The next time you type a question into an AI chatbot, the water used to produce the answer will probably be nowhere near you.
For India, that invisible water footprint is becoming a very physical question.
In July and August, residents and environmental groups in Thane and Visakhapatnam took to the streets over proposed large-scale data centres, raising concerns about water, electricity, land and the impact of such facilities on already stressed local resources.
The protests are among the clearest signs yet that India’s race to build AI infrastructure is beginning to collide with a much older problem: where the country’s water will come from.
India’s data-centre capacity has risen from about 375 MW in 2020 to around 1,575 MW in 2026, according to the Ministry of Electronics and Information Technology.
The government says the sector is expanding beyond established hubs such as Mumbai, Navi Mumbai, Chennai, Hyderabad, Bengaluru, Delhi-NCR and Jamnagar into states including Andhra Pradesh, Madhya Pradesh, Chhattisgarh and West Bengal.
The Council on Energy, Environment and Water (CEEW), in a 2026 assessment, estimates that India’s data centres used approximately 150 billion litres of water annually in 2025 and warns that this could more than double by 2030. It projects installed capacity could reach 4.5–6.5 GW by 2030, with Mumbai currently accounting for roughly a quarter of India’s data-centre facilities, followed by Chennai, Hyderabad and Bengaluru.
These figures are for data centres as a whole, not AI alone.
Not every server is running an AI model. But the rapid growth of high-performance computing is changing the infrastructure equation.
AI workloads require dense concentrations of processors. Those chips generate substantial heat, which has to be removed continuously.
Depending on the cooling system, this can involve water directly through evaporative cooling or indirectly through the electricity and manufacturing supply chains supporting the facility.
A July 2026 study published in Water Research estimated that the global water footprint of AI could reach 4.2–6.6 billion cubic metres annually by 2027.
The researchers identified three major components: water consumed at data centres for cooling, water associated with electricity generation, and water embedded in semiconductor manufacturing. They also noted that two-thirds of data centres built after 2022 are located in water-stressed regions.
But translating a global estimate into a number for “one AI query” is complicated. Water consumption varies dramatically according to the model, workload, server efficiency, cooling technology, climate and electricity source.
Estimates such as “a bottle of water for a handful of prompts” are therefore useful for illustrating the concept, but should not be treated as a universal per-query tariff.
The bigger Indian question is where that water demand lands.
When the data centre arrives in the neighbourhood
In Balkum, Thane, hundreds of residents protested on July 18 against a proposed Amazon data centre spread across roughly 53 acres.
Residents from Balkum, Kolshet and Dhokali said they feared the facility would place additional pressure on water and electricity supplies, apart from concerns over heat, noise, traffic and tree loss.
A July 2026 study published in Water Research estimated that the global water footprint of AI could reach 4.2–6.6 billion cubic metres annually by 2027. Photograph: (ANI)
Protesters claimed the facility could require at least 12 million litres of water a day.
The campaign, led by the citizen group Wake Up Thanekar, subsequently sought answers on the project’s water demand, environmental approvals, sewage infrastructure, thermal impacts and emergency preparedness. Residents also questioned the classification of data centres as essential services and whether that could give such facilities priority access to resources during shortages.
Amazon, however, has disputed the central water-diversion concern, saying its Indian data centres do not use local water and power sources and that it complies with environmental requirements.
The same tension is playing out farther south
In Visakhapatnam, environmental groups and residents have opposed Google’s proposed $15-billion data-centre hub, being developed with the Adani Group. Activists have organised marches and awareness campaigns, with children among those taking part. Their message, “We cannot drink DATA”, turned a debate about AI infrastructure into a question about basic access to water.
The concern is especially sharp because Visakhapatnam already has a supply deficit. According to figures cited by Reuters, the city receives about 410 million litres of water a day against a requirement of 480 million litres, and rationing is common.
The project has also faced legal challenges over its environmental impact and proximity to Kambalakonda Wildlife Sanctuary, located about 860 metres away, according to Reuters. Three cases have also been filed before the National Green Tribunal, while a public interest litigation has been heard by the Andhra Pradesh High Court.
Thanekars gather to protest upcoming Amazon Data Centre in a residential zone! pic.twitter.com/NgRvkl6s70
— Roads of Mumbai (@RoadsOfMumbai) July 18, 2026
Activists have alleged that the project could affect water resources and wildlife; the state has rejected the allegations and said no water meant for rural or residential use would be diverted to the data centres.
Google, meanwhile, has said it will use advanced air cooling to protect local water resources.
India is trying to cool the problem
The Centre says the industry is increasingly adopting direct-to-chip liquid cooling, immersion cooling, adiabatic cooling and closed-loop systems, alongside high-density racks designed to reduce power and water requirements.
Amazon said in June that its operations in India had become “water positive”, claiming that it returns more water to local communities than it consumes across its Indian operations. It also said its Indian data centres do not use water for cooling.
The government has also clarified the regulatory position. In August, the environment ministry told Parliament that AI data centres do not, by themselves, require a separate environmental clearance under the 2006 EIA notification.
However, large data centres falling within specified building or township thresholds do require clearance, during which freshwater availability, water balance, greywater generation and recycling can be examined.
Meanwhile, the geography of the boom is already clear: Mumbai and Navi Mumbai, Chennai, Hyderabad and Bengaluru are established hubs, while Visakhapatnam and other emerging destinations are being positioned for the next wave of investment.
The irony is that some of these locations are also cities where water stress, groundwater depletion, erratic rainfall or competing urban and industrial demand are already part of everyday life.
So the question is not whether an AI prompt literally drains a glass of water somewhere in India.
It is whether millions of prompts, billions of computations and hundreds of new megawatts of computing capacity will create a water demand that is accounted for before the servers arrive, not after residents begin asking where their water has gone.
