Can India’s power system keep up with the explosion in data centres?

India’s power system is headed towards a “paradigm shift” as artificial intelligence (AI)-driven data centres scale up across the country, Samir Chandra Saxena, chairman and managing director of Grid India, said at the India AI Impact Summit at Bharat Mandapam on Tuesday (February 17). While AI runs on algorithms, it also requires electricity, and in large volumes. In India, total installed data centre capacity currently stands at about 1.2 gigawatts (GW) and is projected to grow fourfold by 2030, according to government estimates, as AI-led computing demand accelerates. This rapid expansion could reshape the country’s power landscape as data centres are emerging as large, complex and highly dynamic loads on modern grids. Saxena said the rapid expansion of data centres over the next few years would significantly alter grid planning and operations. Grid risks “These are very intense loads,” he said, unlike the traditional bulk demand seen at the distribution level. “Now these would be at the transmission level because sub-transmission may not be able to cater to this kind of requirement,” he said during a discussion titled, ‘From Insights to Action for Resilient, High-Performance Data Centres’. Abhishek Ranjan, CEO of BSES Rajdhani Power Limited (BRPL); Saumer Phukan of Intel; Deepesh Nanda, CEO Tata Consultancy Services; and Karthik Krishnan of Amazon Web Services were part of the discussion. By 2030, Saxena said, data centre penetration could reach 8–10 GW. “Even that size is going to be very significant,” he noted. Unlike conventional industrial loads, data centres exhibit sharp variability. “It is variable, it is spiky, it has sharp ramps,” he said, adding that they can also make “silent exits” from the grid. “Anything happens on the grid side, the data centres prefer to quietly isolate themselves.” Since most are inverter-based, he warned that a sudden withdrawal of 1–1.5 GW — or even a couple of gigawatts — “simply walking out of the system quietly creates a disturbance kind of situation for the grid.” “This needs to be planned for…We need to handle this in a more planned manner rather than having it in random fashion,” he said. While emphasising the need for meticulous infrastructure planning — including a reliable and robust transmission network, strong connectivity and clear compliance mechanisms — he also highlighted the equally critical challenge of ensuring resource adequacy. “Such a large chunk of demand must be obligated to meet the resource adequacy requirements — not only in terms of primary energy, but also for reserves and balancing requirements,” he said. Saxena said the AI, especially during the prompt phase, is highly unpredictable, which would make forecasting and scheduling challenging for data centres. “I believe that this is something that needs to be handled at the demand side itself, because everything cannot be passed on to the grid. The grid cannot absorb all these risks,” he added. Saxena said many jurisdictions across the world require these data centres to come mandatorily with some form of generation to manage variability and ensure adequacy. “So I think in India, also our systems and code and standards need to evolve further to be able cater to this kind of new requirement before this large chunk comes into the system. And of course, energy storage, and all need to be integrated,” he added. ‘Chaos if not planned’ Echoing the need for planning to meet such huge demands from data centres, Abhishek Ranjan, CEO of BSES Rajdhani Power Limited (BRPL), warned that rapid hyperscale expansion could strain the grid if not carefully mapped. Each hyperscale facility can require close to a gigawatt of power, he noted, underscoring that such demand would necessitate high-voltage grid substations. According to him, planning needs to be there at the level of central level transmission operators like CTU and Grid India, or the State Transmission Utilities, about how and where this high voltage grid substation has to be created. On the supply side, Ranjan noted that the baseload demand of hyperscale data centres is stable and therefore requires long-duration, stable generating sources. While the United States is increasingly relying on nuclear power for baseload supply, India has yet to significantly scale up its nuclear capacity. For now, he said, a mixed approach would be more practical. “Are you going to depend on the grid completely, which might not be possible? So apart from the grid, you should have a local, say, long-duration, maybe six hours or eight hours or nine hours of battery storage, some technology co-located with that centre. some captive generation as well, maybe renewable generation with solar and wind,” he said. Since the connectivity required by these data centres will be huge, he suggested the Central Electricity Authority should create a national plan for possible sites where data centres can open. “That has to be predecided, otherwise it may lead to chaos and unnecessary adding of wires here and there, suboptimal organisation and therefore, the burden of tariff on end consumers,” he added. RE and efficiency Deepesh Nanda, CEO of Tata Consultancy Services, said renewable energy could play a key role in powering upcoming data centres through the open access route, noting that India has surplus generation capacity in parts of the system. Open access in the power sector is a regulatory framework that allows consumers with a connected load of 1 MW or more to purchase electricity directly from competitive suppliers or power exchanges, bypassing local distribution companies (DISCOMs). “Even our pump hydro projects will probably come in line to coincide with the build-out of the data centres. And hence, our ability to offer round-the-clock green electricity at a rate that’s affordable for the term of the contract will be important,” he said. Intel’s Saumer Phukan highlighted that the semiconductor companies are at the centre of the growing power challenge driven by AI, but added that Intel is addressing the issue at multiple levels. At the silicon level, he pointed to innovations under the 18A process — including RibbonFET technology, backside power delivery (power-via) and advanced Foveros packaging — which lower operating voltage, improve efficiency by around 15 per cent and reduce power consumption by minimising data movement between memory and compute. Beyond hardware, he said Intel advocates a “heterogeneous AI” approach, arguing that not every workload requires power-hungry GPUs, and that smarter workload allocation can significantly reduce overall energy demand. Amazon Web Services’ Karthik Krishnan said AI deployment in India is still in its early stages. “AI is scaling, but it hasn’t scaled yet. Many of these problems are in the future — perhaps 18 or 24 months from now. We still have an opportunity to address them,” he said. For hyperscalers, he added, speed and certainty of power availability, assured access to 100 per cent renewable energy, regulatory clarity and long-term price visibility are critical factors when selecting locations for large data centre investments.