Why One London Power Fault Broke Europe Cloud Workloads

Why One London Power Fault Broke Europe Cloud Workloads

Picture tapping your card for morning coffee in London or Madrid and getting a decline. Not because your account is empty, and not because a hacker breached the bank, but because a wire in a data center tripped and nobody told the backup generators to kick on.

That's basically what happened when Google Cloud disruptions in Europe hit the europe-west2 region on October 20, 2026. At 06:53 BST (22:53 US/Pacific on October 19), an external underground cable fault took out utility feed A serving the facility behind europe-west2-b. Backup feed B was already physically opened and key-interlocked for repairs after issues surfaced during a quarterly generator load test on October 14. Within minutes, uninterruptible power supply batteries drained, servers lost power, and ripple effects spread across apps, payment rails, and public services.

Anyone who has run production infrastructure at scale knows how deceptive cloud architecture can look on a whiteboard. Three availability zones per region sounds comforting. Until a single physical fault triggers a cascade that exposes every assumption you skipped testing.

What Actually Broke in the Europe Cloud Outage

Most coverage of Google Cloud disruptions in Europe treats the event like a black box. Google's status page initially noted multiple services experiencing issues and intermittent network failures across europe-west2, with recovery taking roughly six hours from the onset of customer impact to broad workload stabilization.

Here's the twist that makes this incident worth studying closely: the generator failure wasn't a dead engine or an empty fuel tank. According to Google's incident reporting and postmortem analyses of the event, programmable logic controllers (PLCs) responsible for starting the generators checked for voltage on the feeders rather than verifying whether the feeder breakers were closed. Because their sensors sat upstream of the locked-out feed B breaker—where power was still live—both primary and backup PLCs read a healthy utility signal and never issued a start command.

UPS batteries only buy a short window—around five minutes in this case—to bridge the gap while generators start, synchronize, and take load. Once those batteries ran out at 23:00 US/Pacific (07:00 BST), hosts, storage, and network hardware in europe-west2-b went dark.

How a Single-Zone Outage Spills Across a Region

If you're an engineer or IT leader, the obvious question is why a zone-level power loss caused pain across the wider europe-west2 region.

In theory, other zones in London absorb failed traffic cleanly. In practice, three forces push back hard during an abrupt failure:

  • Traffic surges and admission queues: When workloads fail over or retry aggressively into surviving capacity, compute slots, database failovers, and internal queues fill up fast, creating elevated error rates and latency spikes for neighboring services.
  • Cross-zone control-plane dependencies: Even when your compute sits in another zone, synchronous checks, shared metadata paths, or control-plane operations can still touch impaired components.
  • Delayed visibility: Status pages lag behind live engineering triage. When incident updates take hours to name the exact failing zone, teams making split-second migration calls are flying partially blind.

Banks, Payments, and Public Services Felt the Shockwave

Cloud outages rarely stay confined to developer dashboards. Consumers feel them at checkout counters, ATMs, and service desks.

During the October 20 disruption, Revolut saw severe knock-on trouble affecting card payments, top-ups, logins, and transfers across Europe, with dozens of services flagged as partially unavailable on its status page before full restoration later that day. News outlets including The Guardian also reported banking disruptions involving institutions such as Lloyds and HSBC alongside delays affecting UK courts and public-facing services.

Revolut's own technical overview of the October 20 outage offers a remarkably candid look at what happens even when an engineering team thinks they've planned for zone failures. Revolut maintains replicas across three London zones plus another region for broader regional recovery. Yet replication lag left some PostgreSQL clusters read-only after quorum recovery, requiring engineers to verify replica states and chart a safe recovery path. Worse still, their primary instance for visualizing database metrics lived in Zone B—meaning engineers had to rely on logs, command-line tools, and ad hoc monitoring instances spun up in surviving zones before full remediation could move ahead.

That detail hits home for anyone who has ever fought a fire while the smoke alarm melted. Monitoring tools that share fate with the failing zone create blind spots at the exact moment clarity matters most.

Regulation Is Catching Up to Concentration Risk

Incidents like this aren't just operational headaches anymore; they're regulatory flashpoints. Across Europe, financial authorities have been tightening scrutiny on concentration risk—the danger that dozens of banks and payment processors all depend on the same cloud providers or facilities. As The Paypers reported, EU regulators designated Google Cloud among 19 critical third-party computing providers supporting the bloc's financial sector in November 2025.

Under frameworks like the EU's Digital Operational Resilience Act (DORA) and UK Prudential Regulation Authority (PRA) and Financial Conduct Authority (FCA) expectations, financial firms are expected to model "severe but plausible" disruptions and prove their recovery paths actually work. A major cloud zone going down isn't a hypothetical edge case anymore. It's a test of operational maturity.

Why "Just Move to Another Region" Sounds Easier Than It Is

Whenever a regional outage hits, standard advice quickly pops up in status updates and comment threads: migrate workloads to another zone or region.

Honestly, that advice is half-true and half-fantasy if you haven't built for it ahead of time. Re-architecting stateful databases, cross-region data replication, latency budgets, and traffic routing in the middle of an active incident is like changing a tire while doing 70 mph on the motorway.

There's also a real debate among infrastructure engineers about active-active multi-region setups and multi-cloud architectures. Some argue critical payments and authentication paths demand active cross-region or multi-cloud resilience regardless of cost. Others push back, pointing out that multi-cloud adds massive operational complexity, subtle consistency bugs, and higher engineering overhead—sometimes creating more human-induced outages than it prevents. Neither side has a monopoly on common sense; the right trade-off depends on your recovery time objectives, regulatory obligations, and budget.

What's hard to argue against is this: untested failover isn't resilience. It's hope.

Practical Moves to Harden Your Cloud Setup Now

You don't need a massive rewrite overnight to reduce your exposure to Google Cloud disruptions in Europe or similar regional hiccups. Focus first on the bottlenecks that actually slow teams down when infrastructure wobbles.

  • Audit single-zone dependencies first: Check where zonal resources, caches, batch jobs, or single-zone control planes quietly anchor your stack.
  • Decouple observability from primary zones: Keep logging, metrics dashboards, and alerting pipelines isolated in separate zones or regions so your eyes stay open when a zone goes dark.
  • Watch database replication lag and failover paths: Verify how databases behave under load spikes and partial network or power degradation, not just during clean synthetic tests.
  • Run real failover drills: Practice promoting replicas, shifting traffic, and running degraded-mode flows (like stand-in or queued processing) before an outage forces you to improvise.
  • Target high-value journeys first: Prioritize checkout, login, authorization, and core customer support for multi-zone or cross-region resilience rather than over-engineering every internal tool.

Start this week by picking your single most business-critical workflow, tracing every zone and external service it touches, and running a live game-day drill to see where your monitoring and failover assumptions break.

MR

Miguel Rodriguez

Drawing on years of industry experience, Miguel Rodriguez provides thoughtful commentary and well-sourced reporting on the issues that shape our world.