India’s AI race won’t be won by better chips. It’ll be won by better power.

Every nation-state in this great-tech game is searching for a decisive advantage in this geo-technological landscape. Those at the cutting edge of AI research and development have been highlighting that the ability to sustainably deliver clean, reliable, and abundant electricity to data centres and chip fabs—the new factories of the 21st century—will be the decisive advantage.

While this is the conversation taking place between the relevant stakeholders within Washington, Beijing, Riyadh, Abu Dhabi, Singapore and Kuala Lumpur, to name a few, the conversation in New Delhi remains limited to algorithms and chips, both of which are indeed relevant but need the energy infrastructure to sustain AI development at scale. For New Delhi, this is not just a question of technology but of whether its energy sector can keep pace with its sovereign AI ambitions and the stakes in this race.

The Shift from Compute to Kilowatts

As aforementioned, the global conversation surrounding AI development has moved beyond just focusing on algorithms, graphics processing units (GPUs), and compute; the discourse amongst the nations leading the AI race now concerns one of the sector’s single biggest roadblocks: electricity and water. Both are key constraints that will pose challenges to AI development over the next decade. The conversation surrounding water has been well documented. Electricity, on the other hand, has not been deliberated over as much nationally.

Projections from the International Energy Agency (IEA) forecast that the electricity demand for global data centres will more than double by the end of this decade: that is 945 terawatt-hours (TWh) by 2030, up from around 415 terawatt-hours in just 2024. The same report makes clear that AI is the main engine of that increase, with data-centre demand growing far faster than electricity demand in the rest of the economy. In absolute terms, that is an increase of about 530 TWh in six years, implying annual growth of around 15% — more than four times the pace of demand growth in other sectors.

Meanwhile, global strategy and management consulting firm McKinsey & Company estimates that global data centres will require nearly $7 trillion in capital investment by 2030. Of this, around $5.2 trillion will be needed for AI-ready data centre capacity, with a further $1.3 trillion earmarked for power generation, transformers, substations, transmission and other grid infrastructure.

This is not some far-off challenge for India. The Government of India’s own data reveals that the country’s data-centre capacity has grown from about 375 MW in 2020 to around 1,500 MW by 2025. Official projections and industry estimates now point to a several-fold expansion by 2030, with many forecasts clustering in the 8–10 gigawatt (GW) range. India’s Ministry of Power estimates that electricity demand from data centres alone could reach 13.56 GW by 2031–32. Supplying that kind of round-the-clock load, mainly from renewables, would, in practice, require something closer to 40 GW of installed capacity once intermittency and seasonal variability are factored in.

This electricity demand creates challenges that the country may need to think long and hard about to find the best way forward, as AI and data centres are likely to raise India’s peak power demand by an additional 30 GW over the next five to six years. This is the scale of the challenge. New Delhi’s AI ambitions are converging with a power system already grappling with peak demand, transmission bottlenecks, and the need to integrate massive volumes of renewable energy.

How the World is Wiring for AI

China’s “East Data, West Computing” initiative remains the clearest example of integrated planning. It aims to shift more data-centre and computing loads into western provinces with greater land and energy availability, linked to eastern demand centres through long-distance transmission. New data-centre projects in the national hub regions are also being pushed toward high renewable procurement, with policy and market mechanisms targeting at least 80% renewable sourcing.

The United States still dominates the hyperscale ecosystem.

Goldman Sachs Research notes that the US accounts for well over half of global hyperscale capacity, depending on how capacity is measured. But American AI ambitions are increasingly constrained by the hard realities of interconnection queues, local permitting and grid build-out, especially in markets like Virginia, Texas and Arizona.

On the other hand, the Gulf states are taking a more state-directed approach, albeit on a larger scale than many casual references suggest. For instance, Saudi Arabia’s HUMAIN strategy has been reported to be valued at $77 billion, with data-centre ambitions of about 1.9 GW, while the United Arab Emirates’ (UAE) Stargate campus has been described as a multi-phase project scaling toward 5 GW, with investment figures exceeding $30 billion. The broader lesson is not just ambition, but vulnerability: the drone strikes that disrupted AWS facilities in the UAE and Bahrain earlier this year showed that AI infrastructure is now exposed to physical conflict, not just cyber risk.

