India’s AI data centre moment: What we give up to host intelligence
India’s rush toward AI data centres is usually told as a tidy technology story. Compute is running short, global demand for AI is surging, and India — with its land and talent —appears well placed to host the next wave of digital infrastructure. It is an easy story to believe, and a comforting one.
But pause for a moment and ask a different question: is India really making a technology choice here, or is it committing itself to something far more durable? Because once data centres arrive, they don’t just run software — they redraw cities, reorder power grids, lock in public spending, and quietly shift where strategic control sits, long after the technology headlines have moved on. Across the world, AI has entered a new phase. It is no longer constrained primarily in the wall of algorithms or talent.
It is constrained by electricity, cooling, water, and uninterrupted physical capacity. Intelligence at scale has become a fixed-cost system. This is why data centres, not models, have emerged as the real bottleneck in the global AI economy.
Public data already shows how fastly this fixed-cost structure is expanding. In the United States, data centres consumed roughly 4–5% of national electricity in 2023, according to federal energy estimates. Multiple grid operators project that AI-driven demand could push this figure toward 10 per cent within the next few years, concentrated in specific regions rather than spread evenly.
The concentration is what matters. In regions where data centres cluster, they become the dominant load on the grid, forcing utilities to accelerate capital expenditure, delay power-plant retirements, and redesign transmission around a small number of always-on customers. These are not marginal adjustments; they are structural commitments.
Europe has reached similar conclusions. In several countries, grid operators have warned that continued data-centre expansion is colliding with housing supply, industrial electrification, and decarbonisation goals. The response has not been ideological resistance to AI, but administrative slowdown — a recognition that infrastructure has limits.Water tells the same story. Academic studies and municipal disclosures show that large data centres can consume millions of litres of water per day, either directly for cooling or indirectly through electricity generation. What makes this politically volatile is not the absolute volume, but the opacity. Water use is often approved as industrial consumption and becomes visible only when scarcity intensifies.
The fiscal dimension is equally under-discussed. Independent analyses in the U.S. and Europe show that while data centres generate impressive headline investment numbers, their employment density is extremely low, and their long-term public costs — grid upgrades, substations, transmission lines, water infrastructure — are often socialised. Tax incentives are front-loaded; infrastructure costs persist. These are not failures of governance. They are failures of timing. Infrastructure decisions were taken incrementally, before cumulative effects were visible.
India now stands at the beginning of this curve. Publicly available estimates suggest that India’s data centre footprint could triple or quadruple by the end of this decade, requiring tens of millions of square feet of new built-up area. This expansion will not be evenly distributed. It will cluster around a handful of metros and coastal corridors due to connectivity, power availability, and network effects. At that scale, AI infrastructure stops behaving like a sector and starts behaving like urban form. Data centres are land-intensive but people-light. A facility occupying dozens of acres may employ only a few dozen permanent workers. Yet it permanently locks in land, power corridors, and water access. That land is no longer available for housing, logistics, light manufacturing, or civic use. This creates a quiet but consequential trade-off: cities absorb opportunity costs, while the economic upside is often diffuse and invisible to local populations. Electricity makes this trade-off sharper. India’s power system is not a surplus system waiting for demand. It is a negotiated system.
Distribution companies are financially strained. Tariffs are cross-subsidised. Power supply is politically sensitive, tied to households, farmers, and small businesses.
Introducing large, always-on AI loads into this system does not simply add demand. It reorders priority. During heatwaves, fuel price shocks, or grid stress, electricity is rationed — explicitly or implicitly. Once AI data centres are categorised as strategic infrastructure, their protection becomes automatic. The adjustment burden shifts elsewhere, often onto users with the least bargaining power. This is not a hypothetical concern. It is how fixed-cost systems behave when they enter variable-price economies.
Water poses an even more delicate challenge. Many Indian cities already face seasonal shortages and declining groundwater tables. Large AI facilities require continuous cooling, often running year-round. At first, this demand is invisible, approved through technical permits. When scarcity becomes acute, the conflict surfaces — but by then, the infrastructure is immovable. Fiscal policy completes the picture. States competing for AI investment are offering land concessions, electricity discounts, capital subsidies, and tax exemptions. Individually, these incentives appear rational. Collectively, they risk turning states into long-term infrastructure underwriters for globally mobile firms.
This matters because data centres, unlike factories, do not generate thick local supply chains or large employment multipliers. Once built, their economic footprint is narrow, while their infrastructural footprint is deep. Up to this point, the story is about cities, grids, and budgets. But the deeper issue is value capture.
When global firms train and operate AI models in India, India supplies electricity, water, land, and regulatory stability. What it does not automatically receive is control over the intelligence being produced. The models, architectures, and optimisation logic remain proprietary and portable. The infrastructure remains fixed. This creates a structural asymmetry that does not show up in GDP numbers. India may host large volumes of AI activity while capturing a limited share of AI’s strategic value.
India has seen a version of this dynamic before. The telecom revolution delivered cheap connectivity and universal access. Infrastructure expanded rapidly. But platform power — data ownership, pricing leverage, and intelligence — consolidated elsewhere. India became indispensable to global platforms without owning them. AI infrastructure risks repeating this pattern at a deeper level. This time, the subsidy is not just spectrum or market access. It is land, electricity, water, and grid stability — the deepest layers of national infrastructure. There is also a strategic dimension that becomes unavoidable at scale. AI data centres are inherently dual-use. The same compute clusters that train language models can simulate financial systems, optimise cyber operations, or support military applications. The distinction between civilian and strategic compute is eroding.
Countries that recognise this are not rejecting AI infrastructure. They are conditioning it — through transparency requirements, oversight of large training runs, and alignment with national capability goals.
India has not yet articulated such a framework. That does not mean it should slow down. It means it should define terms before path dependence sets in. Because once land is allocated, grids reinforced, and water rights assigned, reversing course becomes politically and economically costly.
Artificial intelligence is physically expensive because intelligence at scale is physically expensive. AI systems convert enormous quantities of electricity into computation and heat. Someone must absorb that burden. Countries that host AI absorb the physical costs. Countries that control AI extract prediction, optimisation, and power. India still has the advantage of foresight. Many countries are now trying to renegotiate AI infrastructure decisions made when scale was underestimated. India can design its approach before those costs harden. The policy implication is not to slow ambition, but to price it honestly.
Large-scale AI data centres should be treated as strategic, city-shaping infrastructure, not routine real-estate projects. Electricity tariffs must transparently reflect grid-upgrade costs. Water use must be disclosed and benchmarked against best-in-class efficiency. Fiscal incentives should be time-bound and conditional on domestic capability creation. And hosting compute must be linked to ownership — of research, systems expertise, and indigenous AI models.
These precautions are not anti-AI. They are pro-sovereignty. If India sets these terms early, AI infrastructure can strengthen national capacity and autonomy. If it does not, the costs will still arrive — later, quietly, and without leverage. That is the real choice embedded in India’s AI moment.
Author is a theoretical physicist at the University of North Carolina at Chapel Hill, US, and the author of the forthcoming book The Last Equation Before Silence; views are personal
