The word “cloud” is a marketing triumph and a geographical lie. There is nothing airy about the infrastructure that moves your AI queries, model weights, and training data around the planet. It is glass and steel, buried in silt on the ocean floor, running for thousands of kilometres between landing stations most people will never see. As generative AI pushes data volumes to levels that would have seemed absurd a few years ago, the industry’s real bottleneck is not chips or GPUs alone — it is the pipes. And the newest of those pipes is going under the sea.
The latest signal comes from a consortium led by Microsoft and Singapore-based Lightstorm, which is building a new undersea cable to carry the traffic that AI and cloud demand are generating. It is a reminder that the AI economy has a physical footprint, and that the map of who controls that footprint is being redrawn right now.
The project
According to TechStartups (July 2, 2026), Microsoft and Lightstorm are leading a consortium to build a subsea cable known as I-2SEA — a roughly 3,600-kilometre link intended to carry the growing data traffic driven by artificial intelligence and cloud computing. The naming and scope point to a system designed less for the internet of a decade ago and more for the AI-heavy internet of the next one.
What makes a project like this notable is not the length alone but the intent behind it. Older subsea cables were built primarily to move consumer traffic — video calls, streaming, web pages. A cable conceived in 2026 is being engineered for a different load profile: large model training runs shuttling data between distant data centres, inference traffic bouncing between regions, and the constant replication of enormous datasets that modern AI systems require. That is a heavier, more sustained demand than the internet was historically designed to carry, and it is why hyperscalers increasingly want to build cables themselves rather than lease capacity on someone else’s.
A consortium structure — pairing a hyperscaler like Microsoft with a specialist infrastructure operator like Lightstorm — is the emerging playbook. The hyperscaler brings the demand and the capital; the infrastructure partner brings the expertise in routing, landing rights, and the unglamorous physical work of laying cable across a seabed.
Why subsea cables are strategic
Industry analysis in 2026 makes the point bluntly: subsea cables form the physical backbone of the internet and, increasingly, the AI economy. Roughly the entirety of intercontinental data traffic travels through them. That makes capacity, resilience, and control of routes a strategic concern rather than a purely technical one.
Start with latency. AI workloads are sensitive to how quickly data moves between locations — where a training cluster sits, where the data lives, and where inference is served all affect performance and cost. A well-placed cable can shave milliseconds off critical paths and open up new options for how workloads are distributed across regions. In an economy where compute is scarce and expensive, the ability to move data efficiently between compute sites is itself a competitive advantage.
Then there is resilience. A single cable is a single point of failure; ships’ anchors, undersea landslides, and deliberate sabotage have all severed cables and knocked regions offline. Building more routes creates redundancy, so traffic can reroute when one path fails. For AI services that businesses now depend on, that redundancy is not a luxury — it is uptime.
Finally, geopolitics. Cables have to land somewhere, cross specific waters, and pass through chokepoints. Whoever owns the cable, and whoever controls the territory it touches, holds leverage — over pricing, over access, and in the worst case over surveillance and disruption. As more of the world’s economic activity flows through AI systems, the routes those systems ride on become instruments of national and corporate power.
The stakes
The most consequential trend in subsea infrastructure is concentration. A growing share of new cable capacity is being funded and owned by a handful of hyperscalers — the same companies that dominate cloud and are racing to dominate AI. When the firms that own the compute also own the pipes connecting it, they accumulate a kind of vertical control over the AI stack that is hard for challengers to match. That is efficient for them and, arguably, good for the users of their platforms in the short term. It is less obviously good for competition.
Security is the second stake. Cables are physically vulnerable and, in an era of heightened geopolitical tension, increasingly seen as targets. Redundancy across multiple routes and operators is the practical defence, which is one reason the sheer number of planned systems matters as much as any single project. A dense, diverse web of cables is more resilient than a few fat pipes owned by few players.
Third is cost and downstream access. The economics of who pays to build a cable shape who can afford to use it. If capacity is built by and for hyperscalers, smaller cloud providers, startups, and businesses in less-connected regions may find themselves buying access on terms set by the incumbents. The seabed, in other words, quietly influences the price of doing AI business on land.
The India read
For India, none of this is abstract. The country has been positioning itself as a data-centre and digital-infrastructure hub, courting hyperscaler investment and building out capacity to serve both its vast domestic market and global workloads. Those ambitions rest, literally, on connectivity — and much of India’s international connectivity runs through subsea cables landing on its coasts.
Treating those links as national infrastructure, not just private commercial assets, is the strategically sound posture. India’s data-centre boom is only as valuable as its ability to move data in and out efficiently and reliably. Sitting near the crossroads of major east–west routes gives the country real geographic advantage; realising it depends on landing rights, diverse cable systems, and policy that encourages capacity to be built rather than bottlenecked.
There is also a sovereignty dimension. As India pushes on AI ambitions and data localisation, the question of who owns and controls the cables serving Indian data centres becomes a matter of national interest. A future in which India’s AI economy runs primarily on infrastructure owned by a few foreign hyperscalers looks different from one in which Indian operators, regulators, and consortium partners have meaningful stakes and redundancy of their own.
The I-2SEA project is a single cable, and it is easy to file it under routine industry news. But it is also a small window into a larger contest. The AI economy has a geography, and that geography is being laid down now, quietly, on the ocean floor. For founders and operators, the lesson is worth internalising: the cloud you build on has a coastline, a seabed, and an owner. It pays to know who they are.
