AI, data centres and India's grid math

In recent years, data centres have moved from operating in the shadows to gaining public attention, particularly due to the rise of artificial intelligence (AI). Four critical questions need to be addressed: Do data centres in India consume more water than similar industrial or commercial facilities? Would a shift to lower water usage affect the reliance on electricity, and what measures are being taken for efficiency?

If they rely more on electricity than the most efficient facilities internationally, usually compared to Nordic countries with cold climates that make cooling almost free, how can this gap be addressed? Lastly, does this infrastructure impact surrounding communities through grid capacity, availability, or cost?

Water usage for data centres is roughly 1.8-1.9 litres per kilowatt-hour (kWh) of electricity consumed, which equates to approximately 15-16 million litres per megawatt (MW) annually. In comparison, a closed-loop air-cooled system, which uses fans and refrigerant loops to reject heat instead of evaporating water, consumes less than 0.2 litre per kWh.

India's nationwide operation currently uses 1.1 million litres per MW, which is about 7% of the technology's water consumption. This finding raises concerns about the industry's water usage.

A 50 MW data centre spans around 4 lakh square feet. A commercial facility with a similar area would require at least 4,000 people to operate, consuming at least 65.7 million litres of water annually. This comparison highlights that a 50 MW data centre uses less water than an equivalent commercial facility employing people.

If heat transfer is managed using mechanical refrigeration, it could potentially consume more electricity. Running compressors harder to save water might also require additional electricity. Ensuring water savings while increasing energy consumption would not be a victory for sustainability. This is where grey and recycled water use can play a significant role in improving efficiency.

A hybrid cooling system, combining closed-loop air-cooled systems with evaporative grey water, reduces electricity consumption by 10%. An air-cooled data centre with a power usage effectiveness (PUE) of 1.45 (the lower the better) is expected to improve to less than 1.3 PUE with grey water usage. However, is this level of PUE sufficient in a climate as hot and humid as many parts of India? Many industry experts have adopted a strategy to decarbonise the electricity used by data centres through renewable capacity.

Concerns about community impact due to large data centres exist. These facilities, which typically draw tens of megawatts continuously, raise questions about whether they compete with resources meant for homes and local businesses. Most large data centres in India connect through dedicated high-voltage substations rather than the local distribution network.

Moreover, a growing share of their power comes from renewable generators under long-term contracts. This shift not only adds new generation capacity but also contributes to improving utility load-factor economics. Well-planned clusters can also stimulate grid strengthening, benefiting the entire region beyond the facility's immediate surroundings.

AI will undoubtedly require substantial physical infrastructure. However, growth should not be an excuse for careless resource use. The challenge lies in building AI infrastructure while managing its footprint effectively. Rigorous planning, transparency, and continuous investment in improvements are essential to achieve this goal.

India can either stand by and wonder why the next generation of technology is being built elsewhere or take proactive steps to build AI infrastructure with renewable energy as a backbone, embed efficiency into the design from the outset, and ensure transparency to prove its effectiveness.