Your uptime does not depend on your network. It depends on whose network your provider bought transit from.
August 21, 2026: Arelion — a global Tier 1 backbone — went down for 1 hour 21 minutes across a 3 hour 10 minute window, disrupting customers in more than 30 countries. Four days earlier, Cox took 24 minutes of outage across seven countries. Both started in Chicago.
WHAT ACTUALLY HAPPENED
The Arelion event began on Chicago nodes around 9:25 PM EDT. Five minutes later Atlanta joined. It cleared, then came back on Chicago. Twenty-five minutes on, Seattle. The final wave hit Chicago, Dallas, San Jose, plus nodes in Sweden and the U.K. before resolving near 12:35 AM EDT.
Impact reached the U.S., France, Singapore, India, Germany, Japan, Brazil, Australia, Egypt and two dozen more.
The Cox outage on August 17 centred on Chicago too, and still touched customers in the U.S., Mexico, U.K., Austria, Canada, Germany and Denmark.
Background rate for the period: 546 global outage events in the week of August 10-16, 534 the following week. This is not exceptional. It is Tuesday.
WHY MULTI-CLOUD DOES NOT SAVE YOU HERE
Two ISPs, two clouds, two regions — and one shared transit provider underneath. Redundancy on paper, single point of failure in the routing table.
Look at the pattern above: a fault on a handful of Chicago nodes propagated to 30+ countries in minutes, then flapped for three hours. That is what a backbone incident looks like from the outside — intermittent, geographically incoherent, and invisible to every monitoring tool that only watches your own equipment.
Your teams will spend the first hour blaming the firewall. Meanwhile the failure is four AS hops away, in a network you have no contract with and no visibility into.
Resilience is not counting your circuits. It is knowing whether they diverge before they reach the core of the internet.
WHAT TO DO NOW
MAP YOUR UPSTREAM AS PATHS — trace which ASNs your links actually traverse, because two providers frequently resolve to one backbone.
DEMAND TRANSIT DIVERSITY IN CONTRACTS — specify distinct upstream carriers, not just distinct last-mile vendors.
DEPLOY EXTERNAL PATH MONITORING — internal SNMP tells you your router is healthy while the internet between you and your users is not.
REHEARSE THE FLAPPING SCENARIO — intermittent, partial, multi-region degradation is far harder to run than a clean failover, so practise that one.
You can only engineer around the dependencies you know you have.
Do you know which Tier 1 networks your business actually rides on today?