Your "Redundant" Connectivity Probably Shares a Duct

The internet is drawn as a mesh and built as a small number of very large pipes.

Between 150 and 200 submarine cable faults are reported worldwide each year, and 70 to 80% of them come from ordinary accidents, commercial fishing gear and ships' anchors rather than sabotage. 1 Most never become public incidents, because there was enough alternative capacity to absorb the loss. The ones that do are the ones where there wasn't.

That distinction, absorbed or not absorbed, is decided long before the cable breaks, and it's the whole of what redundancy is supposed to buy.

Four cables, one anchor

The Red Sea remains the clearest demonstration. In September 2025, multiple systems were damaged in a single event, including SEA-ME-WE-4, IMEWE, FALCON GCX and Europe India Gateway, all of which normally carry traffic between Europe, the Middle East and Asia. The result was wholesale rerouting, with Microsoft forced to move Azure traffic onto alternative paths. Early analysis pointed at commercial shipping, most likely a vessel dragging its anchor across the cables. 2

Four independent systems, one anchor. They were independent in ownership, in contract and on the procurement spreadsheet. They weren't independent in geography, and geography is the only kind that matters to an anchor.

Questions that test diversity properly

"Do you have redundant connectivity?" is close to a useless question. These are better.

Diverse from whom? Two upstreams who both buy wholesale from the same carrier are one upstream with two invoices.

Diverse where? Two paths that converge on the same landing station, the same duct, or the same building aren't diverse at the point of failure. Rights of way are scarce, so different carriers routinely means the same trench for forty kilometres.

Diverse in direction? For a European site, capacity that all exits northwards is one bad quarter in the North Sea away from being no capacity at all.

Diverse in jurisdiction? A cable doesn't have to break for a path to close. Shutdowns, sanctions and routing policy close paths too, and none of them need a repair ship.

The honest test is a single sentence. What's the smallest physical event that removes all of my paths at once? If you can't answer it, you don't know whether you're redundant. You know that you bought two of something.

Two connectivity contexts, not four dots

AVS ISP runs Tirana, Skopje, Amsterdam and Dublin. Those aren't four dots chosen for a coverage graphic. They're two different connectivity contexts.

Amsterdam and Dublin sit on the densest interconnection fabric in Europe, with enormous capacity, enormous choice of transit, and a very large number of other people's eggs in the same basket.

Tirana and Skopje sit on the Southeastern approach. Albania in particular is reached by subsea routes across the Adriatic into Italy and terrestrial paths through Greece rather than the northern trunk. When the northern corridor has a bad day, that's a different failure domain rather than a different rack in the same one.

For customers running their own routing, AS210464 lets you act on this directly rather than hoping we notice. Our BGP community set includes per-region and per-peer-type announcement control, prepending from one to three times, local-preference control for backup routes, and remote-triggered blackholing. You can pull your prefix out of a region yourself, in seconds, from your own router. Sessions start at €5/mo, and every VM comes with a routed /48 of IPv6.

You can also check our claims before spending anything. The looking glass runs live ping, traceroute and BGP route lookups from all four sites with no account. Run a traceroute to Amsterdam and one to Tirana from the same place. If the two paths overlap for most of their length, you've just learned something useful, and it's about your current setup rather than ours.

Nobody is immune

Cables get cut by anchors, fishing gear, dredgers, earthquakes and occasionally on purpose, and no amount of application architecture makes glass on the seabed more robust.

What you can control is whether a cut is a degradation or an outage. That isn't decided during the incident. It's decided months earlier, by whether your copies sit on paths that fail together, and by whether anyone ever checked rather than assuming that two contracts meant two routes.

Sources

1. Media Enquiries & Frequently Asked Questions, International Cable Protection Committee
2. Diving Into the Red Sea Cable Cuts & More Outage News, ThousandEyes