Meanwhile, Singapore, facing limits in land, water and grid expansion, has increasingly relied on a wider regional digital corridor with Malaysia and Indonesia for data-centre expansion. India is trying something similarly ambitious in strategic intent, but with less institutional coordination so far: the IndiaAI Mission, India Semiconductor Mission, National Electricity Plan and broader clean-energy goals all point in the same direction, yet remain spread across multiple ministries and agencies.

The 40 GW Problem: Bridging Policy Silos

To give credit where credit is due, New Delhi’s renewable energy push has been seen as remarkable by industry insiders. Just last year, India added close to 48-50 GW of renewable capacity, pushing total non-fossil capacity to about 267 GW by the end of 2025 and crossing the 50% mark in the installed capacity mix ahead of schedule. This was done half a decade ahead of the 2030 Paris target, with solar the main driver of the increase.

The total installed generation capacity was roughly 520 GW by early 2026, according to official government data. Yet the geography of renewable abundance and digital demand does not neatly overlap. Solar-rich and wind-rich states such as Rajasthan and Gujarat are not the same places where the largest data-centre concentrations are emerging, with Mumbai, the National Capital Region (NCR), Hyderabad and Bengaluru among them. While transmission corridors are indeed expanding, it is not necessarily in sync with the pace of data centre growth in the country.

This mismatch is already becoming a policy concern and is not going unnoticed by Indian industry leaders. At the India AI Impact Summit 2026, held at Bharat Mandapam in New Delhi in February, Deloitte South Asia’s chief growth officer Debasish Mishra pointed out that India’s data-centre capacity is expected to expand from around 1.5 GW in 2025 to 8–10 GW by 2030, with power intensity sharply increasing as AI-driven workloads scale. Arunabha Ghosh, chief executive officer of the Council on Energy, Environment and Water, has argued that India’s data-centre growth is being driven by a form of competitive federalism, with states rolling out their own AI and data-centre policies even as a binding national framework continues to evolve.

Given that India’s data-centre capacity has already roughly quadrupled in five years, and that the next wave of AI-driven growth will lift both absolute load and power density, the backdrop is daunting. The takeaway is not just whether enough power exists in aggregate, but whether it is available at the right quality. The policy challenge is to move away from fragmented incentives and siloed decision-making towards a coherent approach that aligns with investment, resource efficiency, and long-term infrastructure planning.

The Economic Survey 2025–26 clearly captures the dilemma. It notes that limited compute, scarce financial resources and constraints around power and water are binding factors for India’s AI ambitions, and warns that investment in AI infrastructure must be weighed against other pressing development needs. The challenge India faces is not limited to capacity alone. Quality too is a limiting factor.

Data centres require near-round-the-clock uptime, yet many industrial regions in the country continue to face voltage fluctuations, harmonics, and outright power outages. These are unacceptable for data centres and fatal for semiconductor fabrication plants. To put this into perspective: a modern leading-edge fab can draw anywhere between 400 and 600 megawatts (MW) of continuous, ultra-clean power, with near-zero tolerance for dips or surges. A simple, serious power outage can destroy wafers in process, contaminate clean rooms and cost exorbitant amounts of money.

This is why semiconductor policy cannot be treated as only a land-and-subsidy question. The India Semiconductor Mission has now approved 12 projects with a cumulative investment of around Rs 1.64 lakh crore, including one fabrication unit, two compound semiconductor fabrication units, and multiple packaging projects. Whether those projects mature into globally competitive manufacturing will depend heavily on whether fab locations are planned as integrated utility zones rather than isolated incentive sites.

Digital Infrastructure as National Security

The investment wave behind AI infrastructure is already enormous. Industry estimates indicate that global technology companies are investing hundreds of billions of dollars annually in AI-capable infrastructure, while IDC projects that global AI infrastructure spending will reach $487 billion in 2026 and exceed $1 trillion by 2029. This takes the frame far beyond a mere industrial policy debate for India. Everything from AI data centres and semiconductor fabs to cloud support platforms and the electricity system that supports them is increasingly being treated as critical infrastructure, in line with global security trends. These are, after all, strategic technologies with massive national security implications.

