Rohan Sheth, SVP and Head – Colocation, Hyperscalers and Expansion, Yotta Data Services
The artificial intelligence (AI) market in India is projected to reach approximately $17 billion in the near term. Beyond domestic demand, there is a credible opportunity for India to emerge as a global hub for AI services, with infrastructure forming the foundation and reshaping what different users expect from data centres.
AI infrastructure needs
For enterprises and hyperscalers, a data centre providing power, cooling and rack space is sufficient, as they bring their own servers and hardware. AI innovators and start-ups have fundamentally different requirements. What they need is on-demand compute, GPUs, networking and a ready-to-use platform delivered on an opex model.
However, the cost and availability of high-end GPUs is the single biggest hurdle for AI start-ups today. If infrastructure providers can offer GPU access on a pay-per-use basis, alongside training and inferencing capabilities as a platform service, that removes the most significant barrier to their work.
Data centre design for AI workloads
The shift from a cloud data centre to an AI factory is not incremental. It is architectural. A typical enterprise rack historically consumed 2-3 kW on average, within a 5-6 kw design envelope. The same rack footprint today, loaded with the latest GPUs, carries more than 200 kW. That represents a tenfold increase in power density. As a result, the server hall has reduced in relative importance.
Yotta: A case in point
What now dominates the design challenge is power: the number of transformers, generators and chillers required to support 100 MW or 500 MW facilities is the primary constraint, not the dimensions of the server hall.
Yotta experienced this directly. The first data centre in Navi Mumbai was designed as a 30 MW facility for enterprise and hyperscale cloud loads. Because of the shift to AI workloads and the resulting increase in rack density, that same facility had to be upgraded to 50 MWs of IT load capacity.
Role of liquid cooling
Air cooling is effective for up to approximately 15-20 kW per rack. Beyond that threshold, the only viable path is to bring the water source to the rack or to the chip. Direct liquid cooling and immersion cooling are the technologies that enable rack densities of 80 kW, 100 kW, 150 kW and above. In Yotta’s operational facilities across Navi Mumbai and Greater Noida, nearly half of the total load is already on liquid cooling, with the GPU and AI load entirely on liquid cooling infrastructure. That split reflects where demand has moved in just the first one to one and a half years of operating at scale. This reflects rising AI-driven densities and the need for integrated planning across cooling, energy and infrastructure.
Scaling data centres
Yotta’s Panvel campus is a 2.5 GW facility spread across 600 acres. Greater Noida is a 250 MW campus. If grid connectivity at a location is 200-500 MW, that is insufficient. The minimum threshold for serious hyperscaler evaluation is now 500 MW-1 GW, with a 5-10-year planning horizon.
Beyond scale, location strategy is also evolving. With access to extensive land banks through its parent group, Yotta is positioned to expand across multiple markets including Mumbai, Chennai, Bengaluru, Hyderabad and Kolkata, while also exploring edge deployments in Tier II and Tier III cities based on emerging demand patterns.
Alongside scale, efficiency metrics have also expanded. Power usage effectiveness alone is no longer the standard. Water usage effectiveness and carbon usage effectiveness now sit alongside it, because power consumption has increased tenfold and the source of that power – grid or renewable – is now a direct criterion for customer and investor evaluation alike. Shifting to renewable energy is no longer a differentiator. It is a requirement.
From an operational standpoint, a significant share of capacity is already supported by renewable energy, with some campuses sourcing as much as 80-90 per cent of their power from green sources. As AI-driven densities increase, reliance on traditional grid power alone is no longer sufficient, making diversified energy sourcing and long-term sustainability planning critical to maintaining performance, cost efficiency and uptime.
The way forward
With these shifts under way, the industry focus over the next two to three years will move towards execution, aligning capacity with demand and building integrated service capabilities. With demand continuing to accelerate, the opportunity lies not just in building infrastructure, but in delivering it at the right scale, in the right locations and with the right service layers.
