A power event in Google Cloud's London region on Tuesday, October 20, 2026 caused elevated errors and intermittent network failures across multiple Google Cloud services, and Revolut said card payments, balance top-ups, ATM withdrawals and currency exchange were partially unavailable. Google says the cause was a power event in its europe-west2-b zone and that it was "seeing the recovery of services" across most of the region shortly after 1 p.m. local time. A person familiar with the matter told Reuters the Revolut outage was related.
Beyond that, the picture is muddy. Some posts have also pinned other banks and services on Google, but the only wire report we found says it is "not clear" whether the Google outage was behind the bank issues recorded by Downdetector. This post sorts what Google, Revolut and Reuters have said from what is still inference, then covers what builders should do about zone failures. It follows our earlier incident coverage of the Claude outage and network failures in July and the ChatGPT and Codex outage in September.
TL;DR: confirmed vs claimed
| Claim | Status | Source |
|---|---|---|
| Google Cloud had a London-region disruption starting just after 7 a.m. local time | Confirmed by Google | Google status page, quoted by CNN/Reuters |
| Cause was a power event in europe-west2-b | Confirmed by Google | Quoted by the Irish Times |
| Revolut services partly unavailable | Confirmed by Revolut | Revolut's statement and X post, as quoted |
| Revolut outage caused by the Google problem | Reported by Reuters via an unnamed source; Revolut's quoted statements do not name Google | CNN/Reuters |
| HSBC, Barclays, Lloyds outages caused by Google | Not established | Downdetector reports only, per Reuters |
| Customer money and data at risk | Revolut says safe | Revolut statement |
| Cyberattack | No evidence in any source reviewed | n/a |
| Full root cause (why backup power failed) | Not yet published | Google says it is investigating |
Timeline of what happened
- Just after 7 a.m. London time: Google reports "elevated errors and intermittent network failures" in europe-west2, the London region. Downdetector shows a spike in user reports between 7 and 8 a.m., per the Irish Times.
- Morning: Revolut posts that it is "currently experiencing a service disruption affecting multiple services" and that "all our customers' money and data are safe." Customers in the UK and Ireland report trouble logging in, transferring money and paying by card. Revolut has 3.4 million customers in Ireland, per the Irish Times.
- Shortly after 1 p.m.: Google says it is seeing recovery of services across most of the region and names a "power event" as the cause.
- Later: Google says the power event is mitigated and that the issue is "believed to be affecting currently a very small number of customers," with engineers still working on it.
Some secondary sites quote longer Revolut durations and larger counts of affected Google Cloud products. We could not trace those figures to Google, Revolut or Reuters, so we do not rely on them.
Why one zone took down so much
Google Cloud divides each region into zones, which are meant to fail independently. A service built across two or more zones should survive the loss of one. That a single-zone power event reached a bank's customer-facing services tells you one of several things, and the public record does not say which:
- The affected company ran critical components in only one zone.
- It ran across zones, but a shared dependency (a database primary, a queue, an identity service) sat in the failed zone.
- Failover existed but did not trigger or take too long.
- Network failures spread beyond the single zone, as Google's own wording about "intermittent network failures in europe-west2 region" hints.
This is not the first time a Google Cloud zone has lost power. Google's own incident report for us-east5-c in March 2025 describes a utility power loss, a failed battery in the uninterruptible power supply, and a 6 hour 10 minute degradation, with some services needing manual recovery. Google committed then to harden the power failure and recovery path. Whether the London event shares any cause is unknown until Google publishes its analysis.
Why an AI blog covers a bank outage
Because AI products sit on the same few clouds. Model APIs, agents, vector databases and the apps around them run in zones like europe-west2-b. When Google, Amazon or Microsoft has a regional incident, every AI feature routed through that region degrades together. The Irish Times pointed out that last October an AWS problem took down apps and banks, followed a week later by an Azure issue. As more critical workflows become AI-mediated, the concentration is the story. Our map of AI data center backlash, the power and tax fight in Virginia and the hyperscaler nuclear power deals all point to the same fact: power is the binding constraint on cloud and AI capacity, and the grid and backup chain are the weak points.
How to read an outage report like this one
Early incident coverage mixes three kinds of statements, and it helps to separate them. The first is the provider's own status text, which is precise but terse: "power event," "elevated errors," "recovery of services." It tells you the symptom and the scope Google chose to name, not the cause chain. The second is the customer's statement, which describes user impact in business terms (card payments, top-ups, ATM withdrawals) and almost always stresses that money and data are safe. The third is third-party signal, such as Downdetector spikes, which measures how many people complained, not why a service failed.
A Downdetector spike for a bank during a cloud outage is a lead, not a conclusion. Banks share telecom providers, card networks and payment rails, and a mobile-app problem can look identical from the outside. That is why Reuters wrote that it was unclear whether the Google problem explained the HSBC, Barclays and Lloyds reports, and why we keep them in the unconfirmed column until the banks or Google say otherwise.
What to do if you build on a hyperscaler
- Map zone dependencies. For each critical user journey, list every component and which zone it lives in. One single-zone database is enough to fail the journey.
- Test failover for real. A runbook that was never exercised is a hope. Schedule game days where you cut a zone.
- Decide what degrades. Revolut's outage hit specific functions (card payments, top-ups, ATM, FX) while others stayed up. Define in advance what is essential, what can queue, and what can show a read-only mode.
- Do not depend on one region for status communication. Keep your status page and customer messaging outside the cloud that may fail.
- Watch the provider's status feed, but trust your own monitoring. A status page reports what the provider has confirmed, not necessarily what your users see.
- For AI features, plan a fallback. If your agent calls a model hosted in one region, have a second provider or region, and make sure agents fail closed rather than taking actions on partial data. AgentBeam is the agent security platform from the explainx.ai team for stopping AI agents before they take dangerous actions, which matters most when systems are degraded and tool outputs are unreliable.
What to watch next
- Google's incident report. Google usually publishes a root-cause analysis for zone-level incidents; the March 2025 full report came about two weeks after the incident. Look for the power chain, the failure of redundancy, and the remediation list.
- Revolut's post-mortem or regulator questions. UK financial regulators have been pressing banks on operational resilience and critical third parties; a multi-hour payments outage tied to one cloud zone will draw attention.
- Whether the other banks name a cause. HSBC, Lloyds and Barclays have not, in anything we found, tied their issues to Google.
We will update this post as Google and the affected companies publish more. Related background: the neocloud shakeout on how fragile AI capacity can be, and the ChatGPT Work outage on August 31.
Questions readers are asking
Is my money safe? Revolut says all customers' money and data are safe. That is the company's statement; there is no report of loss.
Will Google compensate customers? Google Cloud's service level agreements typically offer service credits for missed uptime targets, but a business's own contract terms decide. We did not find any announcement from Google about credits.
Is the cloud less reliable than data centers you run yourself? Not necessarily. A hyperscaler zone usually has better power and cooling engineering than a small company can afford, but when it fails, thousands of customers fail at once and the failure is out of your control. The skill is designing for it, not avoiding the cloud.
Details reflect reporting and statements available on October 20, 2026 and may change as Google publishes its incident analysis.
Related reading
- Claude outage and network failures: recovery in July
- ChatGPT and Codex outage, paid limits reset
- ChatGPT Work outage on August 31
- The AI data center backlash map
- Hyperscaler nuclear deals for AI data centers
- Virginia data center power tax and AI pricing
- Sources: CNN/Reuters, Irish Times, Google Cloud Service Health