The transatlantic bottleneck for AI isn’t fiber—it’s who gets to land it

AI training and inference do not saturate networks uniformly. They concentrate traffic into the cheapest, lowest-latency paths between large “islands” of compute—typically data centers and cloud regions—then expand outward as models, safety, and user demand scale. The transatlantic subsea layer has become the gating step where that traffic first becomes real, because a cable’s designed capacity only helps if it can be terminated, powered, and extended to domestic interconnect points.

In practice, that means the “forgotten bottleneck” is landing-station access and control: who owns/operates the cable system, who secures landing rights, and who gets to place the domestic handoff inside a data-center-like environment where hyperscalers can interconnect quickly.

A concrete example of the new pattern is Google’s transatlantic Nuvem buildout. Reuters reported that Google’s Nuvem subsea cable system spans roughly 7,000 km and comprises 16 fiber pairs with total design capacity of around 384 terabits per second—then ties the design to specific Atlantic-facing endpoints.

On the landing-station side, Google is also building a dedicated cable landing facility in the Azores. Submarine Networks (via its system page on Google’s landing project) describes a Cable Landing Station (CLS) at Lagoa (São Miguel island), including facility power/size and an expected commissioning timeline, and explicitly frames it as a key connection point for Google’s Nuvem and Sol transatlantic submarine cables.

A cable landing station (CLS) is more than a beach manhole. Equinix’s explainer lays out the internal separation of subsea power and subsea optical data, and it names the core termination subsystems: cable termination equipment (CTE) that separates fiber and power, power feed equipment (PFE) that supplies repeaters for long spans, and submarine line terminal equipment (SLTE) that terminates optical fiber to extend data traffic into domestic networks. It also notes a key operational advantage when the CLS is inside a data center—backhaul can be removed, reducing latency between the subsea termination and interconnection.

This is the supply-chain shift investors should internalize: subsea cable capacity is necessary, but termination location and interconnection proximity determine when that capacity can monetize into cloud traffic and wholesale services.

In the U.S., landing and operating subsea cables is regulated through FCC licensing. The FCC guide on submarine cable landing licenses states that operators must obtain a cable landing license prior to landing a submarine cable that connects the continental U.S. with a foreign country (or other specified U.S. endpoints). It also notes the basis in the Cable Landing License Act of 1921.

That matters for AI buildout because new cable systems don’t just compete on price and capacity—they compete on time-to-landing approval and which entity can legally control the system and its U.S. endpoints.

Ownership is not one thing. It splits across (1) subsea cable system ownership/financing and (2) cable landing station architecture and (3) the domestic network and interconnect environment where traffic can be exchanged.

In this framework, Google’s Nuvem shows a hyperscaler choosing its subsea route and endpoints; Equinix illustrates a venue model where cable termination architecture can be integrated into data centers; and the CLS architecture explainer shows why the “last hundred meters” can become decisive for latency-sensitive traffic.

For the AI investor, the key is to trace the monetization path: AI traffic can only flow into the compute ecosystem when the ownership/control of each layer aligns.

The landing-station bottleneck shows up indirectly in listed fundamentals. If AI hyperscalers add capacity in phases, the beneficiaries tend to be (a) firms that build and operate interconnection-rich environments near landing points and (b) firms that can monetize cable-system build cycles.

Below are directionally relevant scale metrics for three major listed equities in this connectivity ecosystem—Equinix (data center interconnection), Alphabet (Google’s cloud and infrastructure engine), and Prysmian (subsea cable manufacturing and systems).