s symptom note said services were unavailable in the affected zone, while other parts of the region might see elevated errors and intermittent network issues.

Revolut and the Courts Service Confirmed Disruption on the Day

BBC technology reporter Liv McMahon and James Kelly of BBC Your Voice reported that more than 1,500 problem reports about Revolut reached an outage monitoring site shortly after 08:00 BST. Revolut confirmed that some of its services were disrupted by the Google Cloud outage and said they were later fully restored. HM Courts and Tribunals Service said it was aware of an issue affecting the video platform used for remote hearings. Customers of Lloyds Bank, Halifax and Bank of Scotland also reported problems with online banking, though a Lloyds spokesperson did not confirm the cause. Users flagged issues with Royal Mail and DPD as well, and a Preston produce business told the BBC it could not process orders or book parcels with DPD during the outage.

Google’s Workaround Assumed Customers Could Move Workloads to Another Zone

From its second update onward, Google’s advice to affected customers was to migrate impacted workloads to another available zone or region if feasible. At 04:58 Pacific time, Google said it was prioritising recovery of existing workloads and that customers might not yet be able to create new ones. By 05:10 Pacific time, or 13:10 BST, Google reported recovery for most workloads in europe-west2. Its final update, posted shortly after 15:00 BST, said the issue was mitigated for most customers, that a very small number were still affected, and that Google would publish an analysis once its internal investigation was finished.

Why This Matters for Engineering Leadership

The single useful question from this incident is whether your service would have kept running if one zone in its home region had gone dark for six hours. Google’s published workaround was to move the workload, and its own recovery note said new workloads might not be creatable while it restored existing ones. A failover plan that depends on provisioning fresh capacity during the incident is relying on the same control plane that is under strain. Capacity that is already running and already taking traffic in a second zone does not need that step. Engineering leaders can test this directly: list every stateful dependency, including databases, caches, disks and message queues, and record which zones each one actually runs in today. Then schedule a game day that removes one zone and measure what breaks. Google has promised a written analysis, and that document will be worth reading against your own dependency list when it appears.

FAQ