From creating images to drafting emails to posing questions to a chatbot, simple as well as creative tasks seem to have been made quick, effortless, and even fun by artificial intelligence (AI). 

But what many don’t realise is the environmental cost it entails. 

AI relies on data centres, massive computing machines, and mighty infrastructure – all of which run on energy and water.

Google Research in an August 2025 report estimates that an average prompt on its new Gemini AI system requires about 0.26 millilitres of water and 0.24 watt-hours of energy. 

The number may be specific to Google's infrastructure and process, but it shows what it takes to deploy an AI system. 

It may also seem insignificant at first glance - 0.26 millilitres amount to just five drops of water - but let us not forget the scale. The demand for computing power is increasing at a galloping speed, with the number of people using AI tools growing exponentially.

OpenAI CEO Sam Altman has drawn attention to the costs of these minor interactions. In April 2025, he joked that users saying 'please' and 'thank you' to ChatGPT had added tens of millions of dollars to the company's electricity costs, although he described the expense as 'well spent'. 

The innocuous remark, however, illustrated perfectly how small interactions play out at scale.

Nutan Zarapkar of Wonderful WE-R, a water and energy solutions company, says the impact must be assessed across the entire value chain.

“AI's water footprint is small per prompt but large across the value chain, and it is tied to energy, climate and waste,” Zarapkar told The Secretariat.

The 2025 demand for water by data centres in India was estimated at approximately 150 billion litres, which is the average annual consumption of 6.85 lakh households in India - and this is expected to increase two-fold by 2030. 

Behind The Prompt

Data centres, server production, and semiconductors required for AI to function involve electricity, mining, and the processing of critical minerals.

One of the biggest challenges is cooling. Data centres typically produce significant amounts of heat, with some utilising water-based cooling systems to operate at a certain temperature range. 

Evaporative cooling can help to decrease power consumption, but the drawback is water usage. The use of water can be minimised by replacing water with alternative cooling methods like air cooling and/or closed-loop liquid cooling, but much depends on the design and operating conditions.

Aksheshkumar Ajaykumar Shah, Founder and CEO, Cogniify.ai, explains that the water footprint can vary considerably depending on the task and platform. The use of computing power can increase for more complicated tasks such as high-quality images or videos.

“Various interrelated factors affect the water footprint of AI prompts. The first is the model’s design and computational complexity. For instance, generating images or providing lengthy analytical responses may require much more computing power than answering a simple question.”

He adds that cooling systems, electricity generation, and the location of data centres also influence the overall impact. “Performing the same computing task can have very different consequences depending on local temperatures, the cooling technology used, and the availability of water resources.”

Data Centre Challenge

As India's digital infrastructure grows and the nation aims to become a leading data centre hub, the question of AI’s environmental cost is becoming more pertinent.

In mid-2025, India's total data centre capacity was estimated at approximately 1.5 gigawatts, with the number projected to grow to 6.5 gigawatts by 2030, according to an assessment by the Council on Energy, Environment and Water (CEEW) and SYSTEMIQ in February, 2026. 

The expansion will not only call for investment and electricity but also require proper management of water resources.

These forecasts highlight the need for consideration to the geographical position and location while setting up new data centres, especially for those in (or near) urban areas which are already struggling for water supply.

The expansion is also beginning to impact areas outside traditional data-centre hubs. 

Proposals for large-scale investments have been received, such as that by the Adani group at a potential 1GW plant in Odisha. The government of Assam is building a greenfield AI data centre and a disaster recovery centre (DRC) in Guwahati. 

Many claim water and power savings at the project level, but these have not been quantified and published.

Not surprisingly, India has protests around data-centre development. Local communities, farmers and environmental groups have raised concerns over land acquisition, water consumption, and electricity demand. 

Projects linked to big names like Google, Amazon, and Adani in Andhra Pradesh, Maharashtra, and Odisha respectively have attracted controversy every now and then.

Case For Accountability

As digital operations are outsourced to external cloud companies and AI platforms, integrating sustainability into technology procurement and governance may be the need of the hour. 

Jaspreet Singh, Partner, Grant Thornton Bharat, says businesses must look beyond the productivity gains offered by AI and examine the resources required to support it.

“Business leaders should work with technology partners to improve data availability, establish consistent measurement frameworks, and disclose material environmental impacts, including direct and indirect water use, energy consumption, estimation methods, and exposure to water-stressed regions,” he says.

It is, however, difficult to measure these impacts.  Consequently, the water and energy footprint is not attributable to a specific AI application. and instead related to multiple applications and services in data centres. 

Increased transparency by technology companies on water consumption, energy usage, and how energy heats or cools would benefit businesses, say experts. 

Other initiatives include reusing wastewater for cooling, optimising infrastructure, expanding the utilisation of renewable energy, and avoiding areas that are under water stress.

The most important question is: can the infrastructure being built to support the rapid adoption of AI be done sustainably?