Data centres test India’s energy discipline

The AI infrastructure push must be backed with efficiency-linked approvals and smart resource management

In February, India wrapped up its Artificial Intelligence (AI) Summit with the New Delhi Declaration – a bold vision signed by over 92 countries that commits to using AI in an inclusive, equitable, and socially beneficial manner. Alongside that came something more tangible: announcements of over $200 billion in investments for AI infrastructure and hyper-scale data centres.

Two days later, on February 23, the city council of Denver, Colorado, proposed a moratorium on new data centres due to mounting pressure on land, energy, and water resources. Denver is not an outlier. Over the last few years, many regions have paused, capped, or restricted data centre development.

Data centres underpin India’s rapidly expanding digital economy – from Unified Payments Interface (UPI), Aadhaar, and over-the-top (OTT) platforms to cloud software and AI applications. But is the country prepared for the constant and immense electricity demands of data centres? This question is especially crucial because how India scales AI infrastructure will influence the pace and direction of its energy transition.

The Ministry of Power (MoP) estimates that electricity demand from data centres could reach 13.6 gigawatts (GW) by 2031-32, roughly equivalent to the power demand of five Bengaluru-sized cities. At this scale, data centres begin to resemble heavy industries (such as steel plants or aluminium smelters) but without the regulatory scrutiny that usually accompanies them. Adding to this, these centres are geographically concentrated in resource-stressed urban hubs grappling with constrained transmission networks. Mumbai hosts over 70 data centres; Hyderabad, Delhi NCR, Bengaluru, and Chennai host around 30 each.

Fortunately for India, countries such as Ireland, Singapore, and the United States experienced the data centre boom in the early 2010s, giving us a template on what mistakes to avoid and what solutions could work. Ireland offers perhaps the clearest cautionary tale. Once known as the data centre capital, the country imposed restrictions on new data centre connections in key regions in 2021. Multiple rounds of public consultations were conducted, with an electricity policy for data centres – the Large Energy User Action Plan – issued in January 2026. The policy requires data centres to provide on-site or local generation and/or storage capacity to match their demand and mandates that at least 80% of annual electricity consumption be met through renewable energy (RE).

Singapore, too, paused new approvals for data centres between 2019 and 2022. During that period, the government worked with industry and academia to develop mandatory energy efficiency standards for IT equipment, aiming to reduce energy consumption by at least 30%. Today, Singapore offers targeted incentives such as tax exemptions and funding for data centres that can demonstrate measurable carbon reductions.

In 2020, the Amsterdam Duurzaam Digital policy mandated a Power Usage Effectiveness (PUE) of 1.2 or lower for data centres. Amsterdam also stopped approvals for hyper-scale data centres (above 70 megawatt [MW] and 10 hectares) in 2024.

In the US, approvals are largely handled at a state level. In Georgia, a special commission on data centre energy planning has been proposed to review both electricity and water use, as well as to develop new energy efficiency guidelines. The state has also proposed a bill to halt all projects until March 2027, giving planning departments time to set the necessary policies for regulating data centres. Virginia will review several bills on data centres, including bills proposing a temporary moratorium and stricter planning restrictions on water and energy use. In Texas, data centres are required to participate in demand management programmes or disconnect from the grid during supply shortages.

China launched a three-year action plan in 2021 with clear quantitative targets, followed by several updates since. These targets include a PUE below 1.25, RE requirements of up to 80% in national hub cities, and deliberate siting of data centres near large solar and wind farms.

India’s learning: Plan, then execute

The demand for data centres is real and rising. India’s challenge is, therefore, not whether to build data centres, but how, where, and at what pace.

First, planning is not optional. Resource adequacy and long-term load forecasting must account for data centres. This requires coordination between the Central Electricity Authority, distribution companies, and state load dispatch centres. India already plans and forecasts the electricity requirements for other large bulk consumers such as railways, metro systems, and heavy industries. Data centres, emerging as a comparable and inflexible base load, remain outside this framework.

Second, approvals must be linked to grid readiness and energy efficiency standards, and connectivity should depend on transmission availability. Fast-track, single-window clearances for projects may look attractive, but they are risky for infrastructure that locks in demand for decades.

Finally, India needs to rethink its incentive structure for data centres. Today, central and state governments offer generous subsidies for AI infrastructure – 20-year tax holidays, capital investment credits, and power tariff concessions. Instead of blanket support, incentives should be tied to performance obligations, such as on-site or local storage, participation in demand-response markets, RE procurement, and progressively lower PUE benchmarks. As small modular reactor technologies scale, this option can also be incorporated.

To capitalise on the AI boom, India must act with intent, building capacity in phases, developing a regulatory framework, and conserving already stressed local resources.

(The writer is a Senior Associate in the Energy Policy and Regulations group at the Centre for Study of Science, Technology and Policy [CSTEP], a research-based think tank)

(Disclaimer: The views expressed above are the author's own. They do not necessarily reflect the views of DH)