Inside India’s AI data-centre boom: What it takes to build one

Existing players, including Yotta, CtrlS, and Sify Technologies, are also aggressively expanding AI-focused capacity across Mumbai, Hyderabad and Bengaluru.

“A lot of AI computing will be required at present. GPU (Graphics Processing Unit) computing is increasing exponentially. I don't see that saturating anytime soon. It's going to only increase in the coming years,” said Vishnu Subramanian, Head of Product and marketing at GPU cloud provider E2E Networks (former founder and CEO of Jarvislabs AI).

What’s an AI data center?

An AI data centre is a specialised version of a traditional data centre, one built to handle the high-intensity computational requirements of AI and machine learning workloads. This demands high-performance hardware such as GPUs and TPUs, low-latency networking and massive storage for training and deploying models.

It also means significantly higher energy and space requirements, along with more advanced cooling systems and a highly skilled engineering workforce to build and operate such facilities.

What goes into constructing a massive AI data centre?

According to S Anjani Kumar, Partner at Deloitte India, even the “grey space” of an AI data centre, which includes supporting IT hardware, requires capital expenditure on heavy-duty power infrastructure such as high-voltage substations, backup generators, chillers and advanced cooling systems. The “white space” then houses high-density racks of GPUs, servers, memory/storage, network gear, and orchestration software.

“In short, building a $15 billion AI data centre in India means an integrated megaproject of construction, power, networking and supercomputing gear,” said Kumar.

An AI data centre of this scale typically needs multiple gigawatts of power at full capacity.

Naresh Singh, Senior Director Analyst at Gartner, said the engineering involved is immense. These facilities will not only house hundreds of thousands of GPUs and CPUs, but also require extensive specialised networking, both inside the data centre and externally for connectivity.

“Apart from multiple technology providers that will come into play, for India, this also translates into very good opportunities for several service providers, including local companies. Operating such a data centre also means employment generation for several staff apart from service opportunity for allied offerings (data centre, cloud, AI, etc.),” Singh told Moneycontrol.

Such large-scale projects also depend heavily on state support: land allocation, data-centre parks, policy incentives and a local skilled workforce.

However, the single biggest hurdle is power. AI data centres are extremely power-hungry, which means each project drives substantial additional demand for transformers, substations, and transmission lines. “Utilities and companies often collaborate on this, with developers investing in on-site generators, microgrids and battery storage to supplement the grid,” Kumar explained.

Why existing data centres can’t simply be repurposed for AI?

With hyperscalers like Google Cloud, Microsoft and Amazon Web Services already operating large data-centre footprints in India, one question naturally arises: why can’t that capacity be repurposed for AI?

Sunil Gupta, Co-founder and CEO of Yotta Data Services, says that while not impossible, it is extremely difficult to re-engineer traditional cloud data centres for AI workloads.

“This is due to a simple reason,” he said. A traditional data-centre rack consumes 6–8 kW of power, and rarely crosses that limit. But once GPUs are added for AI workloads, that same rack requires at least 50 kW and can go up to 200 kW.

“The same rack now requires 8–15X more power. Then you have to deliver that high density of power in a small footprint of a rack. This will also generate that much more amount of heat,” Gupta said.

This means traditional cooling systems fail. AI facilities shift from standard liquid cooling to more advanced chip-level cooling, a significant engineering upgrade.

Yotta has repurposed parts of its facilities to cater to AI demand. “My NM1 data centre in Navi Mumbai has grown from 30-megawatt data centre to 54 MW now. Around 50% of this data centre’s capacity, that would be about three out of the six floors, are now fully flushed with GPUs,” Gupta said.

“In my case, I was lucky that my building footprint is very big. So I had a lot of space to keep extra generators, UPS and transformers, etc.”

The financial gap is significant as well. In a legacy CPU data centre, non-IT capex stands at $5–6 million per MW, and IT capex (CPUs) adds another $5–6 million per MW, around $10–12 million per MW in total. For an AI data centre, that number soars because GPUs alone cost $35–50 million per MW.

Yotta's AI data centre in India

Gupta said Yotta’s NM1 data centre currently hosts the largest GPU cluster in the country, followed by E2E Networks, which has deployed around 1,000 GPUs in L&T’s Chennai facility.

“Almost 70 percent of GPUs in India are running in that data centre NM1. Even my own data centre in Delhi doesn’t have GPUs. I have put all my GPUs in the Mumbai data centre. I understand that Amazon and Google would have brought in GPUs in India to some quantity, so maybe one or two of their data centres. They might have been running at least thousand plus GPUs. But that’s it,” he said, highlighting the still nascent stage of AI data-centre capacity in India.

Why states want AI data centres?

Deloitte estimates that India will need an additional 45–50 million sq. ft of real estate and 40–45 terawatt-hours of power by 2030 to meet demand from upcoming data-centre expansions.

Overall capacity is projected to grow to 6–10 GW by 2030 across various scenarios.

While these projects attract large FDI inflows and stimulate local spending during construction, their long-term job creation is more limited.

According to Gupta, a colocation data centre may require over 2,000 people during construction—engineers, construction workers, electricians, technicians. Once operational, however, even a 54 MW facility needs fewer than 50 employees.

Subramanian of E2E Networks said the more meaningful impact is the creation of high-skill engineering roles in a domain where India still has limited expertise. “There’s going to be a third-order effect, too. Most of these data centres will hire from India, enabling a bigger advantage as people who work there will get a sense of the know-how,” he said.

Kumar added that during steady-state operations, such a facility will require “50+ permanent staff (facility managers, engineers, maintenance, etc.). Additionally, it is estimated that for every job within a data centre, around 3.5 jobs are created in the surrounding economy.”

As the Digital Personal Data Protection (DPDP) Rules come into effect, the demand for data centres to support India’s data localisation needs is set to accelerate. State governments are already vying for these investments with incentives and dedicated data-centre policies. Meanwhile, global technology companies are doubling down on India as one of their fastest-growing markets.