India is Building an AI Fibre Network. But Can It Survive a Cable Cut?

India’s AI infrastructure buildout has entered a new conversation. Data centres and GPU clusters are expanding beyond traditional technology hubs, while private infrastructure companies are planning to build fibre networks linking these facilities to each other and to global cloud markets.

Lightstorm’s ₹2,500 crore debt facility from IndusInd Bank is being used to accelerate its ‘AI Superhighway’, a terrestrial network connecting AI and data centre hubs across India and extending towards Singapore and Malaysia through its I-2SEA submarine cable system.

But the connectivity layer beneath this expansion remains vulnerable.

Subsea cables carry roughly 95% of intercontinental digital communications, according to a July 2026 report by the Takshashila Institution titled ‘Navigating India’s Subsea Cable Governance’. India consumes around 20% of intercontinental internet traffic but hosts only 18 of roughly 600 subsea cables globally. Eleven of India’s 18 cables are more than 20 years old and approaching the end of their expected 25-year operating lives.

“Largely, yes,” Anwesha Sen, Assistant Programme Manager at The Takshashila Institution and author of the report, tells AIM when asked whether new terrestrial AI-fibre links risk strengthening the domestic layer while leaving the subsea bottleneck untouched. “Fibre links between India’s AI data centre hubs strengthen connectivity for AI inference but leave the pre-existing subsea chokepoints untouched.”

Concentration Remains a Problem

India’s subsea network was built for a different era of the internet, with landing infrastructure shaped by commercial and financial considerations rather than today’s distributed AI workloads.

Sen points to the concentration of cable landing infrastructure around Mumbai. At least 13 of India’s 18 cables land within a six-kilometre stretch at Versova, Mumbai, according to her report. That creates concentration risk even when individual cables have redundancy.

“The first set was built at a different time for different purposes,” Sen says. One purpose was connecting Mumbai, India’s financial capital, to the rest of the world, resulting in the concentration of landing points there.

The architecture is now changing. I-2SEA, an approximately 3,600-km system being developed by Lightstorm, Microsoft, Singtel and Tata Communications, is designed to connect India’s east coast with Singapore and Malaysia, linking AI and hyperscaler clusters around Hyderabad and Chennai with Southeast Asia. It is targeted to go live in Q4 2029.

For Lightstorm, the argument is that AI requires a different connectivity architecture.

“The bottleneck isn’t compute. It never was,” says Amajit Gupta, Group CEO and MD at Lightstorm. He argues that while India has attracted investment into chip clusters, data centres and GPUs, the network layer has not kept pace.

AI training and inference depend on large volumes of data moving rapidly between distributed compute resources. Gupta says a temporary drop in throughput can do more than slow an ordinary enterprise application, as it can leave expensive GPU capacity idle.

Lightstorm currently operates more than 30,000 km of terrestrial fibre and 21,000 km of subsea infrastructure, connecting more than 100 data centres globally. Its ₹2,500 crore debt facility “on a 10-year, fully underwritten basis,” is intended to fund the next phase of the AI Superhighway, including links between India’s southern AI compute corridor and hyperscale hubs in Singapore and Malaysia.

Redundancy is Not the Same As Resilience

Cable systems are designed to reroute traffic when one cable fails. A cut, therefore, does not necessarily mean a nationwide blackout. The more immediate impact can be increased latency, congestion and degraded performance.

“The real effect is latency and degraded performance,” Sen says. “AI infrastructure — heavily reliant on high-speed, high-volume data transfer, particularly for inference—would be especially sensitive to a prolonged disruption.”

The problem worsens when several cables share the same geographical corridor. India’s international routes also rely heavily on the Strait of Hormuz and Bab el-Mandeb, both strategically important and geographically constrained maritime chokepoints.

According to reports, India’s per-capita fibre penetration lags significantly behind China, Japan, and the US. “That gap was manageable when networks were carrying consumer data. It becomes structurally significant when AI workloads with their requirements for low-latency, consistent, high-throughput transport are what the network must carry,” Gupta highlights.

Which is why Lightstorm argues that route and gateway diversity can reduce this exposure. Gupta says more than 92% of the company’s terrestrial network runs on utility-grade fibre, while I-2SEA uses dual Indian landing points and a carrier-neutral design. He says the system is designed to create “more routes, more owners, more options” when it becomes operational.

But new routes do not eliminate the wider repair problem.

Gupta states that raw bandwidth alone isn’t enough for AI workloads. These require consistency, low jitter, minimal packet loss, and predictable latency at scale. That is why, he says, the company is creating the AI Superhighway, designed specifically for AI transport rather than adapted from legacy networking.

“Through our Polarin NaaS platform, hyperscalers, neoscalers and enterprises can provision and monitor global connectivity on-demand, with real-time network visibility. The goal is infrastructure that stays ahead of the demand curve, not one that is perpetually catching up to it,” the group’s CEO adds.

India’s Repair Gap

India currently has no Indian-flagged, Indian-owned or permanently India-stationed cable repair vessel, according to Sen’s report. Repairs would depend on foreign fleets operating from locations such as Singapore or the UAE, with an average repair timeline of four to five months.

During geopolitical tension, foreign vessels could also face higher insurance costs or refuse to enter risky areas. That could further delay repairs.

“You cannot eliminate subsea cuts or geopolitical risk entirely. But you can eliminate the conditions under which a single event brings the whole system down. Diversity in routes and in gateways is the engineering answer to concentration risk,” Gupta adds. 

India has previously examined creating an Indian-flagged submarine cable repair vessel and cable depots on the eastern and western coasts. But the capability gap remains.

The regulatory framework is evolving too. The Department of Telecommunications has mechanisms for establishing cable landing stations, subject to security clearance and government permissions. The Telecom Regulatory Authority of India has also recommended frameworks for fair and non-discriminatory access to essential facilities at cable landing stations.

Yet, regulation alone does not create resilience.

Governance Must Catch Up

Sen’s report recommends treating subsea cables and cable landing stations as ‘Critical National Infrastructure’. It argues that this could enable stronger physical security, protected cable zones and faster emergency access for repair vessels.

The report also proposes a ‘Green Channel’ for cable repairs, allowing vessel entry, equipment imports, and maritime permissions to be pre-cleared, as well as stronger coordination between telecom operators, the Navy, and the Coast Guard.

Sen argues that resilience cannot be left entirely to private companies because the consequences of a systemic outage extend beyond any individual operator.

“Overall reforms need to be government-led through public-private partnerships,” she says. Her proposed model includes government support for an Indian repair fleet, pre-cleared land for new landing sites and emergency coordination, while hyperscalers and telecom operators co-invest in repair infrastructure.

Lightstorm’s buildout showcases both the opportunity and the gap. Its terrestrial network is designed to connect India’s AI clusters, while I-2SEA adds a new international route towards Southeast Asia. Gupta says the company is targeting corridors where cloud, hyperscaler and enterprise AI demand is already concentrating rather than building fibre speculatively.

But India’s AI ambitions increasingly depend on connectivity between domestic compute clusters and international AI hubs. That means resilience cannot stop at the data centre gate.

While a new fibre route can reduce congestion, a new submarine cable can add international capacity. However, neither, on its own, guarantees continuity during a major disruption.