Data center outages reveal hidden web of dependencies
The gist
When the power goes out, it’s not just the servers that crash—hidden dependencies can ripple across banks, airlines, and even the courts.
What to know
- UK blackouts and a Google Cloud London outage in late 2026 showed that data center failures often start with external utility breakdowns, not just on-site glitches.
- Operators are now mapping every critical dependency—utilities, telecoms, cloud, and fuel—and testing end-to-end recovery, after silent failures and delayed telemetry exposed hidden weak links.
- Recent outages hit Revolut, HSBC, Royal Mail, and even grounded over 1,300 US flights, proving that one facility’s hiccup can escalate into nationwide chaos.
Beyond the Power Switch
Restoring electricity is just the start—modern blackouts trigger multi-day failures across payments, telecoms, and logistics, exposing how deeply everyday services depend on fragile external networks.
November 2026 sharpened a lesson the sector had been inching toward for months: data center business interruption often begins outside the building. Insurance Business captured that shift directly in October with the warning that “data center business interruption exposure starts outside the building,” and the UK blackout planning assumptions made the point concrete by showing how loss of utility power quickly becomes a wider failure of telecoms, water, fuel, transport and financial services rather than a single-site electrical problem.
The practical consequence is that restoration of a facility’s own power no longer marks the end of interruption risk. As ADS Advance noted, “That means a business with power restored after 48 hours could still face problems with payments, communications or deliveries,” and gave the example that “A shop may have lighting but still be unable to take card payments”; the same planning scenario estimated that “around 60% of electricity demand could be restored within two days,” with “planned disconnections on days five and six” for some customers, and “Full, stable restoration could take more than seven days.”
Resilience Starts with Dependencies
Operators now dissect every external link—utilities, cloud, fuel—to anticipate and contain outages, shifting from hardware checklists to dynamic, scenario-driven risk maps that drive real-world recovery plans.
Operators are rewriting resilience assessments around dependency maps rather than equipment lists, extending analysis beyond the facility to utilities, telecoms, cloud providers, fuel suppliers and other third parties, as TechTarget argues. That broader lens changes the recovery question too: Michael Campbell of Fusion Risk Management says leaders should, for each critical service, “identify its dependencies, tolerance for disruption and the minimum conditions for safe recovery,” a discipline that turns third-party exposure into concrete contingency planning instead of assuming failover will work when shared external links are stressed.
That expanded assessment is driving more scenario-based testing and earlier intervention, because external shocks now feed directly into continuity planning rather than sitting outside it as background noise. Supply Chain Dive notes that “The WTO’s October 2026 trade outlook identifies higher energy prices and disruption to transport routes from the Middle East conflict as pressures on global trade,” while “UN Trade and Development has also highlighted” how shipping disruption ripples through energy, freight, insurance and input costs; in response, “Crisis24 AiiA delivers anticipatory intelligence tailored to an organization’s business priorities… Earlier understanding gives organizations more time” to evaluate alternatives, qualify backup sources and adjust operations before failures cascade.
Full-Stack Monitoring Prevents Blindspots
Integrated telemetry and rigorous failover drills are exposing silent, cascading failures before they hit production, forcing organizations to confront and fix hidden weak spots across their digital infrastructure.
The practical shift in resilience is that monitoring now has to see across the full stack and prove redundancy paths before production depends on them. VP IT’s multicloud guide argues that teams should map application tiers to availability tiers, then “test test” with failover and failback, using end-to-end instrumentation and automation from application through infrastructure so hidden dependency gaps are exposed in rehearsal rather than during an outage, while explicit RPO and budget choices define which weak links matter most and where added redundancy is justified.
That only works if telemetry is smart enough to catch subtle failure modes and late signals that ordinary dashboards miss. In the TimescaleDB case study, engineers deliberately modeled “Every device at one specific site goes completely silent,” “A counter reset… its lifetime energy WH total drops back to zero,” and “a stuck state… report that it's broken every 10 seconds for an entire day, but it never goes offline”; they also note, “We tell time scale to update this every 5 minutes. But notice the end offset of 2 hours,” so delayed data does not mask weak links before they cascade into service loss.
Single Faults, Nationwide Fallout
A lone data center glitch can cripple payments, deliveries, and air travel for days, with ripple effects that outlast downtime and erode public trust—even in the absence of cyberattacks.
The October failures made the mechanism of interruption visible: on 20 October, “a power event at the datacentre hosting Google Cloud’s europe-west2-b zone took that zone offline,” and Revolut, HSBC, DPD and Royal Mail were among the organisations reported to have experienced disruption, while the video platform for remote court hearings in England and Wales also failed. Computer Weekly said Google’s own report showed the blast radius extended beyond the darkened zone, with “secondary failover and queuing delays across infrastructure that had not lost power,” turning a single external power event into payments, deliveries and justice-system disruption.
Capacity sharpened the point by focusing on spread rather than source: Google’s incident report said the europe-west2-b outage caused around six hours of disruption, and Carvalho argued that “how far the problem travelled” mattered because “One fault in one facility, and a good part of the UK’s digital morning stopped.” The same pattern appeared in November outside the UK: on November 12, Microsoft 365 services including Outlook, Teams and Copilot went down for users in 74 countries after a hidden network-capacity problem surfaced, and on November 13 the FAA “ordered a nationwide ground stop after radar and communications instability spread across its air route traffic control centers,” halting departures for “more than four hours,” with “More than 1,300 flights were canceled and more than 10,000 delayed,” underscoring that dependency failures—not cyberattacks—can outlast downtime and widen into economy-scale interruption. “None of the disruptions has been linked to a cyberattack,” yet “the costs also outlast the downtime: a four-hour ground stop became three days of cancellations,” and “every outage erodes public trust while giving opportunistic attackers a ready-made smokescreen.”



