A New Spine for the India Southeast Asia Corridor
The physical layer of the AI economy rarely makes headlines, but it is where the constraints are hardest and the timelines longest. On July 2, a consortium led by Singapore based Lightstorm and including Microsoft, Singtel, and Tata Communications announced the start of construction on I-2SEA, a submarine cable system connecting India, Malaysia, and Singapore. The roughly 3,600 kilometer route will run from Singapore to Machilipatnam on India's east coast, with onward connectivity to Hyderabad, and it is targeted to be ready for service in the fourth quarter of 2029.
What distinguishes I-2SEA from the generations of subsea cable that preceded it is its stated purpose. This system is designed from the ground up to serve hyperscalers, GPU infrastructure providers, and enterprises running AI training and inference workloads across the India and Southeast Asia corridor. That is a different design brief than the consumer internet cables of the past decade. AI traffic patterns, with their bursts of east to west data movement between training clusters and inference endpoints, impose demands on capacity and latency that legacy routes were never optimized to meet.
Who Is Building I-2SEA
The consortium structure is telling. Lightstorm is the majority owner and driving force, but the presence of Microsoft as a named partner signals that this is hyperscaler infrastructure as much as it is carrier infrastructure. Singtel and Tata Communications bring deep regional operating experience and existing landing station relationships, while NEC Corporation has been appointed system supplier and ASEAN Cableship Pte Ltd will handle marine installation. It is a coalition that spans cloud demand, regional telecom muscle, and specialist subsea engineering.
Amajit Gupta, Group CEO and Managing Director of Lightstorm, framed the project as an extension of the company's existing terrestrial AI transport network. Lightstorm works around a single mission, interconnecting intelligence, he said, adding that with its SmartNet AI Fabric already delivering AI ready transport across data centers and GPU clusters in India, the company can now offer the natural extension of that platform into the subsea domain. The pitch is continuity, a single fabric running from Indian GPU halls out across the ocean to Southeast Asia's cloud gateways.
Hyderabad and Chennai Meet Singapore
The route is a map of where India's digital infrastructure is actually concentrating. The cable lands on India's east coast, home to the fastest growing AI and hyperscaler data center clusters in Hyderabad and Chennai, and connects them directly to Singapore and to Malaysia's emerging data center corridor around Kuala Lumpur. Those two endpoints are not arbitrary. Singapore has long been the region's premier interconnection hub, and Johor and Kuala Lumpur have become the overflow valve as Singapore's power and land constraints push capacity across the border.
For India, the strategic logic is about escaping a bottleneck. The country has enormous AI ambitions and a rapidly expanding base of data center capacity, but its international connectivity has historically leaned on western routes through Mumbai. A high capacity eastern link to Singapore gives Indian AI infrastructure a direct, low latency path into the Southeast Asian cloud ecosystem and, through it, to the broader Asia Pacific. That reorientation matters for sovereignty conscious enterprises that want their data to traverse a predictable, regionally controlled path.
Why Hyperscalers Are Buying Subsea
A decade ago, submarine cables were built by consortia of telecom carriers and leased to everyone else. Today the hyperscalers are increasingly principal investors, and Microsoft's participation in I-2SEA fits that pattern. When a cloud provider commits to a cable, it is buying certainty. It is locking in the capacity and the route diversity that its AI and cloud services will need years before the demand fully materializes, because the alternative, discovering a connectivity shortfall after the data centers are built, is far more expensive.
The economics of AI make this pre commitment rational. Training runs and inference services move staggering volumes of data, and the cost of insufficient bandwidth is not a slow website but a stalled workload and an idle GPU cluster worth tens of millions of dollars. By taking an ownership stake, a hyperscaler converts a variable, market priced input into a fixed, controlled asset. I-2SEA is one more data point in a clear trend, that the companies building the AI cloud increasingly want to own the pipes that connect it, not merely rent them.
Engineering for Resilience at Depth
The technical details of I-2SEA reveal a preoccupation with resilience. The system uses optimized route planning combined with a deep burial strategy that targets three meter depth across the buried sections of the network. That depth is a deliberate defense against the leading cause of subsea cable faults, which is not exotic sabotage but the mundane hazard of fishing trawlers and ship anchors dragging across the seabed. Burying the cable deeper puts it below the reach of most of that activity.
Resilience is not a luxury feature for AI infrastructure, it is a requirement. A cable cut that might once have meant slower email now means degraded model serving and interrupted training for the enterprises that depend on the route. The emphasis on uptime reflects the reality that these systems are becoming load bearing for economic activity, not just communication. Designing for a high level of protection from the outset is cheaper and more reliable than retrofitting redundancy after the first expensive outage teaches the lesson.
The Long Runway to 2029
The sobering counterpoint to all this ambition is the timeline. I-2SEA is not expected to enter service until the fourth quarter of 2029, more than three years from the start of construction. Subsea projects run on the physics of ships, seabed surveys, and manufacturing lead times, and no amount of AI urgency compresses them much. That gap between the speed of AI demand and the speed of physical infrastructure is one of the defining tensions of this build out, and it shapes how enterprises should plan.
For technology leaders, the takeaway is to think in the same multi year horizons the infrastructure demands. The connectivity, power, and data center capacity that AI workloads will need in 2029 is being committed now, and the organizations that plan their regional footprint around these announced routes will have an advantage over those that assume capacity will simply appear on demand. I-2SEA is a reminder that the AI economy runs on assets with long lead times, and that foresight at the physical layer is a genuine competitive edge.
