Big tech companies talk about artificial intelligence as if it lives entirely in some magical cloud. They focus on the software, the models, and the chatbots. But the physical reality of AI is incredibly heavy, hot, and grounded in massive chunks of infrastructure. If you don't have physical cables running under the ocean to move petabytes of data, your advanced AI model is essentially useless.
That is exactly why a new consortium involving Microsoft, Lightstorm, Tata Communications, and Singtel just signed the papers to build a brand new 3,600-kilometer undersea cable system named I-2SEA. Running directly from Singapore and Malaysia to the eastern coast of India, this project represents a massive shift in how global tech infrastructure is built. It is targeted to be ready for service by the fourth quarter of 2029.
Most tech news stories just copy and paste the press release, call it a day, and move on. They tell you that it will boost data speeds or help companies run workloads. But if you want to understand why these massive corporations are spending hundreds of millions of dollars on a specific stretch of the ocean floor, you have to look at the unique demands of modern AI training and the structural bottlenecks currently choking global data networks.
The Real Route and Why the Locations Matter
Most international subsea cables landing in India hit Mumbai on the west coast or traditional landing points in Chennai on the east coast. The I-2SEA system is doing something completely different. It will feature dual landing points in India, hitting Machilipatnam in Andhra Pradesh and a new, distinct location in South Chennai.
Selecting Machilipatnam is a deliberate engineering choice. It provides the absolute shortest physical subsea access to Hyderabad. Why does Hyderabad matter so much? Because Hyderabad and Chennai are rapidly becoming the primary clusters for hyperscalers and GPU infrastructure providers in South Asia. Tech companies like Meta and Alphabet have already made major data center moves in these regions.
By avoiding the crowded, traditional landing hubs, the consortium achieves two major things.
First, they dramatically reduce the physical distance data has to travel over land to reach the massive server farms in the interior of the country. Light traveling through fiber-optic cables moves incredibly fast, but every extra kilometer of glass adds microseconds of delay. When you are syncing thousands of Nvidia GPUs across different regions to train a single foundational model, those microseconds turn into massive financial costs.
Second, this choice offers true path diversity. If an anchor drags across a traditional cable cluster near Mumbai or central Chennai, a huge chunk of India's international internet traffic can drop instantly. This new path ensures that AI training workloads remain active even if other regional infrastructure suffers physical damage.
The Brutal Physics of Training AI Over the Ocean
To understand why Microsoft and Tata Communications are investing in a purpose-built system like I-2SEA, you have to look at how training an AI model differs from loading a webpage or streaming a movie.
When you stream a video, the data flows mostly one way. If a few data packets arrive slightly out of order or face a tiny delay, the video buffer handles it. You don't notice a thing.
AI training is a completely different animal.
Distributed training involves breaking a massive neural network across thousands of distinct chips located in different data centers. These chips constantly talk to each other. They compute weights, pass gradients back and forth, and must remain perfectly synchronized.
The Problem with Jitter and Latency
If one cluster of GPUs has to wait for data from another cluster across the ocean, the entire training run stops. This is known as an idle state. You are essentially burning millions of dollars worth of electricity while expensive hardware waits around for data packets to arrive.
Two main factors ruin this process.
- Latency: The absolute time it takes for a data packet to travel from point A to point B.
- Jitter: The variation in that arrival time. High jitter means some packets arrive fast and others arrive slow, destroying the tight synchronization required for parallel GPU computing.
The I-2SEA cable is engineered from scratch to resolve this. Lightstorm is integrating its terrestrial network, which spans over 30,000 kilometers in India, directly into the subsea system. By combining the underwater cable with their low-latency domestic fiber network, they can tie AI regions across Singapore, Malaysia, and India into a unified end-to-end network. This specific design directly targets low-jitter, loss-optimized data transport, ensuring that cross-border training runs run smoothly without hitting packet-delivery bottlenecks.
The Unsung Hero of the Project is Three Meters of Mud
When people read about high-tech subsea networks, they focus on the lasers, the fiber strands, and the terabits of capacity. But the most critical detail of the I-2SEA announcement is actually its physical burial strategy.
