Data centres in India pivot to strategic artificial intelligence focus

India is expected to undergo a major shift in its data centre business, with 40-50 per cent of its capacity dedicated to artificial intelligence (AI) and graphics processing unit (GPU) workloads by 2030, even as overall capacity trebles to 3 gigawatt (Gw) during the same period.

Global Cloud players such as Microsoft, Amazon Web Services (AWS), and Google are set to become major players in owning their own captive data centres, which will generate over 1 Gw of capacity in the next five years.
 

Currently, their captive capacity accounts for 10 per cent of the total live data centre capacity, but they have been aggressively negotiating and acquiring land in various locations across the country to build larger captive data centres.
 

Sunil Gupta, managing director and chief executive officer of Yotta Data Services, the first data centre provider in India to offer Nvidia GPUs for processing to clients, says, “We estimate that by 2030, we will exceed 3 Gw of data centre capacity, of which 40-50 per cent will be AI/GPU-based workloads. We expect the Big Three to own 30 per cent of this total capacity.”
 

Vivek Dahiya, head of the Asia-Pacific (APAC) data centre advisory team at Cushman & Wakefield, says, “India’s total data capacity crossed 1 Gw in the first quarter of this year. We expect it to reach 3 Gw within five years. We also see a big opportunity for India to become an offshore AI data centre hub, although it will take some time.”
 

The reason, according to Dahiya, lies in the cost-effectiveness of building and operating AI-led data centres, primarily due to lower costs compared to many Asean and APAC locations, especially since AI requires vast capacity for data processing. The consultancy firm outlines three distinct advantages that India offers: cheaper land and construction costs, affordable power (particularly from renewable sources), and lower employee costs. 
 

For instance, Cushman & Wakefield estimates that building a data centre with a capacity of 1 megawatt (Mw) costs $6-8 million in India, compared to $10 million in Indonesia and Australia and $8 million in Thailand.
 

Of course, trebling capacity will necessitate investments, with the price tag higher for data centres processing workloads on GPUs for AI. For example, a data centre with a typical information technology load would cost about $5 million per Mw, excluding land costs. 
 

In contrast, a data centre designed for GPU workloads could require an additional investment of $30 million per Mw. Thus, for a 24 Mw data centre exclusively dedicated to GPU workloads, the capital expenditure for building would be $120 million. If it accommodates 16,000 high-end GPUs, the total cost would reach $720 million.
 

Global Cloud players are ramping up their efforts. Microsoft has been particularly aggressive in acquiring land — a key expense 

in establishing a data centre campus. 
 

Reports indicate that the company has purchased two parcels of land in Hinjawadi, Pune, as well as a parcel in the Pimpri-Chinchwad area of Pune in 2022. They have also made acquisitions in Hyderabad.
 

Similarly, Google and AWS are building new capacity in Mumbai. Google is currently negotiating to acquire 22.5 acres in Navi Mumbai; if successful, this will be their first captive data centre. AWS is also exploring a land parcel near Mumbai.

However, the challenge in building AI-based data centres is obtaining enough GPUs from Nvidia to lease to clients. In India, Yotta is leading the way with an order of 16,000 GPUs from Nvidia, of which 4,000 have already been delivered. Gupta expects the remainder of the GPUs to arrive within the next few months and adds that they are considering acquiring an additional 32,000 GPUs.
 

The Ministry of Electronics and Information Technology has also signed an agreement with Nvidia for 10,000 GPUs, which will be allocated to startups.

BACKBONE OF DIGITAL TRANSFORMATION

- 40-50% of data centre capacity will be used for AI/GPU workloads by 2030
 

- Overall data centre capacity is expected to treble by 2030 to 3 Gw
 

- Global Cloud giants — Google, Microsoft, and AWS — will control a third of the capacity through data centres built and owned by them by 2030
 

- They are actively scouting and acquiring land for these developments