India’s race to cool humans and AI machines

On April 28, 2026, on the coast near Visakhapatnam, Google broke ground on a gigawatt-scale AI data centre campus; its largest India investment to date. Three weeks later and roughly 800 km down that same coastline, Chennai residents were losing electricity for 40 minutes to an hour, every night. Delhi and Noida had cuts after dark too. Punjab moved to formal load-shedding.

India’s power ministry projects AI data centres will add 26.3 GW of load by 2032. That’s substantial. It’s also roughly a sixth of the peak load room air conditioners are expected to add over a similar period. Its estimated cooling alone could account for more than 180 GW of Indian peak demand by 2035, more than a third of the national total. Air conditioning is the far larger claimant and gets a fraction of the attention, because data centres come with press releases and air conditioners come with bills.

The real competition is for a scarce product: firm, clean power after dark. A data centre runs flat, around the clock. An air conditioner in a Chennai bedroom runs hardest at 11 at night. Both want power in the hours when India’s 162 GW of solar produce nothing until we get great at long-term energy storage. Until then, every marginal unit will go to whoever bids highest.

A hyperscaler will pay whatever it takes, because electricity is a small fraction of a data centre’s cost. A household paying up to 8% of its income to run a cheap air conditioner is making a different calculation. It means the real question for this decade isn’t whether India can generate enough electricity. It’s the ability to deliver it to various segments at the right volume and costs at the time of need.

The day the grid was tested

At 3:45 pm on May 21, with Delhi at 43.6°C, the grid reached a record peak of 270.82 GW, up from 214.9 GW 50 days earlier. A decade ago, a summer like this meant blackouts across half the country. This time, the grid held.

The generation mix at that moment is worth exploring. It was 62.8% thermal, 22% solar, 5% wind, 5.8% hydro. Then the sun went down, Grid-India logged a peak deficit of roughly 2.57

GW, and Chennai went dark. A country that has largely solved daytime generation is now asked to add a hundred-plus million air conditioners alongside the earlier-mentioned 26.3 GW of servers over the next decade, all producing massive round-the-clock demand.

Not every number in that week’s headlines was about generation, though. In early March, the Indian crude basket touched $113 a barrel after the Strait of Hormuz closed amid regional conflict. India’s crude import dependence hit 88.7% and keeps rising. That’s a different problem, moving on a different clock, and conflating the two is the most common error in writing about Indian energy.

The heat is the hinge

Roughly only 8% of India’s 300-million households own an air conditioner, placing the nation where China sat in the mid-1990s. That gap is exactly where the earlier 180 GW figure comes from: researchers project India will add 130 to 150 million room air conditioners between 2025 and 2035, arriving on the grid at the same time as AI data centres.

The moral case for cooling 300 million homes is not contestable. The question is whether they arrive as cheap, inefficient units or good ones that can contribute to managing India’s energy demands.

And the burden isn’t shared evenly. Poor households everywhere spend up to 8% of their income on cooling electricity, compared with 0.2–2.5% for high-income households. They often run the cheapest, most power-hungry machines in the leakiest homes, on the parts of the grid most likely to fail first. That’s the loop that was visible in this summer’s issues. Heat drives demand, demand forces loss-making distribution companies onto the spot market, and the losses starve the grid of upgrades that would have prevented the next outage.

What the machines want

The AI build is real and fast, and there is a sovereign strategic case. No serious economy in the 2030s can go without domestic compute, and India hosts a tiny fraction of the world’s data centre capacity relative to the data it generates.

But the load itself is harder to serve than the announcements suggest. First, its shape: data centre loads are flat and round-the-clock, indifferent to sunrise, and it falls exactly on the profile solar serves worst.

There’s an optimistic take as well. Data centres are creditworthy, long-lived buyers of energy. India’s distribution companies, carrying lakhs of crores of accumulated losses, are terrible counterparties for investing in new capacity. A hyperscaler is an excellent one, and its demand could become the anchor tenancy that finances India’s first real tranche of firm renewables-plus-storage, with households also inheriting the cost benefits.

There’s a second constraint on data centres that has nothing to do with the grid. The heat they generate on-site. Rack densities in AI clusters have risen to the point where conventional air cooling can no longer keep chips within operating temperature, and the industry is shifting towards liquid cooling delivered at the chip and rack level. India needs its own answers rather than depend on a fully imported solution.

The wider bet

Cooling and compute will decide the next five years, but they sit atop a larger set of choices India must get right over the next 15. Solar is the genuine success story. Capacity has grown roughly 57-fold since 2014. But it stops producing exactly when cooling demand peaks. That makes long-duration storage the most important gap in the system. Coal is not going away and is a strategic insurance against this summer’s kind of volatility. Nuclear is finally opening to private capital, though targets remain distant. Green hydrogen has so far been mostly aspiration. None of these gaps close from a policy document alone—which is where a newer layer of the market comes in.

That newer layer is private innovation, showing up in every category, not just cooling. In long-duration storage, young Indian companies are attempting chemistries built to escape lithium’s import dependence: Sthyr Energy is building a zinc-based battery designed to hold power for weeks, aimed at the seasonal gap solar cannot fill. In rooftop solar, a handful of companies are trying to turn a fragmented installation business into a standardised, serviceable consumer product.

A third of installed Indian air conditioners stop cooling properly on the hottest days, because they’re rated against test conditions imported from milder climates—India’s official efficiency label is measured up to 43°C, and Delhi hit 43.6°C the day the grid set its record, so at this level ACs draw more power for less cooling. Entrepreneurs are noticing, like Optimist, a startup, which has launched a “built for India” air conditioner with the highest efficiency rating on the Indian household market. They are working towards mass-market 1-tonne and 2-tonne variants in the coming months with significantly lower energy consumption than what is available today.

On the data-centre side of the same problem, KuhlTherm, an Ahmedabad-based startup founded in 2025, builds precision liquid-cooling hardware and the software to run them. They are engineered specifically for the retrofit-heavy, price-sensitive conditions of Indian data centres rather than imported wholesale from denser, cooler markets.

The honest answer to what fixes our cooling needs is neither government alone nor markets alone. Government must do what only it can: raise minimum standards for cooling, fund and de-risk storage at a scale no company can finance alone, keep distribution companies solvent enough to invest, and decide whether AI data centre contracts can be incentives to invest in long-term green capacity.

Private innovation and investors must do what government structurally cannot. Take risks on unproven technology and bet on smart young technologists to find solutions. Neither moves the needle alone.

Google’s Visakhapatnam campus is due to start operating around 2028. The same year, researchers warn India’s cooling-driven shortages could begin. Two claims on the same grid, arriving in the same summer. Whether India is ready for both depends less on any one company or policy than on whether the state and its innovators actually move together.

(The author is managing director, Arkam Ventures. He is also the co-author of Winning Middle India: The Story of India's New-age Entrepreneurs. Views are personal.)