COBRAcable

COBRAcable is one of the more unusual entries in our monitoring set. Most submarine cables we track were built to move data — Google and Meta and Microsoft paid billions for glass fiber under the ocean so that web requests could bounce between continents a few milliseconds faster. COBRAcable was built to move electricity. Specifically, 700 megawatts of it, across 304 kilometers of North Sea seabed, between the Dutch coast at Eemshaven and the Jutland shore near Endrup, Denmark. The fiber we ping is a telecom byproduct — spare glass bundled into a high-voltage direct-current power cable because, having laid the trench anyway, you might as well include some communications capacity.

The cable was commissioned in September 2019 as a joint project between TenneT, the Dutch transmission system operator, and Energinet, its Danish counterpart. It was the first direct electrical interconnection between the Netherlands and Denmark, and it is part of a broader push to stitch together the national power grids of the countries around the North Sea into a single high-voltage mesh. Relined, a fully-owned subsidiary of TenneT, markets the spare fiber capacity commercially — which is why COBRAcable shows up in public submarine cable databases alongside "real" telecom cables built for the purpose.

Why a power cable ends up in our measurements

The short answer is that the physical cable is indistinguishable, from an optical fiber point of view, from any other submarine fiber. Once the glass is laid, light travels through it the same way whether the copper next to it is carrying 700 megawatts or nothing at all. Operators like TenneT historically need fiber alongside their long-distance transmission lines for control signaling, protective relaying, and remote operation of the converter stations at each end. Reselling the leftover capacity turns a cost center into a revenue line, and Relined has been doing exactly that along TenneT's Dutch rights-of-way for over a decade.

What makes this especially interesting for anyone who cares about the European internet backbone is that the Netherlands–Denmark fiber path was, until COBRAcable, almost always routed through Germany. Most traffic between Amsterdam and Copenhagen takes a terrestrial path via Hamburg. COBRAcable opened up a direct submarine option — shorter in geography, technically independent of the German backbone, and operated by a pair of national grid operators rather than by a traditional telco consortium.

Our measurements

GeoCables monitors COBRAcable through RIPE Atlas measurements using probes hosted in and near the cable's landing areas. Over the last 30 days we collected 97 cable health checks, of which 78 returned usable ping RTT values. The remainder were traceroutes without a successful terminal ping, or scheduled checks where the probe was temporarily offline.

Those two rows look like they are measuring two different cables. They are not. They are measuring the same 304 kilometers of fiber from two different places in the Netherlands, to the same Danish target IP. The difference — consistently 25 to 30 milliseconds on a good day — comes almost entirely from how each probe's upstream network has chosen to reach Denmark.

Probe A appears to be peered in a way that lets it reach Denmark over a short path, probably via a direct session at AMS-IX or a regional IXP in the north of the Netherlands. Probe B's upstream provider hands traffic off over a longer route, plausibly transiting through Frankfurt or another European hub before finally arriving at the Danish endpoint. We are not measuring COBRAcable in isolation — we are measuring COBRAcable plus several hundred kilometers of peering and transit on either side, and the BGP decisions that determine which hops a packet actually visits.

This split view is one of the most useful things about observing a cable from more than one probe. It is a reminder that latency numbers are always a product of the physical path and the routing layer, and that for a short cable like COBRAcable the routing layer completely dominates.

Theory vs reality: where do those milliseconds go?

The fiber path between Eemshaven and Endrup is about 304 kilometers. Light in glass travels at approximately 200,000 km/s. The theoretical minimum round-trip time for a packet traversing the pure fiber is therefore:

(304 km × 2) ÷ 200,000 km/s ≈ 3 ms

Three milliseconds. That is all the actual submarine cable contributes to the total RTT. Our cleanest observed value is 13.8 ms, and our 30-day average on the longer path is closer to 48 ms. In other words, on a short cable like this one, the fiber itself accounts for less than a quarter of the best-case measurement, and less than ten percent of the typical measurement. Almost everything we observe is overhead happening on dry land — in routers, switches, peering sessions, and the converter stations where the optical path meets the IP layer.

This is the opposite of the Equiano situation, where an 80-millisecond fiber floor dominates a 205-millisecond total. On long transoceanic cables, physics wins. On short regional cables like COBRAcable, routing wins, and the cable itself is more or less a constant tucked inside a much larger variable.

The April 2026 latency cluster

Between April 5 and April 7, 2026, our anomaly detector fired four alerts on the COBRAcable corridor, all on the same probe and all pointing at the same Danish target endpoint. Three were classified as warning and one as critical. The critical alert coincided with a brief window in which our ping RTT touched 871 milliseconds, an order of magnitude above the 30-day baseline on that path.

We do not know what caused the cluster. The measurements resumed normal behavior within hours, and subsequent checks on April 8 and later showed latency back in the 44–50 ms range. A single-probe event on a single target IP is not enough to conclude that the cable itself was degraded — it could equally be a transient congestion event on the upstream transit network, a router reconvergence after a BGP session flap, or maintenance on a peering port. What we can say is that the event was visible, was flagged automatically, and resolved without operator intervention. That is what the monitoring system is supposed to do.

What we are watching

• Second-vantage confirmation. The 14 ms versus 45 ms gap between our two NL probes is large enough that we want more observations from each before we call it stable. A probe physically closer to the Eemshaven landing station would help us pin down how much of the remaining latency is cable-adjacent and how much is long-haul transit.
• Seasonal patterns. COBRAcable is a power cable first. Maintenance windows on the HVDC converter stations at either end can in principle affect the fiber pair if the fiber shares conduit or equipment rooms. Correlating any future outages with publicly-announced TenneT or Energinet grid works would be an interesting exercise.
• Traffic symmetry. Most of our measurements so far are launched from the Dutch side. We would like to get comparable traffic launched from a probe inside Denmark toward a Dutch target, to check whether the routing asymmetry we observe on Equiano also appears on this much shorter path.

Design notes

COBRAcable is a bipolar high-voltage direct-current link operating at approximately ±320 kV, with a nameplate transmission capacity of 700 megawatts in either direction. The two converter stations — one at Eemshaven in Groningen province, the other at Endrup in western Jutland — translate between the HVDC link and the respective national 400 kV alternating-current grids. A core commercial driver of the project is the ability to send Danish wind surplus to the Dutch market when North Sea winds are favorable, and to run the link in reverse when Dutch solar or gas generation is cheaper than the marginal Danish unit.

The fiber that we monitor sits inside the cable sheath alongside the HVDC conductors. Relined manages it as a dark-fiber product, typically leasing strands to operators who want a short, physically diverse path between the Netherlands and continental Europe's northernmost peninsula. That makes COBRAcable, in our database, simultaneously a submarine cable, a power interconnector, a climate-policy artifact, and a piece of commercial telecom infrastructure — all 304 kilometers of it.