MAREA

MAREA is the submarine cable you have probably heard of without realizing it. When commissioned in February 2018, it was the highest-capacity trans-Atlantic cable in service and the first major trans-Atlantic system owned by a partnership of cloud companies rather than by a traditional telco consortium. Microsoft and Meta (then still Facebook) put up the capital; Telxius, the infrastructure arm of Telefónica, handled construction and operates the cable. It runs roughly 6,605 kilometers between Bilbao on Spain's Bay of Biscay coast and Virginia Beach on the US mid-Atlantic shore — deliberately avoiding the congested Lisbon–New York corridor that carried most earlier trans-Atlantic traffic.

The name is worth a moment. "Marea" is Spanish for "tide." The cable earned its name during the laying phase in 2017, when Hurricane Harvey passed near the installation vessel. The work continued. It is a slightly unusual thing for industrial infrastructure to be named after a natural force it had to survive, and the choice signaled something about what this cable was supposed to mean: not a quiet backstop, but a marquee project by its owners.

Why a cable to Virginia Beach matters

Before MAREA landed, almost all hyperscaler trans-Atlantic traffic concentrated on New York and New Jersey landing points. That made the metropolitan fiber trenches underneath the NYC region some of the most heavily shared cable real estate on Earth — every new submarine system was fighting for rights-of-way with legacy telegraph cables, electrical infrastructure, dredging zones, and commercial shipping anchor fields. MAREA pushed the primary hyperscaler trans-Atlantic entry point roughly 350 kilometers south, to a town on the Virginia coast that had almost no existing submarine cable infrastructure when construction began.

Two reasons drove the Virginia Beach choice. First, the undersea cable density offshore is far lower than around New York, which made the engineering easier and the trench maintenance contracts cheaper. Second, the Virginia shelf has a more predictable hurricane shelter profile than the Long Island approach, and the inland backhaul from the landing station to the Ashburn data center cluster runs through less-congested terrestrial fiber corridors than the equivalent path from Long Island to northern New Jersey.

The downstream effect is bigger than the cable itself. In the years after MAREA, Microsoft and Amazon built massive hyperscale data center complexes in Virginia Beach and the surrounding counties, specifically because MAREA's landing station was the fastest and lowest-latency way to reach Europe from that part of the US. The cable didn't just carry traffic — it reshaped the economic geography of the North American data center market. It is one of the few submarine cables about which you can reasonably say that it pulled an entire hyperscale cluster into existence.

Our measurements: a detective story

The honest version of what happened when we first pulled up MAREA's raw measurements is that the numbers looked suspiciously good. An average RTT of 82 milliseconds for a 6,600 km trans-Atlantic cable would have put MAREA almost at the physical floor of what light in glass can achieve — a multiplier of only 1.2× theoretical — which is dramatically better than any other long-haul cable in our database. Something was off.

We sorted our 90 most-recent cable health checks for MAREA by target IP address and found this:

Fifty-three of our ninety measurements were hitting a.root-servers.net, which is an anycast address operated by Verisign. Anycast means the IP is advertised from many physical locations simultaneously, and routers deliver your packet to whichever advertiser is topologically nearest. For a probe in northern Spain, the nearest a-root instance lives in Madrid or London or Frankfurt. Our probes were never actually crossing the Atlantic for those measurements — they were pinging a European DNS mirror that happened to share the same globally-routable IP address as a server in Virginia.

Another 21 measurements were launched at verified non-anycast targets that simply never returned a ping response inside our collection window, leaving us with 16 usable trans-Atlantic samples: 15 from a Cox Communications endpoint in Virginia, and one from a secondary US transit provider. That is a small sample. It is also a clean one.

The cleaned measurements tell a very consistent story:

An 11-sample standard deviation under one millisecond is about as stable as a measured internet RTT ever gets. The underlying fiber is doing its job, the BGP path is not flapping, there is no congestion on the intermediate transit networks, and no router along the way is queueing packets for meaningful amounts of time. The cable delivers a trans-Atlantic round trip that is boring in the best possible way.

