India’s AI buildout is redrawing the data centre map. What does that mean for network operators?

For most of the last decade, India's data centre story was about Mumbai and Chennai. The bulk of the capacity sat in and around these two cities, the subsea cables landed there, and network planning largely meant connecting them to each other and to the world.

That picture has changed as hyperscalers set up cloud regions and content distribution across the country. AI facilities are now expected to accelerate the shift dramatically, because site decisions today rely more than ever on the availability of power and large land parcels. Those conditions are found on the outskirts of major metropolitan hubs and increasingly in smaller cities, and as data centre sites become more distributed, connectivity becomes the critical factor in getting this ecosystem up and running.

Ciena's 2026 Wave Services Report tracks this pattern globally. New data centres are coming up along corridors between established hubs and in regions with previously limited digital relevance, because large power capacity, land and local incentives are easier to secure there than in congested metros. Site selection decisions that once took years are now being made in months. The report describes this as a shift from hub-centric to corridor-centric growth, with power availability rather than legacy digital geography deciding where compute lives.

In October 2025, Google announced an AI hub in Visakhapatnam, an investment of approximately $15 billion between 2026 and 2030 that includes gigawatt-scale compute capacity and a new international subsea gateway. As the government noted, a gateway on the East Coast provides route diversity that complements existing cable landings in Mumbai and Chennai. For a country whose international traffic has largely entered from the West Coast and the Southeast, a major landing point on the East Coast changes how traffic will flow within the country.

The scale of what is coming shows up most clearly in power planning. According to information shared in Parliament, India’s data centre capacity grew from about 375 megawatts in 2020 to around 1,500 MW by 2025, and the Ministry of Power estimated in March 2026 that electricity demand from data centres would reach 13.56 GW by 2031-32. By July, the government had revised that projection sharply upward, telling Parliament that AI data centres alone are expected to add 26.3 GW of electricity demand by 2031-32, to be integrated into the grid and served primarily by renewable energy. When official demand estimates double within four months, it says something important about the pace of investment. Capacity on that scale will be built wherever reliable power can be secured, which means new sites well beyond today’s hubs, and the routes connecting them will carry a growing share of the country’s digital traffic.

New connectivity patterns are creating new possibilities

Each of these facilities needs high-capacity connectivity to other data centres, cloud regions, enterprises and users before it can serve any purpose. These connections, known as wave services, are high-capacity optical links sold and managed by network operators. The Wave Services Report projects that global revenue from wave services will grow at an 18% compound annual rate and exceed $50 billion by 2030. What makes the forecast interesting is where the growth comes from: more links, higher capacities, flexible service options and stricter reliability requirements, all of which translate into revenue operators can book, rather than simply more traffic to carry at higher cost.

What are the network implications?

The most useful way to think about this is that AI does not simply add traffic to the networks India already has. It adds new layers of connectivity with its own economics.

The first layer sits between AI data centres themselves. As training workloads outgrow single facilities, models are increasingly trained across multiple sites, and the connections between them must move enormous volumes of data with minimal delay. When compute is distributed this way, the network determines how efficiently very expensive GPU clusters are utilised.

The second layer emerges as AI moves from training to inference, where models serve applications, enterprises and users. Here the character of traffic changes. Machines increasingly exchange data with other machines, with more traffic flowing upstream and laterally between facilities rather than simply downstream to users. That changes what operators are asked to sell. They will need to offer a broader mix of high-capacity connections at finer granularity, with stricter commitments on latency and availability. Customers will also expect stronger security and data sovereignty assurances. Over time, operators will need to turn services up and down far faster than traditional provisioning cycles allow, eventually on demand through software interfaces.

Together, these layers explain why the redrawn map matters. The value of a route will depend not only on the capacity it carries today, but on how many training, inference and international paths it can serve as the ecosystem fills in. Operators that plan their networks around where power, compute and cable gateways are converging, and around the distinct service requirements of each layer, will be charting the future course of India’s digital traffic.

(The author is CTO and Vice President, Solutions Engineering Asia Pacific, India and Japan, Ciena. Views are personal.)