Gondwana-1

Based on 23 RIPE Atlas measurements from GeoCables monitoring infrastructure, March–April 2026.

Gondwana-1 is the submarine cable that connects New Caledonia to Australia. It came into service in mid-2008, runs 2,151 km from landings on the New Caledonian east coast (Poindimie on Grande Terre, with branches to Mouly on Ouvéa and Xepenehe on Lifou) to Sydney, and is now in its eighteenth year of operation. For most of its existence it has been the only commercial submarine cable serving the territory of approximately 270,000 people. Every video call, every web page, every cloud-stored file accessed by a New Caledonian household, business, or government office crosses this single fibre on its way to and from the rest of the world.

Across 23 measurements between Noumea and Sydney, the minimum round-trip we observe is 22.76 ms, the maximum is 22.93 ms, and the average is 22.85 ms. The spread between the slowest and fastest packet across our entire measurement window is 0.17 ms — one hundred and seventy microseconds. The standard deviation is 0.05 ms.

The 50-microsecond standard deviation

That number deserves its own paragraph. Of the cables we have published so far — fifty-nine, spanning every major ocean basin and every architectural style from 1999-vintage rings to 2025 hyperscaler trunks — Gondwana-1's standard deviation is among the lowest we have ever recorded on an intercontinental path. For comparison, the quietest direction we measured on ARCOS-1, the Caribbean ring system we published earlier today, came in at 0.18 ms — itself extraordinary. Gondwana-1 is more than three times tighter than that. Most modern long-haul cables, even those carrying hyperscaler traffic on dedicated wavelengths, settle to standard deviations in the 1–10 ms range after BGP convergence and routing-policy stabilisation. Single-digit milliseconds are good. Sub-millisecond is rare. Fifty microseconds is the kind of variance you get when you measure the same physical signal twice with a slightly imperfect clock.

That last analogy is, in fact, almost literally what is happening. The 0.05 ms standard deviation is approaching the noise floor of the measurement process itself — the resolution of system clocks on the probe and the response endpoint, the jitter introduced by the Linux kernel scheduling the user-space ping process, the small variations in the time the network card takes to push the response packet onto the wire. When the measured network path itself contributes essentially zero variance, what you are looking at in the standard deviation column is the instrument, not the network.

The physical floor at 1.081×

The theoretical physics floor for the 2,151 km cable, accounting for the optical refractive index of single-mode fibre and a typical 10–15% non-straight cable-route allowance, lands at 21.05 ms. Our measured minimum of 22.76 ms is 1.081× that floor — a margin of just 1.71 ms above the absolute fastest a packet could travel between Noumea and Sydney through this fibre. That margin of less than two milliseconds is consumed by the protocol overheads we cannot eliminate from RTT measurements: the latency added by ping packet construction, the response generation at the target, and the round trip through routing equipment at each end of the cable. There is essentially no room left in the latency budget for routing inefficiency, BGP detours, or alternate-path selection.

The reason is simple. There is no alternate path. From Noumea, when a packet is destined for Sydney or anywhere reachable through Sydney's peering exchanges (which is, effectively, the rest of the global internet for New Caledonia), there is exactly one cable available, exactly one upstream ISP arrangement at the Australian end, and effectively one BGP-announced route at any given moment. The packet leaves Noumea, traverses Gondwana-1's 2,151 km of fibre, terminates at Sydney, and is handed to the Australian internet from there. Reverse-direction packets follow the symmetric path back. There is no decision to make, no preference to set, no alternative to consider. Routing, in this context, is not a choice — it is a single path, and the latency we measure is the cable.

What total dependency looks like in numbers

The 0.05 ms standard deviation and the 1.081× floor multiplier are different ways of saying the same thing: New Caledonia's internet has no redundancy at the international level. When most countries' submarine measurements show variance, it is because their international connectivity has been engineered for resilience — multiple cables, multiple peering points, multiple commercial alternatives. Each of those alternatives is slightly different in physical path length, in equipment latency, in operator policy, and operators select among them based on cost, reliability, and current load. The variance we measure is the visible signature of that engineered diversity. New Caledonia does not have it.

This is not a unique situation among small Pacific island states. Tonga Cable, the cable that connects the Kingdom of Tonga to Fiji and from there to the wider internet, has a similar single-cable architecture — and the consequences of that became very visible when the Hunga Tonga–Hunga Haʻapai eruption of January 2022 severed it and left the country mostly cut off for several weeks. Coral Sea Cable System, the 2020-vintage cable serving the Solomon Islands and Papua New Guinea via a Sydney landing, was specifically built to address the same kind of vulnerability for those neighbouring island states. Bulikula, the 2026 hub-and-spoke system being built across Polynesia and Micronesia, is the latest answer to the question of how to give Pacific islands meaningful redundancy without building dozens of expensive low-traffic cables.

Eighteen years of being the only thread

Gondwana-1 is mid-life by the design standards of its generation. It will need replacement or major upgrade by approximately 2030–2032 to remain commercially viable for the modern bandwidth-hungry application stack. The 2008 vintage means original transponder technology in the 100 Gbps-per-wavelength range; modern coherent optics deliver several times that capacity per wavelength using the same fibre, so an upgrade in the form of new terminal equipment can extend the cable's useful capacity life without re-laying fibre. New Caledonia's network operator OPT (Office des Postes et Télécommunications) has discussed replacement and redundancy options publicly for several years, and a Pacific Connect-style additional cable extending from Vanuatu through New Caledonia and onward — providing a second international route entirely independent of the Gondwana-1 path — has been in feasibility studies. As of our measurements, no such alternative is yet in service.

The closest comparable systems we have published, in terms of single-thread small-economy dependency, are Tonga Cable (2013, 826 km, ~100,000 people primarily depend on it) and Coral Sea Cable System (2020, 4,700 km, several Pacific economies share it). Of those, Tonga Cable's single-direction RTT measurements and standard deviations are the structurally similar reference: a Pacific island country, one cable, no alternatives, and consequently a measurement profile that is almost too clean.

What we watch

Two things. First, the standard deviation. As long as it stays at 0.05 ms — as long as our 23-measurement window keeps spreading over 0.17 ms total — Gondwana-1 remains the only commercial route between Noumea and Sydney, and we are looking at the cable directly. The first sign that an alternative path has come into service will be a sudden jump in standard deviation as packets begin to be split between two physically distinct routes with slightly different lengths. If we ever see Noumea→Sydney standard deviation step up to even 0.5 ms, that's the news: redundancy has arrived in the New Caledonian internet.

Second, the floor multiplier. If a future cable lands at Noumea on a shorter path to Sydney — say, a direct great-circle route that shaves 100 km off Gondwana-1's 2,151 km — our measured minimum could drop below 22.76 ms, and we would see the floor multiplier fall toward 1.0× on the new cable while staying at 1.081× on Gondwana-1. That kind of split between primary and backup cable is the standard mid-life transition for a long-running submarine system, and Gondwana-1 will reach that point in the next handful of years. For now, eighteen years in, it remains the only thread, and the 0.05 ms standard deviation is the proof.