The AI Energy Challenge
Artificial intelligence isn't just software; it's a physical process demanding immense computational power, which translates directly to electricity consumption. Unlike traditional computing, AI workloads, especially the training of large models, run
on power-hungry graphics processing units (GPUs) packed densely into server racks. These AI-specialised servers are causing a dramatic surge in energy use. According to Gartner, global data centre electricity consumption is projected to grow 26% year-over-year, reaching 565 terawatt-hours in 2026. This is largely driven by AI-optimised servers, which are expected to account for nearly a third of data centre power consumption in 2026. The International Energy Agency (IEA) forecasts that by 2030, data centres could consume almost 3% of global electricity demand. This rising demand is straining local power grids, with reports of wholesale electricity costs surging near data centre hotspots.
Measuring What Matters: PUE in the AI Era
For years, the key metric for data centre efficiency has been Power Usage Effectiveness (PUE). Developed by The Green Grid, PUE measures the ratio of a facility's total power consumption to the energy used by its IT equipment. An ideal PUE is 1.0, where all energy goes directly to computing tasks. Any value higher than 1.0 represents energy spent on overhead like cooling and power distribution. While the industry average PUE in 2025 was around 1.5, tech giants like Google have achieved fleet-wide averages as low as 1.09. However, the unique demands of AI—with its intense, variable power draws and high-density heat generation—are testing the limits of this traditional metric. Experts now argue that while PUE is still useful, it doesn't capture the full picture of efficiency, such as water usage or how effectively the computing resources themselves are utilized.
The Cooling Conundrum: From Air to Liquid
A significant portion of a data centre's energy overhead—often 30-40%—goes towards cooling. Traditional air-cooling methods are struggling to keep up with the intense heat generated by high-density AI racks. This has sparked a rapid shift towards liquid cooling solutions. Liquids are far more efficient at transferring heat than air; water, for example, has about four times the heat capacity of air. There are several approaches, including direct-to-chip cooling, where liquid is piped directly to hot components like CPUs and GPUs, and immersion cooling, where entire servers are submerged in a specialized, non-conductive fluid. These methods not only manage heat more effectively but also significantly cut energy consumption, reduce operational costs, and allow for even denser server configurations.
Smarter, Not Just Stronger Hardware
The race for efficiency extends to the very silicon that powers AI. Chipmakers are increasingly focused on performance-per-watt, not just raw processing power. In a landmark 2025 announcement, rivals Nvidia and Intel revealed a partnership to co-develop custom chips for data centres, aiming to tightly integrate their respective strengths in GPU and CPU architecture. This collaboration is geared towards creating more powerful and energy-efficient solutions for AI workloads. Nvidia’s newer architectures, for instance, are being designed with 100% liquid cooling from the ground up, allowing them to run more efficiently by managing heat at its source. By focusing on specialized, integrated designs, the industry hopes to curb the runaway energy growth of AI computation.
India's High-Stakes Balancing Act
India is poised to become a major hub in the global data centre expansion. The nation's data centre capacity is projected to triple between mid-2026 and 2029, fueled by massive investments and a booming digital economy. The Indian AI data centre market is expected to grow at a compound annual growth rate of 28.2% between 2026 and 2033. However, this rapid growth presents significant hurdles, primarily concerning the power grid and infrastructure. As AI adoption increases, data centres are projected to consume a growing share of the country's electricity. To sustain this expansion, India will need a coherent national policy that balances growth with sustainability, ensuring that land, connectivity, and, most importantly, power are available to meet the unprecedented demand.
