Cautious optimism on data centres

Tamil Nadu is recalibrating its industrial policy on data centres at a crucial time. Artificial intelligence has triggered a massive global infrastructure build-out, with recent incentives from the Centre sweetening the deal. TN, which has so far incentivised the industry with tax breaks and subsidies under its 2021 data centre policy, is allowing the policy to lapse. The govt has signalled it wants to step away from what it calls a "rat race".The state's stance has shifted over the past year—from celebrating large data centre investments to raising concerns about environmental costs and limited job creation. The core argument is that power- and water-intensive data centres generate relatively little employment. AI data centres host graphics processing units (GPUs), used to train or run large language models (LLMs), among other things. These facilities require more power and water for cooling than traditional data centres. The issue is sensitive in Chennai, where many data centre clusters are located in areas vulnerable to floods and water stress.But, some industry observers suggest the state's strategy might be counter-intuitive, especially as competition from other states intensifies. A senior executive advising companies on data centre investments said TN's grid infra is already under strain after years of underinvestment. Despite a single-window clearance system on paper, bureaucratic delays, limited transmission capacity and insufficient renewable energy availability remain concerns."That is why the state cannot currently compete with others such as AP, Gujarat and Maharashtra," the executive said, requesting anonymity. Another industry analyst said investors are also concerned about the lack of firm and green power availability.AI-led racks can consume more power than traditional IT racks—sometimes up to three times as much. Conventional racks used for storage and routine applications typically need about 5–10kW (kilowatts) per rack, but AI systems running high-performance GPUs and specialised processors can push power density to 30kW or more. The higher density also increases cooling requirements and power infra needs.Meenu Singhal, MD for the India region at Socomec, a power solutions company, said the growth of AI workloads is reshaping how operators think about energy infra. "In cities such as Mumbai, Chennai and Hyderabad, the growth of AI-led data centres has changed our perception of power infra," he said. "Compared with traditional data processing, AI workloads require much more power and are also highly sensitive to disruptions, which makes reliable and continuous power systems essential."Abhishek Anand, MD of Insignia Policy Research, said the economic impact of AI data centres is evolving, though govts can incentivise them if fiscal space allows. "It should not crowd out incentives for labour-intensive sectors," he said. Pointing to higher costs, infra gaps and regulatory barriers, Anand said govts should make it easier for the industry to buy power directly from generators and allow private distribution players at least in select cities such as Chennai.Equinix India operates data centres in Chennai that provide interconnection services linking cloud providers and internet exchanges. Manoj Paul, the company's MD, said hosting data centres offers strategic value by supporting industries that depend on digital infra, attracting anchor investments and spurring related sectors."The surrounding ecosystem of connectivity, energy, construction and digital services fuels broader economic activity. Data centres also create demand for equipment such as UPS systems, transformers and cooling infra. Several global vendors have set up manufacturing facilities to supply this equipment, which is creating jobs," he said. State industries minister T R B Rajaa said the govt is prioritising infra investments that align with efficient use of resources and long-term economic value. "Our policy will be to nudge resource-intensive data centres to move to peripheral areas with more headroom for additional infra, water and power. Our approach is to focus on incentives tied to location and renewables rather than mere tax breaks. As a govt, we recognise digital infrastructure is vital, but we are also careful not to enter any rat race to host hyperscale data centres," he said."We are also increasing our focus on innovation rather than storage infra. Partnerships such as the one with Sarvam AI reflect our strategy of building capability in AI research, compute and talent. In that context, smaller and more efficient AI-oriented data centres located in appropriate industrial corridors make more sense," he added.Despite the current policy encouraging data centre development beyond Chennai, most investments have remained concentrated around the city due to its strategic advantages, including proximity to end users and international cable landing stations. Infra gaps elsewhere have also played a role. Larsen & Toubro's Vyoma facility in Sriperumbudur hosts the latest NVIDIA GPUs and is preparing for next-generation B300 AI processors. The facility is expanding capacity at the same location and serves enterprises, SaaS providers and large language model developers.When asked about moving data centres beyond Chennai, Prashant Chiranjive Jain, head of the corporate centre at L&T, said AI training workloads are not only power-intensive but also highly variable, creating sudden swings in electricity demand that can affect grid stability."Placing data centres closer to renewable energy sources enables behind-the-meter power arrangements, reducing transmission losses and lowering capital expenditure. However, these advantages must be weighed against the lack of robust grid interconnectivity to meet contingency power requirements and the intermittency of renewable power," he said. Paul of Equinix said gigawatt-scale power supply sourced from multiple substations would be essential.(With inputs from G Balachandar)