We discovered the anycast contamination while researching this article and updated our aggregation pipeline the same day. The RTT numbers you see on the MAREA cable page — and on every other cable page across GeoCables — now exclude anycast targets from averages and from the health dashboards. The change also improved the displayed averages for several other trans-Atlantic and trans-Pacific cables where anycast targets had been pulling the reported latency downward. The methodology is now honest about what it is measuring.

Theory vs reality: best-in-class trans-Atlantic

The Bilbao–Virginia Beach great-circle distance is about 6,605 kilometers. Light in glass travels at roughly 200,000 km/s, so the theoretical minimum round-trip time on a perfect straight-line fiber is:

(6,605 km × 2) ÷ 200,000 km/s ≈ 66 ms

Our measured average is 129 ms. The ratio of measured to theoretical is therefore about 1.95× — significantly tighter than the 2.5× we see on Equiano's longer Sesimbra–Cape Town corridor, and dramatically tighter than the 3× to 15× we see on short regional cables like COBRAcable where routing overhead completely dominates. For a trans-oceanic path, this is about as close to the physical floor as you can reasonably get.

There are several specific reasons MAREA runs tight:

• Disaggregated optics. MAREA was one of the first production trans-Atlantic systems designed as an open line system, meaning the cable supports transponders from multiple vendors rather than being locked to a single supplier's optical equipment. This matters for latency because modern coherent DSPs can be upgraded independently of the wet plant, and the owners keep them current.
• Short terrestrial backhaul. The Bilbao landing station has direct fiber backhaul to inland Spanish internet exchanges without a long terrestrial detour. On the US side, Virginia Beach now hosts multiple neutral interconnect facilities that put MAREA's US termination within tens of microseconds of the major peering networks. Both endpoints minimize the land-side overhead that inflates RTT on older cables.
• No branching units on the main trunk. MAREA is a clean point-to-point link with no mid-cable drops. Branching units introduce small amounts of additional latency and more importantly can create BGP path ambiguity when traffic is diverted to a spur. The clean trunk topology means that every packet traverses the same physical path every time.
• Mature transit relationships. By 2026 the networks on either end have had almost eight years to establish short-path peering arrangements specifically for MAREA traffic. The earliest trans-Atlantic cables typically took years to settle into their optimal BGP paths.

What we are watching

• More vantages from the Americas side. All of our current usable MAREA measurements are launched from Europe. Getting a probe physically near the Virginia Beach landing station to ping European targets would let us check symmetry and confirm that the tight ratio holds in both directions.
• Capacity upgrade signals. MAREA was designed for about 200 Tbps at launch, and upgrades since then have raised the real usable throughput several times without physical changes to the wet plant. Unfortunately that kind of progress is invisible in RTT data. What we can track are small shifts in latency that correlate with DSP upgrade events, and we would like to see whether those events leave a signature.
• The Virginia Beach ecosystem. MAREA is no longer the only hyperscaler cable at Virginia Beach — BRUSA, Dunant, and Confluence-1 also land there. Over time we want to compare their measured RTT profiles from the same European probes to see whether they share any transit properties or whether each cable ends up with its own consistently different path even when the endpoints are in the same building.

Design notes

MAREA is a point-to-point subsea system with 8 fiber pairs, commissioned in February 2018. At launch its design capacity was 160 Tbps, and it has since been upgraded to well over that through transponder refreshes without any changes to the underwater plant. The physical cable was manufactured and laid by TE SubCom (now known as SubCom), the same supplier behind most of the major modern hyperscaler trans-Atlantic systems. The installation vessel was the CS Durable, and the cable was spliced and energized in a multi-phase operation during 2017 and early 2018 that weathered, among other things, Hurricane Harvey passing over the installation area.

The ownership structure is instructive. Microsoft and Meta each hold a large share as anchor customers, but the cable is operated by Telxius — Telefónica's infrastructure subsidiary — which brings traditional telco operations experience to a project that would otherwise be run by software companies. This hybrid model has since been copied by other hyperscaler cable projects (Havfrue, Dunant, Equiano), and MAREA is in many ways the template for what modern trans-Atlantic hyperscaler infrastructure looks like. Not quite a private cable, not quite a consortium cable: a deliberate collaboration between the cloud buyers of capacity and a telco that knows how to run wet plant.