AI data centres need power all night, and Indian solar stops at sunset

Nikhil Singh
3 Min Read
Data center infrastructure in the United States. Photo: DOE / National Renewable Energy Laboratory (NREL), licensed Public domain via Wikimedia Commons.

Mumbai – Indian data centre electricity consumption is forecast to rise from about 10 terawatt hours in 2025 to 191 by 2040, a near twenty-fold increase, and artificial intelligence workloads are driving most of it.

Operational capacity is projected to grow from 2.2 GW in 2025 to 12 GW by 2030. Capacity dedicated to artificial intelligence is expected to rise roughly twenty-four fold, from 275 MW to about 6,546 MW. In the first half of 2026, absorption reached 101 MW, with pre-committed hyperscale capacity accounting for 82 per cent of it.

The load shape is the problem

A data centre runs flat, all day and all night. Solar generates in a daytime window. These do not match, and a data centre cannot reduce output to follow supply the way a factory can shift a shift.

So a facility claiming renewable supply on an annual netting basis is, in practice, drawing coal-fired power overnight and buying enough daytime renewable generation to balance the books. That is legitimate under current accounting and increasingly unpersuasive to buyers moving toward hourly matching.

What the sector is doing about it

Power demand from these facilities is projected to add around 26.3 GW by 2031-32, and renewables are expected to meet much of it. Roughly a quarter of Indian data centres hold green certification, with 30 to 40 per cent expected by 2030.

Firm low-carbon supply is the missing piece, which is part of why industrial groups are showing interest in small modular nuclear reactors. A data centre is precisely the customer that values round-the-clock output over the lowest headline tariff.

Water deserves more scrutiny than it gets

Data centre cooling consumes water, and Indian ambient temperatures make cooling loads higher than in temperate markets. Disclosure of water usage effectiveness is far less common than disclosure of power usage effectiveness, and in a water-stressed country that is the wrong way round.

Source: India data centre market analysis for 2026, and power demand projections for artificial intelligence workloads to 2031-32.

Share This Article
Follow:
Nikhil Singh is a Senior Correspondent at The ESG India, covering ESG, CSR, sustainability and government policy across India. He reports on corporate accountability, climate action and social impact.