OPINION | AI gold rush could anchor India’s energy transition if the grid keeps up

India is adding renewable energy faster than it can build the wires to carry it. In the first quarter of 2026 alone, roughly 300GWh of clean power was generated and then wasted, curtailed because the transmission system could not move it. That gap is the quiet constraint on the country's entire clean-power ambition, and a new kind of electricity customer is about to test it harder than anything before—the AI data centre. For two decades, data centres were an unremarkable load, under 1 per cent of national consumption. The grid was planned on that assumption. The assumption no longer holds.

A conventional server rack draws 5 to 10 kilowatts; a rack that trains AI models draws 20 to 30, and the densest clusters are climbing toward 100. A single cabinet is drawing what a small neighbourhood once did, every hour of every day. India's installed data-centre capacity is projected to reach 8 to 10 gigawatts by 2030, with announced AI-linked investments running past USD 200 billion over the coming decade.

The stakes here are not only an energy story. Every major economy is now competing to host AI infrastructure, and India is one of the few large economies with the resources, the capital and the corporate buyers to provide clean energy at speed and scale.

So, India needs to capture this build-out. The obvious response is that AI needs power, so we should build more renewables. That is true, but misses the wider implication. The more important point is financial, and it runs the other way. Handled correctly, this demand can lower the cost of building clean energy across the whole sector, not only for the data centres. But that is contingent on two things standing between a renewable plant and the customer that wants its output – transmission and regulation.

Historically, the primary buyer of renewable power has been the state distribution company. To start with, the discoms have been financially stressed. The stress now shows up in the contracts themselves. Around 40 GW of auctioned renewable capacity is sitting without signed PPAs, as the discoms are hesitant to commit. When a developer signs a 25-year contract with a buyer holding weak credit, lenders demand higher returns and shorter tenures. That extra cost gets baked directly into the tariffs.

This is where AI demand alters the baseline. The global tech giants, or hyperscalers, building this capacity alongside large domestic conglomerates carry investment-grade balance sheets. That lets a renewable project raise debt more effectively, which lowers the cost of the power it produces. Once this is scaled up across enough projects, the effect becomes structural.AI demand can make the entire business of building renewables look safer to lenders, and safer means cheaper to finance. That is the real opportunity, and it is larger than the data centres that triggered it. Their public mandates for carbon-free operations make this clean power a contractual requirement rather than just a corporate preference. The shift is already there. Indian data centres draw an estimated 30 to 35 per cent of their power from renewables through power purchase agreements, with leading operators well above that, and players such as AdaniConneX and Nxtra by Airtel have committed to fully renewable operations by 2030. Those commitments are turning into contracts, with Amazon signing hybrid solar-wind agreements of 600 MW, and Nxtra scaling up to a green-power deal of more than 200 MW.

There is, however, a catch. A facility that runs every hour cannot rely on power that wanes at the whims of the weather gods. It needs round-the-clock clean power bundled with storage, which in this decade means batteries, where India is starting from a low base. Firm clean power costs more than plain solar, but AI operators, for whom electricity can be close to half of operating cost, are among the few buyers willing to pay for reliability.

So the demand is real, valuable, and durable. S&P Global puts the additional renewable capacity needed to meet it at 15 to 30 GW within a few years. Whether it can be served comes down to the grid, and here the picture is murkier. A solar or wind project is built in 18 to 24 months; the transmission line to carry its power takes 36 to 60 months, slowed down by land acquisition, rights of way and clearances. Build-out has been running at about 80 per cent of target. Generation races ahead, the wires lag, and the gap shows up as renewable plants standing built and idle, waiting for a connection. The problem has a geography, too. The best renewable resources sit in the interior, while 90 per cent of data centres cluster in four major coastal and urban hubs, requiring massive high-voltage corridors to bridge the gap. Mumbai is the sharpest case. Its historically islanded grid failed in 2020 and 2022. The Kharghar-Vikhroli 400 KV line came up in late 2023, adding about 1000 MW of import capacity, and the Mumbai Urja Marg inter-state line was brought up in 2024 for roughly 2000 MW more.

Regulation adds a quieter obstacle. Data centres typically keep a full grid connection from the utility for reliability while buying much of their actual power directly from private generators. The utility is left holding contracted capacity it does not fully sell, and recovers those fixed costs through surcharges on direct purchase. Layered on top, the discount for moving renewable power across state lines is being phased out, just as the geography of this build-out makes long cross-state flows unavoidable. None of these decisions is irrational by itself. Taken together, though, they make the country's own stated goal harder to reach.

That difficulty shows up in one of two ways. Renewables get built but cannot reach the data centres, and AI campuses run on fossil fuels while they wait. Or, the clusters draw so heavily on local networks that they crowd out homes and industry, and the backlash lands on the renewable sector rather than the grid. Avoiding both is mostly a matter of sequencing. Transmission has to come first, because it has been planned reactively for years, and that cannot serve a load being announced faster than it is built. The grid for the data-centre clusters of 2030 must be planned now, ahead of firm demand, accepting that some corridors run under-used for a few years. An under-utilised line is a small cost. A stranded AI campus is not. Beyond that, incentives should steer new campuses toward the renewable-rich states where possible, and round-the-clock clean power, backed by battery storage, must become a standard product.

The timing comes down to simple arithmetic. A transmission line takes three to five years, and data-centre demand reaches its 2030 scale in roughly that window. The grid to carry it has to be sanctioned and under construction within two to three years, or it will not be there when the load arrives. And India's target of 500 GW of non-fossil capacity by 2030 assumes that it will be.

India is short of none of the ingredients. The capital is committed, the demand is real and creditworthy, the resource is in the ground. What we must supply is delivery, which means the wires, the storage and the regulatory steadiness that turn an announcement into contracted clean megawatts. Get that right, and the AI build-out becomes the anchor that pulls India's energy transition forward and lowers the cost of clean power for everyone along the way. Get it wrong, and the same demand is met by diesel and gas, and a rare opportunity becomes a missed one. The gold rush is real; whether India captures it depends on whether the grid can keep up.

The writer is General Manager - Business Development & Regulatory, AMPIN Energy Transition.

The opinions expressed in this article are those of the author and do not purport to reflect the opinions or views of THE WEEK.