India's data centre push inland is a template for Global South

Narendra Sen has a humble origin story. His family comes from a village near Indore. They moved to the city in 1997 and everyone had to work, so Sen started a cyber-cafe. He slowly expanded into hosting servers. Today, the founder of Indore-based data centre company RackBank presides over a 10-megawatt facility on the outskirts of Madhya Pradesh’s capital, with two more data centres under construction – one of them in Raipur. He’s also secured land for a fourth site.

Sen did not chase Mumbai’s eastern seaboard or Chennai’s coastlines – both popular because they are landing sites for submarine cables that are the backbone of our global internet. Instead he chased land prices. “In Indore we paid roughly 30 lakh an acre,” Sen notes. “A similar plot in Mumbai is nudging 30 crore. If you buy expensive, you have to sell expensive. India’s startups cannot afford that.”



Cheap land is only the first link in Sen’s affordability crusade. Because power is the single largest operating cost for a data centre provider, RackBank negotiated a state subsidy that cuts its tariff to around 6 per kilowatt-hour – less than half the prevailing rate in Mumbai. “We reckon we can run 40% cheaper than a tier-one data centre yet meet the same Uptime Tier IV standards (the highest level of data centre reliability),” he insists. “That is what will democratise compute for colleges, fintechs and AI developers who simply cannot pay Western hyperscale prices.”

The economics matter because India is racing into an era in which every AI chatbot, logistics algorithm and streaming platform is underpinned by specialised silicon – and that silicon is ravenous. While a conventional rack might draw 10 kW, Sen’s next Indore data centre is being designed for 80-200 kW per rack, cooled by direct-tochip liquid loops and full-immersion baths built in-house. “General-purpose centres are yesterday’s technology,” he says. “Accelerated computing needs different architecture, and it can be done just as safely in central India as on the coast. AI can live anywhere if the power and cooling are right.”

A NATIONAL RETHINK

Sen’s philosophy – put capacity where land and electrons are plentiful rather than where submarine cables land – is rapidly becoming industry orthodoxy. Vipul Kumar , vice-president for edge and network at CtrlS Datacentres, says the momentum is unmistakable. “We have already established edge data-centre facilities in Patna and Lucknow and are expanding into Ahmedabad, Bhubaneswar, Guwahati, Bhopal, Jaipur and Nagpur,” he says.

For Kumar, the logic is strategic as well as economic. Decentralising data-centres, he says, is vital for India’s digital future. “Our facilities in these emerging cities are purpose-built to deliver low latency, comply with data-localisation requirements and offer cost-effective solutions,” he says.

Ashish Arora, chief executive of Airtel-owned Nxtra, says that with the growing demand from OTT platforms and content delivery network providers for low-latency and high-performance content delivery, edge data-centre networks have become crucial. The company today has over 120 edge facilities across over 65 cities, complemented by 14 hyperscale data centres, to ensure seamless streaming experiences even in smaller towns.
Fintech platform Infibeam Avenues is staking its future on a constellation of small facilities. Chairman Vishal Mehta says they are planning a network of 1to 2-megawatt data centres in at least ten cities, each costing 20-50 crore. The decentralised model, he says, spreads risk, supports low-latency AI workloads and can break even within 24 months.

Mehta adds that the shift inland is already reshaping corporate real estate. “Global capability centres (GCCs), once concentrated in major metros, are increasingly looking at tier-2 and tier-3 cities. The availability of advanced digital infrastructure, combined with cost savings and talent pools, is turning places such as Gandhinagar, Kochi and Jaipur into innovation hubs,” he says.

ECONOMICS OF THE HEARTLAND


As we said earlier, the economics of the heartland is compelling. Land in Vidarbha or Bundelkhand costs a fraction of Mumbai’s asking price; power is cheaper and, crucially, more abundant.

Latency is another driver. A widely used rule of thumb equates one millisecond of round-trip delay to roughly 100 miles of fibre. A server farm in Indore sits almost equidistant from Delhi, Mumbai, Hyderabad and Ahmedabad – meaning half the population can be reached in under 10 ms. “For a streaming service or a payment gateway that is good enough,” Sen says.

The final catalyst is AI. Training a large language model can devour more electricity than a thousand Indian homes use in a year and demands carefully choreographed GPUs. “AI is already a different war,” Sen says. “America leads, China follows, and India must create its own sovereign compute layer if we want to stay in the game.”

RackBank plans to deploy Nvidia’s forthcoming Blackwell chips within six months and market a sovereign “supercloud” to Indian developers. CtrlS is integrating renewableenergy hybrids; Nxtra is piloting fuel-cells; Infibeam’s pods are designed for phased GPU upgrades as demand grows.

Collectively these experiments could become templates for fastgrowing regions from Africa to Latin America. “If compute stays expensive, only big corporates innovate,” Sen argues. “Put affordable GPUs in tier-two cities and a college kid can fine-tune a healthcare model on local data. That is how you discover the UPI or Aadhaar of the AI era.” And this, he argues, applies to many countries around the world.

DEALING WITH POWER RELIABILITY

None of the pioneers underestimate the obstacles. Reliable grid power, fibre backhaul and skilled technicians are perennial headaches. “Power infrastructure reliability remains a primary concern in many locations,” Kumar admits, noting the need for hefty investments in redundancy. Mehta flags the same issue: “Many tier-2 regions still lack the grid stability and renewable integration needed to support large-scale data-centre operations.”
Yet, technical ingenuity is helping to close the credibility gap. RackBank’s pre-engineered, single-storey sheds can be assembled in six months – onethird the time a vertical citycentre tower might take – and its patented immersion tanks can slash cooling power by 60%. CtrlS and Nxtra are pursuing similar modular playbooks; Infibeam distributes its IT load across multiple micro-grids to avoid single points of failure.