Data Centres, People and the Planet: India’s AI Boom Needs a Grid Strategy

In the AI economy, electricity is fast becoming the currency that decides who gets to compete at all, and how it is generated, stored, shared and accounted for will shape whether India’s data centres strengthen or strain the systems millions depend on.

October 21, 2026. By News Bureau

Tags:

articles

Tanya Singhal

Global Energy Alliance for People and Planet

AI economy

AI infrastructure

Economic Survey 2025-26

Discoms

MERC

India is entering an era of rapid data centre expansion, which can bring investment, jobs, and new AI infrastructure. But it will also require large amounts of electricity, placing new demands on an already complex power system. The real question is how India should build them in a way that works for people, the planet, and the power system.

The United States offers an early example of what happens when demand grows faster than infrastructure. In July 2024, xAI brought its Colossus AI data centre online in Memphis in just 122 days. But this timeline was not sufficient for the municipal utility to provide the 150 megawatts the facility needed. So xAI turned to onsite generation, installing dozens of gas turbines. This led to disputes over permits, pollution, and noise and a larger debate about what happens when large data centres try to work around the public grid.

India has a chance to prepare before that scale arrives. The Economic Survey 2025-26 makes a similar point, advocating for prioritising decentralised, application-driven systems over capital-intensive frontier models to avoid fragile dependencies in artificial intelligence. The central idea is to take bottom-up, multiple sector-specific approaches grounded in open and interoperable systems to promote collaboration and shared innovation.

The answer is not to slow India’s digital ambitions. It is to make sure that large data centres are part of the grid strategy from the outset. And now with states already competing to attract larger AI and data centre investments, this is more urgent than ever. For instance, Andhra Pradesh is backing Google’s planned USD 15 billion AI Hub in Visakhapatnam, including a 1 GW hyperscale data centre campus. The state has allocated around 600 acres for the project. The plan leans heavily on renewable power. The harder question is how that renewable-heavy model will deliver firm, round-the-clock power at gigawatt scale.

Vizag is not an isolated case. And the renewable push behind it is not purely voluntary. India’s Renewable Consumption Obligation, revised by the Ministry of Power in 2025, requires large captive and open-access consumers to source a rising share of their power from renewables. Any gigawatt data centre qualifies. In principle, this keeps new data centres off the grid as pure, heavy consumers. But reliability has a cost: a data centre built to stay online through a cloudy day or a still afternoon needs more solar and wind capacity than it typically uses. The surplus is the harder question.

The obvious answer is to send that surplus back into the grid, turning a data centre from a pure source of demand into a source of flexibility as well. But a distribution network built for one-way delivery cannot simply absorb power flowing the other way. Reverse power flow into a network built for one-way delivery is already straining parts of India, serious enough that Rajasthan's DISCOMs recently restricted new rooftop-solar connections to manage the resulting voltage instability. A gigawatt data centre exporting power unpredictably would raise the same risk at a far greater scale. And absorbing that surplus safely depends on something more basic than a mandate: real-time visibility over the entire network. Without knowing, feeder by feeder, where power is flowing and where capacity exists, a DISCOM cannot tell whether it is safe to take the surplus in, or where to channel it once it does. Get this right, and that surplus becomes a genuine grid asset, improving system utilisation and lowering costs over time. Get it wrong, and it becomes instability nobody asked for.

This also raises the question of whether India’s power distribution companies (DISCOMs) have a concrete answer, and their finances suggest real caution is warranted. The sector still carried INR 6.47 lakh crore in accumulated losses and 7.26 lakh crore in borrowings in FY 2024-25, a legacy of years of underpricing and weak collections that a single good year does not erase. DISCOMs did record a modest collective profit that same year, the first in over a decade and a small fall in AT& C losses, but that gain amounts to a fraction of the debt still on their books. Utilities carrying that much unresolved debt cannot absorb a new gigawatt of firm power, in either direction, without significant planning and investment.

Maharashtra offers an early glimpse of this challenge. A dispute over how renewable power could be banked and used led Maharashtra Electricity Regulatory Commission (MERC) to restrict banked solar-hour energy to the same hours. Renewable energy stakeholders challenged the move, and the Bombay High Court intervened. The case shows how difficult it is becoming for DISCOMs to manage a grid where customers generate, store, and use power at different times. A rooftop solar consumer is a small version of this challenge. A gigawatt-scale data centre will take it much further. DISCOMs will need to forecast large new loads, plan network upgrades, manage renewable power, and recover the cost of providing reliable supply. The question is whether they are ready for that scale.

India’s advantage lies in being a late mover. Gigawatt hyperscale campuses can still fit into that advantage, but only if three things happen before the next one breaks ground: First, DISCOM finances and capabilities need to be strengthened on their own terms, not treated as an afterthought to industrial policy. Second, large new loads should come with clear flexibility and storage commitments as part of land and power approvals, not as a courtesy. Third, states should decide, deliberately, whether gigawatt campuses will operate largely as standalone power ecosystems or become genuine partners in building the grid flexibility to absorb what they generate.

India’s AI ambitions and its development targets do not have to compete. But that outcome is not automatic. In the AI economy, electricity is fast becoming the currency that decides who gets to compete at all, and how it is generated, stored, shared and accounted for will shape whether India’s data centres strengthen or strain the systems millions depend on. India’s advantage is that it is building at a moment when the risks are already visible elsewhere. The opportunity is to ensure that the infrastructure powering the AI economy also strengthens the grid, conserves resources, and creates value beyond the walls of the data centre itself.

- Tanya Singhal, Vice President and Country Head - India at the Global Energy Alliance for People and Planet