India’s Jamnagar and Thane have emerged among the world’s locations most exposed to data centre-driven water stress, as the rapid expansion of artificial intelligence (AI) infrastructure threatens to sharply increase cooling-water demand in regions already facing water constraints, according to Rystad Energy.
The pressure is expected to intensify over the next four years. Regions facing high or extremely high water stress are projected to account for 34% of the data centre sector’s global direct water consumption by 2030. However, mandating the least water-intensive cooling technologies in such locations could reduce consumption by as much as 45%, Rystad said. Jamnagar and Thane, along with Reeves County in Texas, are among locations where data centre water consumption is high relative to local water stress, highlighting the potential conflict between AI infrastructure expansion and water availability.
Globally, direct water consumption by data centres could nearly triple to 644 billion litres a year by 2030 from 222 billion litres in 2025 if water-saving measures fail to keep pace with the AI-led expansion in computing capacity.
An aggressive water-efficiency pathway could limit annual consumption to about 388 billion litres by 2030, while a moderate scenario would put it at around 543 billion litres. The projections underline the scale of potential savings from cooling technology and data centre design choices.
“Data centres consumed 222 billion liters of water directly for cooling in 2025, but that figure could nearly triple to just under 644 billion liters by 2030 in our risked central case,” said Minh Khoi Le, global head of data centre and hydrogen research at Rystad Energy.
“More water-efficient pathways could bring this down to 543 billion liters in the moderate case or as low as 388 billion liters in an aggressive water-saving scenario,” Le said.
The estimates cover only water consumed directly for cooling. The electricity used by data centres carries a separate, indirect water footprint, adding another layer to the resource challenge.
Cooling technology will therefore be critical as AI workloads expand. AI servers generate significantly more heat than conventional computing equipment, but rack-level liquid cooling can reduce the burden on facility-level cooling systems and enable greater use of dry cooling. Dry cooling can save around 2.15 litres of water for every kWh of IT load, but requires an additional 0.30-0.74 kWh of electricity.
That trade-off can shift water consumption away from the data centre itself to the power system. In the US, indirect water consumption associated with electricity supply can be more than twice direct consumption.
Water efficiency also varies sharply between operators. AWS reported average direct-site water-use effectiveness of 0.12 litres/kWh in 2025, compared with Meta at 0.19 litres/kWh, Microsoft at 0.27, Digital Realty at 0.59 and Equinix at 0.91.
Even within AWS, the figure ranged from 0.02 litres/kWh in Stockholm to 2.85 litres/kWh in Jakarta, a spread of more than 100 times, highlighting how climate and cooling architecture can materially alter data centres’ water footprint.