EXA Infrastructure is preparing another large increase in transatlantic network capacity, announcing a 6,552km subsea cable between New Jersey and Somerset as hyperscale cloud and AI infrastructure providers increasingly buy entire fibre pairs rather than smaller slices of existing systems.

EXA Infrastructure expects the Meridian cable to enter service in the fourth quarter of 2029, linking the United States with a new landing station at Brean before connecting through terrestrial routes to Slough and the company’s wider European backbone. The system is being designed with 24 fibre pairs and more than 500Tbps of scalable capacity.

The project has reached Contract in Force with subsea supplier Xtera, taking it beyond an early route proposal, while EXA says it has secured significant anchor commitments from hyperscale and neocloud customers. Financial terms and the identities of those customers have not been disclosed.

For a network due to enter service three years from now, that early demand provides a useful indication of how large infrastructure buyers are planning for AI and distributed-cloud traffic. Subsea cables have long development cycles, so the networks carrying late-decade workloads have to be financed and ordered before anyone can know precisely how much traffic those workloads will generate.

Large buyers want more control over fibre

EXA says the pattern of demand is changing alongside overall traffic growth. Instead of purchasing individual wavelengths on a carrier’s network, the largest customers are increasingly seeking spectrum or complete fibre pairs, giving them greater control over how capacity is lit, upgraded, and operated over the life of the cable.

That model suits companies whose network requirements have become infrastructure-scale in their own right. Hyperscalers interconnect data centres across continents, AI providers move large datasets and model workloads between compute regions, financial institutions depend on resilient low-latency routes, while research organisations exchange increasingly large scientific datasets.

A complete fibre pair provides more flexibility than buying a fixed amount of managed bandwidth, but it also shifts more responsibility towards the customer or its infrastructure partners. Buyers need optical systems, network engineering, capacity planning, resilience arrangements, and enough demand to justify committing to a much larger block of infrastructure.

Meridian is designed to feed directly into EXA’s owned European network rather than ending as an isolated landing asset. From Brean, diverse terrestrial paths are planned towards Slough and onwards into the company’s continental backbone, allowing customers to purchase the ocean crossing as part of a wider route into European data-centre markets.

That integration is becoming more important as data-centre geography changes. London remains a large connectivity hub, but power availability, land constraints, national cloud strategies, and new AI facilities are spreading high-capacity infrastructure across more European regions.

Route diversity becomes part of cloud resilience

Subsea capacity is also a resilience asset because a small number of cable failures can affect large volumes of international traffic. Operators and major customers therefore look for geographic diversity between landing points and terrestrial routes so that one incident does not remove several supposedly separate paths at once.

Brean adds another UK landing point to EXA’s portfolio, while Meridian will become the company’s ninth transatlantic cable. EXA describes it as the first transatlantic system built by a single operator in a decade, a claim that also reflects how frequently new cable projects have been structured through consortia or partnerships.

The investment sits alongside EXA’s wider attempt to expand the physical footprint over which it has direct control. Techopia reported earlier this month that Ofcom is considering granting EXA statutory network-building powers in the UK, which would affect terrestrial infrastructure rather than the subsea system itself.

Together, the projects illustrate how cloud and AI growth is spilling into assets with very different investment cycles from software. Model capabilities can change within months, while planning, manufacturing, laying, connecting, and commissioning an ocean cable takes years.

That creates the risk of both shortage and overbuild. Too little capacity can constrain routes or raise the value of existing fibre, while excessive construction leaves expensive infrastructure underused. Anchor commitments reduce some of that uncertainty by demonstrating demand before service begins, although EXA has not disclosed how much of Meridian’s proposed capacity is contracted.

The claimed capacity of more than 500Tbps should also be read as the designed scale of the system rather than traffic that will appear immediately in 2029. Optical technology, customer equipment, service configuration, and actual demand will determine how much capacity is lit.

Meridian nevertheless extends the infrastructure race surrounding AI beyond processors and data centres. Computing clusters still need to exchange training data, replicate cloud services, connect users, and move workloads between regions, leaving long-haul fibre as one of the less visible constraints beneath digital infrastructure expansion.