Why Indian Data Centres Have a Long Way to Become AI-Ready

Traditional data centres were never designed for AI workloads. They were built for variable CPU loads, lower rack density, and air cooling. But AI workloads need dense GPU clusters, near-continuous utilisation, faster networking, stronger cooling, and tighter data controls. 

The demand is compounded by companies increasingly sanctioning AI tools for their employees. According to a Deloitte survey shared with AIM, AI workloads could reach over 50% of total Indian data centre capacity by 2033. 

Talking to AIM, Ashish Arora, CEO of the data centre company Nxtra by Airtel, believes that AI's next phase of growth will depend not only on compute power but also on resilient, secure, and sustainable infrastructure capable of supporting mission-critical workloads across industries.

"Our goal is to make AI deployments predictable, telecom-grade, secure, and cost-effective at scale while remaining sustainable by design, so enterprises can focus on creating value rather than operating infrastructure," he says.

Operational Challenges 

According to Cushman & Wakefield’s report, India’s operational data centre capacity stands at ~1,600 MW; however, AI-specific capacity is estimated at only around 20–25%. This figure is expected to grow exponentially as India’s AI market balloons to $131 billion by 2032, registering a compound annual growth rate of 39%, per Avendus Capital. 

“The challenge for enterprises now is whether enough AI-ready capacity will be commissioned in their preferred geography and on their preferred timeline,” Kumar adds.

Retrofitting older sites is expensive and slow. Operators must either invest in enhanced power, cooling, MEP, and structural systems or risk owning sites that cannot support future AI demand.

Anjani Kumar, Partner, Deloitte India, observes that traditional enterprise racks typically run at 5–15 kW, while AI training and large-scale inference are moving toward 30 kW+ per rack. Advanced AI clusters are already reaching 80–100 kW, with some next-generation liquid-cooled deployments moving beyond 100 kW per rack. 

Cooling often accounts for about 40% of data centre power consumption, and AI-led demand could need 40–45 TWh of additional power, per the Deloitte survey. Operators need advanced cooling architectures such as direct-to-chip liquid cooling and immersion cooling; however, this increases upfront costs and requires facility redesign.  

Liquid cooling is expected to become dominant in India as it is typically closed-loop and uses very little water. However, it can exacerbate local water stress. “More data centres mean more freshwater diversion for cooling, often coinciding with India’s recurrent water shortages and summer heat stress,” the Council on Energy, Environment, and Water (CEEW) noted in its blog.

Moreover, established data centre hubs such as Mumbai, Chennai, Delhi-NCR, and Bengaluru are already under pressure. They account for most of India’s operational capacity, but local transmission capacity, substation readiness, and grid redundancy remain key constraints. Future AI-led capacity is also expected to cluster in specific zones outside metropolitan areas, where large campuses of 100–200 MW+ will require stronger power planning upfront.

“Locating renewable energy plants farther away would entail substantial investments in transmission infrastructure, adding pressure on the already strained finances of distribution companies,” CEEW noted.

However, India’s AI data-centre growth essentially depends on imported GPUs, with the country only being able to secure a little over 38,000 GPUs through the IndiaAI Mission.  Currently, most India-bound AI accelerators are NVIDIA-led today, while AMD and custom-chip adoption remain limited, according to Kumar. 

“API-based cloud AI is faster and cheaper at low usage. However, as usage grows, token costs become a recurring operating cost and can make dedicated digital infrastructure more economical,” he mentions.

Need for Sovereign Data Centres

India generates nearly 20% of the world’s data but hosts only about 3% of global data centre capacity. This creates a sovereignty gap because the demand for Indian data and AI workloads is growing faster than domestic infrastructure. Sovereign AI data centres are needed to process, store, and govern more of this value within India. 

“Sovereign AI data centres are critical for India. They are foundational infrastructure for how India trains, hosts, governs, and scales AI. Without them, India's data, models, public services, and regulated industries remain dependent on infrastructure controlled outside its borders,” Kumar adds.

Indian enterprises are increasingly concerned about foreign-controlled AI stacks. According to Deloitte, around 82% of Indian firms report that 21–60% of their AI stack is foreign-owned or foreign-controlled, and 77% are at least moderately concerned. 

India’s DPDP framework, RBI localisation rules, and sector-specific compliance requirements make auditable infrastructure important for sensitive workloads in sectors including banking, insurance, healthcare, telecom, and the public sector. 

Kumar highlights that India is backing local foundational and large language models from teams such as Sarvam AI, BharatGen, Gnani.ai, and Socket. These models need Indian datasets, local evaluation, retraining capacity and inference hosting. Meanwhile, platforms such as AIKosh and Bhashini are strategic assets because they support AI in Indian languages and local contexts. 

Nxtra also believes that investments in sovereign digital infrastructure will be central to India's AI economy as they would help enterprises move beyond isolated pilots toward enterprise-scale AI deployment.

The company operates a network of 15 hyperscale data centres and over 120 edge data centres in India. Its portfolio spans startups, GCCs, hyperscalers, and enterprises modernising mission-critical workloads.

In March, it secured $1 billion in investments to expand its network, targeting a 25% market share. Nxtra, along with Google and AdaniConneX, recently broke ground in Visakhapatnam to build a gigawatt-scale AI data centre.

We provide secure and compliant facilities protected by multiple layers of physical security and advanced cybersecurity controls, ensuring the safety of critical workloads and sensitive data,” Arora shares.

At the same time, the company prioritises energy-efficient operations. The company’s core data centres now run on 49% renewable energy, contracting 483 GW of renewable energy so far.

Arora notes that standardised infrastructure enables organisations to shift investments away from infrastructure management towards building AI-driven products and business innovation. "Scale converts infrastructure from a barrier into an enabler. Our approach is to capture these efficiencies so organisations can adopt AI more ethically, affordably, faster and with confidence," he states.

Looking ahead, Nxtra expects three major trends to shape India's AI infrastructure landscape: the emergence of sovereign AI ecosystems for governments and regulated industries, data centres evolving into integrated innovation platforms combining compute, storage, networking, and security, and wider enterprise AI adoption driven by reliable, low-latency infrastructure supporting intelligent automation, edge computing, and real-time decision-making.

Large-scale data centre networks in India are primed to significantly reduce the cost of enterprise AI adoption by standardising infrastructure, improving operational efficiencies, shortening deployment timelines, and lowering security and compliance costs.