A boy in Uttar Pradesh sank a new handpump and found the water table lower than his grandfather remembered. In another village, mothers line up at a municipal tanker while a gleaming campus of servers hums miles away.
There’s nothing mystical about these juxtapositions. They are the practical consequences of an industry that aims to turn India into a global hub for artificial intelligence: vast campuses of servers, hungry for electricity and water, springing up across states where communities already feel pressure on basic resources.
Dollars, megawatts and a national push
Big-name cloud providers and local conglomerates are ploughing billions into Indian data‑centre capacity. The country’s installed capacity today is measured in low single‑digit gigawatts, but forecasts — and company commitments — point toward an explosion: analysts and industry groups expect several more gigawatts by 2030 and double‑digit gigawatts by 2035 in some projections.
Policy plays a role. Reports from consultancies suggest India could leapfrog competitors by offering novel models such as “data embassies” and purpose‑built “data cities” to attract sovereign and corporate workloads. Those ideas promise tax clarity, sovereign hosting and concentrated ecosystems — attractive to governments and enterprises that want scale and security.
But incentives that bring capital do not erase physics. Servers need power to compute and water to cool. AI workloads are not like a Netflix stream; training dense models can draw thousands of megawatt‑hours and dramatically increase a facility’s cooling needs. Even routine, everyday cloud operations add up when multiplied across millions of queries and millions of racks.
The environmental ledger: energy, water, waste
Three pressures are already obvious.
First, energy. India still generates the majority of electricity from fossil fuels. Unless new data centers secure large, reliable blocks of renewables and storage, each new facility risks locking in additional carbon emissions for years.
Second, water. Many modern cooling systems — especially for high‑density AI racks — use evaporative or water‑based chilling. A handful of large campuses can consume millions of litres daily; in Bengaluru, for example, citywide data centres already draw millions of litres each day while neighborhoods ration supply. Put a cluster of AI factories in a water‑stressed district and you produce local contention faster than any regulatory debate can catch up.
Third, e‑waste. High‑performance servers and accelerators have short commercial lifespans. When hardware spins through cycles every two to three years, piles of circuit boards and heat sinks accumulate. Existing collection and recycling rules in India are improving but remain focused on consumer electronics; data‑centre specific flows remain poorly tracked.
These are not speculative worries. Activists and villagers in Telugu states and in parts of Maharashtra have raised objections about wells, land use and the opacity around resource deals. The stories echo across state lines: flourishing digital infrastructure at the edges of communities that see few of the economic upside but bear many of the environmental costs.
Efficiency as industrial policy
There’s a practical response at hand: design and standards. India’s data‑centre operators can aim for higher benchmarks — lower PUE (power usage effectiveness), aggressive on‑site or contracted renewables, and advanced cooling such as liquid immersion — to squeeze more computing out of every litre of water and every megawatt.
Materials and engineering choices matter too. Some operators point to higher copper intensity and better thermal management as ways to improve electrical efficiency and reliability. Others are piloting liquid cooling and closed‑loop systems that dramatically reduce water use compared with open evaporative towers.
The goal is to avoid simply exporting old, dirty patterns of growth into a new sector. If AI workloads are sited with foresight — matching high‑density compute to renewable‑rich grids, or clustering heavy compute where water is abundant and managed — the carbon and water footprints shrink. That, however, requires planning: hard standards, transparent disclosures about water and energy use, and enforcement capacity that’s currently patchy.
Governance gaps: disclosure, assessment and community voice
Right now India’s regulatory framework for environmental reviews was built for conventional industry. Data centres frequently slip through the cracks — treated as construction projects rather than as continuous, thirsty industrial operations with unique risks. There is limited mandatory disclosure of water and energy metrics, and environmental impact assessments often do not capture the intensity or the cumulative effect of multiple campuses in a region.
Public participation is uneven. By the time communities learn about a planned campus, contracts may be signed and foundations poured. That breeds mistrust and legal challenges, even as the economy gains jobs and ancillary activity.
Regulators and policymakers facing these tradeoffs have a menu of options: mandate resource‑use disclosures, tie incentives to verifiable renewable procurement and low PUE, and require lifecycle plans for e‑waste tied to reliable take‑back systems. International best practice also shows the value of regional planning: cluster facilities where the grid, water and workforce can sustainably support them.
New models — and hard choices
Proposals like data cities and sovereign hosting can concentrate infrastructure and give governments levers to shape sustainability outcomes, but they are not silver bullets. They require legal frameworks for tax neutrality, diplomatic protections and cybersecurity guarantees. Done well, they can create managed zones with high environmental standards; done poorly, they could become tax shelters that externalize local costs.
Technology can help in other ways. Operators experimenting with advanced cooling and higher rack densities are part of a global learning curve; the next wave of hardware — more efficient processors and purpose‑built AI accelerators — will change the economics and the resource profile. Even software advances matter: model optimizations and smarter scheduling can reduce the number of training cycles and therefore the energy consumed.
It’s worth noting that the push for compute has spurred creative thinking beyond planet Earth: companies are exploring distributed and novel hosting ideas, from floating platforms to orbital concepts. Those projects remain speculative, but they underline a central truth — compute wants to go where cost, security and resource availability align. (For a read on one of these high‑ambition experiments, see Google’s Project Suncatcher.)[/news/google-suncatcher-space-datacenters]
Research and commercial models for AI — like some of the new image and generative systems — also affect infrastructure demand. As models such as Microsoft’s MAI‑Image‑1 and others proliferate, their training and serving requirements will ripple through the data‑centre market and resource planning in ways policymakers should factor into decisions today. [/news/microsoft-mai-image-1]
A practical horizon
India’s ambition to become a digital and AI powerhouse is legitimate and strategically important. The path to that future doesn’t run only through tax incentives and land allotments; it must run through engineering, standards and accountable governance.
There will be tradeoffs. Some districts will gain jobs and new supply chains; others will wrestle with water and land stress. The quality of the outcome will depend on choices made now: whether contracts require green power, whether cooling tech is chosen to conserve water, whether communities have a voice in planning, and whether e‑waste is treated as a pipeline to value rather than a burial problem.
If India gets those pieces right, it can build digital capacity that’s both competitive and sustainable. If it doesn’t, the country risks letting the next industrial revolution bloom at the expense of the people who live beside the servers.