Is the race to build AI in developing nations worth the water it takes?

Artificial intelligence is being touted as a way to revolutionize everything from health care to education in the global south. But as low- and middle-income nations rush to harness AI through server farms and data centers, it is creating a silent crisis — a strain on already stretched water resources. In Bengaluru, the “Silicon Valley of Asia,” the AI dream is clashing with the people’s most basic right: access to water. Earlier this year, India’s minister of technology, Ashwini Vaishnaw, announced a national push to build an indigenous AI platform with 18,000 servers by the end of the year. The move followed Chinese tech giant Lenovo’s announcement last year that it would begin manufacturing AI servers in the city. But Bengaluru — which is home to over 13 million people and more than a dozen technology campuses — has been facing acute water shortages in recent years due to overexploitation and erratic monsoon. These shortages have caused recurring protests, especially during the summer months of April to June, by the residents, including tech employees, demanding a stable water supply. At the heart of the issue lies the huge resource demand of AI infrastructure. Unlike traditional IT systems, the new wave of AI requires intensive computational power, which in turn demands large server farms and data centers. These facilities generate massive heat and require ultra-clean water to cool their equipment efficiently. On average, a one megawatt small data center can consume up to 18,000 gallons of water per day. Bengaluru is already home to around 22 data centers, with more on the way as India courts global tech firms and AI developers. But experts warn that this level of water consumption is alarming in a city already grappling with severe water shortages. “There has been substantial pressure on the already suffering water systems in Bengaluru. Groundwater here is already over-exploited and the water from the shallow aquifers only last a few months and the deep aquifers do not replenish. People are dependent on water tankers, and are facing floods/drought conditions year on year,” said Khushbu Birawat, researcher and curator at Paani.Earth, a nonprofit that works on water and river rights. A recent United Nations working paper warns that the rapid digitization of economies in the global south risks outpacing natural resource capacities, warning of what it calls drought-like situations where water scarcity is accelerated by unregulated technology expansion. The report cites Google’s planned data center in Uruguay and Meta’s planned data centers in the Netherlands and Arizona as examples that stoked tensions within local communities. Experts warn that this growing demand for energy and water-intensive AI infrastructure could worsen water scarcity and mimic Taiwan’s 2021 chipmaking crisis, which was triggered by severe water shortages forcing the government to reduce water supply to industries, including chip manufacturers, thus affecting global semiconductor chip supply. Bhargavi S. Rao, a senior fellow at the Environment Support Group, a Bengaluru-based nonprofit that focuses on environmental and social justice issues, warned that without sustainable water use policies, Bengaluru could face droughts even more severe than those Taiwan endured. She added that government AI investments are sources of resource inequality, and without proper planning, the water is simply taken away from people and given to machines. “No matter the water situation in Bengaluru, the government makes sure that the tech industry survives in the city, because that’s a feather in its cap,” Rao said. “For this reason, the government provides companies with water, and the local people face the brunt of it.” The water shortages are not only about water demands and policy, but also about accountability. Experts say India lacks clear regulations governing water use in data centers. While the Ministry of Electronics and Information Technology has announced infrastructure incentives for server and cloud infrastructure, environmental safeguards have lagged. “It is important to require transparent water use reports from tech firms, mandating environmental and social impact assessments for new data centers, and setting limits or incentives for water-efficient technologies,” Sundeep Narwani, an AI expert and co-founder of the Narrative Research Lab, an AI-run project, said. In a recent blog post Sam Altman, CEO of OpenAI, said that an average ChatGPT query uses “about 0.000085 gallons of water; roughly one fifteenth of a teaspoon” but research conducted by the University of California, Riverside, last year found that ChatGPT requires up to 1,408 milliliters of water, or about three 16.9 oz bottles to generate a 100-word email. It takes 40 to 50 milliliters of water to generate a three-word "You are welcome," their research found. In India, local communities are already feeling the pressure. In Whitefield and Electronic City, two major tech corridors of Bengaluru, residents complain about declining water tables and reduced availability of groundwater. Some housing societies have to pay high rates for private water tankers, and the cost of living has surged due to the competition over resources. "We used to have groundwater up to 200 feet. Now it’s gone. People around us have dug up to 1000 feet to find water. The data center next to us runs 24/7 with ACs and backup generators. What do we do?" said Karthik M., a resident of Marathahalli. Schools and clinics in these areas also struggle to stay open during peak summer months due to water shortages. “AI may be the future, but our children are the present,” said Revathi Naik, a teacher in Sarjapur. “If tech companies and the government want to build here, they must take responsibility for the basics.”