The consortium announced that the buried sections of the cable will target a depth of three meters below the seabed across the entire network. That is significantly deeper than standard practice for many older maritime systems.
Why go through the immense expense of burying a cable three meters deep in ocean mud? Because the shallow waters of the Malacca Strait and the Bay of Bengal are some of the busiest shipping lanes on earth.
The Real Threats Under the Sea
Commercial fishing trawlers drop heavy nets that scrape along the ocean floor. Giant cargo ships drop massive anchors that can slice through an armored telecom cable like butter. Sharks occasionally bite cables, though that is mostly a myth driven by old videos; human maritime activity causes over 80% of all subsea cable faults.
When an undersea cable breaks, you can't just send a local technician to fix it. A specialized cable repair ship must sail out to the location, drop robotic tools to find the broken ends, haul them up to the surface, splice the microscopic glass fibers together in a sterile lab on the ship, and drop it back down. This process can take weeks or even months depending on weather conditions and ship availability.
For a company running standard enterprise cloud services, a cable break means rerouting traffic through a different path, which might slow things down slightly. For a tech giant running live AI inference models or multi-week training clusters, a sudden drop in connectivity can corrupt data and ruin the entire project. Deeper burial means fewer cuts, higher uptime, and predictable reliability for the companies paying for this capacity.
India's Explosive Data Center Growth Is Demanding More Fiber
This infrastructure build isn't happening in a vacuum. India is currently seeing an unprecedented surge in data center development.
Data from Macquarie Equity Research highlights the sheer scale of this growth. India's operational data center capacity sits around 1.4 gigawatts. Based on projects currently under construction, that capacity is on track to double by 2027. If developers fast-track their long-term plans, the country's data center capacity could expand five-fold by 2030.
Right now, India has about 17 active submarine cables handling its international data traffic. While that sounds like a lot, the maximum potential capacity across those lines cannot keep pace with a five-fold increase in server farms filled with power-hungry AI chips. At least 10 more subsea cables have been announced to help bridge the gap, but I-2SEA is distinct because it is being shaped specifically for the AI and hyperscale market from day one.
The Corporate Power Play Behind the Consortium
The makeup of this consortium tells you everything you need to know about where the tech industry is moving.
- Lightstorm: The majority owner. They are backed by I Squared Capital and are positioning themselves as a core network provider for AI architecture. They are even planning an initial public offering in India by mid-2027, eyeing a valuation up to 1.5 billion dollars.
- Microsoft: The hyperscale titan desperately needing infrastructure to support its massive global investments in OpenAI and its own Azure AI services.
- Tata Communications and Singtel: The established telecom giants with deep regional roots, owning the landing stations and existing connectivity infrastructure that makes these international agreements actually work.
- NEC Corporation: The Japanese industrial giant chosen to manufacture and supply the physical cable system.
- ASEAN Cableship: The maritime installation specialist tasked with the brutal physical work of laying the line across the ocean floor.
By joining forces under a Joint Build Agreement, these companies share the immense financial risk of building trans-oceanic infrastructure while securing guaranteed, private lanes of data capacity for their own future needs.
Your Next Steps to Prepare for the AI Infrastructure Shift
If you are a technology leader, network engineer, or enterprise decision-maker, you shouldn't look at this project as just another piece of foreign tech news. It is a clear indicator of how data traffic patterns are shifting. Here is what you should do next to adapt.
First, evaluate your regional data footprint. If your long-term plans rely on deploying AI models or using major cloud services in South Asia, you need to align your data center strategies with the upcoming landing points in Machilipatnam and South Chennai. Cities like Hyderabad are no longer secondary hubs; they are becoming central infrastructure nodes.
Second, audit your network architecture for path diversity. Relying entirely on traditional data routes through Mumbai leaves your operations exposed to regional cable congestion and physical maritime cuts. Look into how you can use newer, carrier-neutral landing paths to ensure your systems remain online when older networks experience outages.
Finally, keep a close eye on capacity commitments. The I-2SEA consortium has officially opened the system for commercial capacity commitments ahead of its 2029 launch. If your organization handles massive hyperscale transfers or distributed workloads across the Singapore-India corridor, securing dark fiber or dedicated bandwidth early is the only way to avoid surging infrastructure costs down the road.
