Why AI Is So Power-Hungry
A traditional data centre, the kind that hosts websites or stores your cloud documents, has significant but relatively predictable power needs. An AI data centre is a different beast entirely. The process of training large language models and other complex
AI systems requires immense computational effort. This work is handled by racks of specialised chips, primarily Graphics Processing Units (GPUs), which consume three to five times more power than the Central Processing Units (CPUs) in conventional servers. A single rack of servers in a traditional facility might draw 10-15 kilowatts. In an AI data centre, that same rack can demand 50-100 kilowatts or more. This intense activity generates a huge amount of heat, which in turn requires constant, energy-intensive cooling to prevent the high-performance hardware from overheating. This combination of high-powered computation and the necessary cooling infrastructure is what makes AI's energy footprint so massive.
The Staggering Scale of Consumption
The numbers are startling. Globally, data centre electricity demand is projected to more than double between 2024 and 2030, rising from roughly 415 terawatt-hours (TWh) to 945 TWh, with AI being the primary driver of this growth. To put that into perspective, 945 TWh is more than the total electricity consumption of Japan today. Some estimates suggest a single modern AI data centre can use as much power as 100,000 homes. The United States currently dominates data centre power usage, consuming nearly 40% of the global total in 2025—more electricity than was used to power the entire United Kingdom. This unprecedented growth is ending decades of relatively flat electricity demand in developed nations, putting immense strain on existing power grids. For a rapidly developing country like India, the implications are even more profound.
An Opportunity and Challenge for India
India's data centre market is in the midst of a massive expansion, fuelled by digitalisation, data localisation policies, and the global AI boom. The country's data centre capacity is projected to surge from around 1.4 gigawatts (GW) in 2024 to as much as 10 GW by 2030. This positions India as a key future hub for digital infrastructure. However, this growth presents a formidable challenge. Power availability is already becoming a key constraint determining where new facilities can be built. The immense new demand puts pressure on a grid that is already working to meet the needs of a growing population and economy. Key data centre hubs like Mumbai and Chennai face risks related to both grid capacity and the availability of resources like water for cooling. Ensuring this digital expansion is powered sustainably is no longer just an environmental goal; it's a core business and infrastructure necessity.
The Search for Cleaner Power
Tech giants are acutely aware of this problem and are making public commitments to power their operations with 100% renewable energy. The primary method is through Power Purchase Agreements (PPAs), where a company agrees to buy electricity from a renewable energy project, like a solar or wind farm, for a fixed term. In India, companies like CtrlS and AdaniConneX are making similar pledges. However, matching annual energy consumption with renewables is different from ensuring every hour of operation is carbon-free, which is the new goalpost. This has led to exploring more diverse energy portfolios, including on-site solar, battery storage systems, and even, in some global discussions, next-generation technologies like small modular nuclear reactors (SMRs) to provide constant, carbon-free power.
Efficiency Is Also a Critical Piece
Alongside finding cleaner sources of power, making data centres themselves more efficient is a major focus. The industry uses a metric called Power Usage Effectiveness (PUE), which measures how much energy is used for overheads like cooling versus actual computing. A perfect score is 1.0. While leading hyperscale facilities can achieve a PUE as low as 1.1, the industry average has remained stubbornly high at around 1.56. This means for every watt used for computing, another half a watt is spent on just running the building. Closing this gap is crucial. Innovations like direct liquid cooling, which is far more efficient than traditional air cooling, are becoming essential for managing the heat from dense AI server racks and can cut cooling energy by 30-40%. Indian operators are also adopting these advanced cooling methods to achieve better efficiency.