New Delhi’s SHANTI Act, which paved the way for deeper private-sector participation in the country’s nuclear sector, has also helped strengthen the policy case for small modular reactors as reliable future power sources for energy-intensive digital infrastructure. There are both strategic and practical stakes at play here: take the example of a power disruption at a large AI campus: not only would it interrupt commercial cloud services, but it could also adversely affect the nation’s financial system, cybersecurity operations, defence-linked computing, and public digital services. Digital infrastructure is becoming part of the geopolitical battlespace, as seen earlier this year with the kinetic targeting of data-centre assets in the Gulf.

The challenges that remain unique to India are far sharper than those of richer economies because the baseline is still relatively low. Official government figures put India’s per capita electricity consumption at 1,460 kWh in 2024–25, while newer reporting suggests it may already have risen to about 1,538 kWh. Even so, that remains far below advanced-economy levels, meaning India must scale AI and digital infrastructure while still expanding basic energy access, industrialisation and household demand.

The Illusion of Efficiency

Some analysts argue that AI electricity demand is overstated because hardware and model efficiency are improving quickly. There is truth in that. TSMC has reported a 44% reduction in peak power consumption for EUV lithography tools through a dynamic energy-saving programme.

Researchers at the University of Cambridge have also demonstrated hafnium oxide memristors that could reduce energy use in some AI-relevant computing tasks by up to 70%.

Model economics are improving too, but claims here require care. DeepSeek’s technical and commercial model showed that frontier-adjacent systems can be trained and served at dramatically lower cost than many earlier Western models, though some of the more dramatic published comparisons are difficult to verify precisely. Liquid cooling, better power electronics and improved facility design can also reduce overhead losses and improve efficiency at scale.

But efficiency does not automatically reduce total demand. As compute gets cheaper, users tend to consume more of it, a dynamic often described as the Jevons paradox.

McKinsey and the IEA both suggest that lower unit costs are likely to accelerate adoption faster than they reduce aggregate electricity demand, especially as more capable reasoning models and AI agents increase inference intensity.

For India’s planners, the implication is straightforward: infrastructure should be sized against high-demand scenarios, not optimistic assumptions that efficiency alone will solve the problem.

The 2035 Crossroads: Leadership, Friction, or Failure

One path is for India to become an energy-aligned AI leader. In this scenario, the central government establishes a formal mechanism for joint planning among MeitY, the Ministry of Power, MNRE, water authorities, and state governments, to align on siting, grid access, water, and cooling for AI and semiconductor clusters. Under this model, dedicated clean firm power —including storage, gas peaking, hydro, and eventually SMRs—supports major compute and fab clusters, allowing India to become a serious location for sovereign AI infrastructure and advanced electronics manufacturing.

A second path is AI growth with grid friction. Investment keeps coming, but storage, transmission and power-quality upgrades lag behind demand. Data centres then rely more heavily on diesel and gas backup, while semiconductor projects face lower yields, higher operating costs and investor caution.

The third path is a missed opportunity as aforementioned policy silos persist, infrastructure planning remains fragmented and water and transmission constraints deter large-scale investments. In that scenario, India remains strong in downstream digital applications but weak in domestic compute and semiconductor manufacturing capacity, becoming more dependent on foreign cloud and imported chips.

The most important number in this debate may still be the 40 GW problem. If India eventually needs to supply around 13 GW of continuous data-centre load through predominantly renewable electricity, the associated generation, storage, and transmission build-out will be far larger than the headline load figure suggests.

Other under-reported issues remain equally important: the water intensity of fab projects in stressed regions, the role of state electricity commissions in shaping industrial tariffs and, therefore, data-centre geography, and the emerging security risk posed by the concentration of high-value digital assets. These are not side issues to India’s AI future. They are the terrain on which that future will be built or lost.

India’s success in AI and semiconductors in the next decade may depend less on how many models it trains or how many chips it designs, and more on whether it can build the power system those models and chips will run on. Countries spent the past decade competing for chips. The next decade will be a competition for the electricity that keeps those chips running. Nations that fail to build resilient power systems may discover that their AI ambitions are constrained not by computing power, but by the grid beneath it.

Aritra Banerjee is a senior journalist specialising in the intersection of defence, space, emerging technologies, industrial policy, and the global energy transition. His work explores how technological innovation, strategic industries, and geopolitics are reshaping national security and economic competitiveness in the twenty-first century.

Disclaimer: The views expressed are the author’s own and do not reflect the official policy or position of Financial Express.