AI infrastructure is not like roads or bridges. Technology is changing rapidly, aging rapidly, consuming enormous amounts of energy, and is highly dependent on the changing priorities of private technology companies. What seems simple on paper is much more difficult to manage in practice.
India has the advantage of seeing others leading the way. This is important, because once projects reach a certain size, choices become more difficult to reverse. The real limitation today is not talent or ideas. It involves constant power, cooling, water, stable networks and available land. These are the pressure points. Large computer centers are not implicitly in the background; they rely heavily on the country’s physical systems.
A single modern facility can use as much electricity as a small city, and it needs that supply without a break. It does not turn off during peak demand or summer shortages. The cooling runs day and night and often uses local water. Once built, the ripple effects spread silently. Transmission lines are strengthened. Older power plants stay open longer. Water is diverted. Zoning rules change. None of it feels shocking at this point. It happens step by step.
Lessons from abroad
Countries that adopted early are now seeing the full costs of those choices. What started as a digital bet has steadily changed networks, water systems and land use – with physical consequences long after the headlines faded.
The US offers a clear warning signal. In 2023, data centers used about 176 terawatt hours of electricity – about 4.4 percent of the country’s total demand. That share is expected to rise sharply, possibly reaching 10 to 12 percent within a few years, as heavy computing loads increase.
The real problem, however, is not the national average. It is where this demand is concentrated. In Northern Virginia, home to the world’s largest cluster of data centers, these facilities already consume more than a quarter of the region’s electricity. To keep up, utilities have rushed to expand the power grid, postponed the closure of fossil fuel plants and reworked the way risk is priced.
The impact is already visible. Household electricity bills are rising faster than the national average, and network planning now revolves around the demands of large computing facilities rather than homes or small businesses. What started as a technology wave has grown into a broader energy and pricing challenge.
Local governments first welcomed these facilities because of the tax revenue. Over time, a different picture emerged: upgrades to the electricity grid, substations and transmission costs were largely shared across the system, while job creation remained modest compared to energy consumption. The infrastructure remained. The leverage disappeared.
Ireland offers a parallel lesson. By 2022, data centers accounted for more than 20 percent of the country’s electricity demand, heavily concentrated around Dublin. Grid authorities warned that further expansion would jeopardize stability and climate goals. The answer was practical and not ideological: new grid connections were interrupted in key areas. What had been sold as digital growth was redefined as a hard infrastructure limit.
Water tells the same story, only later and sharper. In The Dalles, Oregon, Google’s facilities sometimes used nearly 30 percent of the city’s water in an area prone to drought. Use grew quietly under industrial permits, and public concern only emerged when the scarcity became apparent – after contracts and infrastructure had already been committed.
The lesson is not that data centers are harmful. Its costs accumulate over time and spread widely, while its benefits emerge early and are picked up by a few.
The dynamics in India
India must now ask itself what this dynamic would look like at home. Electricity in India is not a simple market commodity. It is a social agreement. Distribution companies are under financial pressure, tariffs are being crucified and the power supply is deeply political: households, farmers and small businesses are being held in balance. Adding large, always-on data centers to this system not only increases demand. It changes who gets priority.
During heat waves, fuel shocks or mains stress, difficult choices are inevitable. Once large computing facilities are labeled “strategic infrastructure,” their access to power is rarely questioned. The adjustment costs shift elsewhere – through outages, higher rates or reduced reliability for regular users – often without public debate.
Water is an even bigger concern. Many Indian cities are already experiencing seasonal shortages and groundwater depletion is widespread. Large data centers can consume water on the scale of thousands of homes, but this often remains invisible until the shortage becomes acute. In the context of India, the allocation of water to the global computing workload is not just a technical decision. It is a social and political one.
Fiscal policy adds another layer. In the US and Europe, governments offered tax breaks, energy rebates and infrastructure support to attract data centers. The first investment figures looked impressive. Over time, many regions found that government costs persisted while revenues declined – especially given the low employment these facilities generate. Indian states, many of which are already under fiscal pressure, will face similar arithmetic problems.
So far this seems like an infrastructure story. But the deeper problem is where value accumulates.
When global companies operate large AI systems in India, India provides electricity, water, land, network stability and regulatory certainty. What it does not automatically obtain is control over the intelligence produced. The models, systems and optimization logic remain proprietary and mobile. The infrastructure remains. The intelligence can leave.
This imbalance is not clearly reflected in the GDP figures. India could host large amounts of AI activity while capturing only a small portion of its strategic value. India has seen this pattern before. The telecom revolution brought cheap connectivity at scale, but the power of the platform – data control and pricing – was consolidated elsewhere. India became essential to global platforms without owning them.
AI infrastructure threatens to repeat this mistake at a deeper level. This time the subsidy is not just about market access. It concerns energy, water, land and grid stability.
There is also a strategic dimension. Large computer clusters are inherently dual-use. The same systems that power commercial applications can also be used for surveillance, cyber operations or influence campaigns. Countries that respond seriously are not rejecting AI infrastructure; they set conditions around transparency, supervision and national capacity building.
Set the conditions
India still has an advantage that others lack: time. The US and Ireland learned these lessons after losing leverage. India can still define the terms – before the scale makes these choices irreversible.
India does not need to slow down its ambitions, but it must price them fairly. Large data centers should be treated as strategic infrastructure, not routine real estate. Energy and water costs must be transparent and limited, incentives must be time-bound and government subsidies must be linked to domestic capacity, and not to permanent concessions. Above all, hosting computations must be accompanied by the possession of skills, systems and models.
These safeguards do not block growth. They decide who benefits from it. Set the conditions early, and infrastructure builds national strength. Delay them, and the costs will still come – quietly and without any influence.
The writer is a physicist at the University of North Carolina at Chapel Hill, USA
Published on February 21, 2026
#scaling #data #centers