The artificial intelligence (AI)-driven data centre boom is fundamentally reshaping the US water landscape. Surging electricity demand is driving this transformation, shifting water-related risks upstream to power generation and affecting communities that never anticipated becoming focal points for AI infrastructure.

A new insight report, The Water-Power Nexus: How Data Centers are Reshaping the US Water Landscape, published by Bluefield Research, reveals that the most significant water implications of AI infrastructure are not occurring on-site within data centre facilities, but at electric power plants. By 2030, 72% of total water consumption associated with data centres is expected to occur off-site, linked to electricity generation — more than double the projected on-site cooling demand. Data centres are forecast to account for 8.9% of total US electricity demand by the end of the decade, up from 4.4% in 2023.

“While data centres use water directly for on-site cooling, the rapid growth in electricity demand is driving the sector’s overall water use outlook,” said Amber Walsh, research director at Bluefield Research. “Power — not on-site cooling — is becoming the defining water risk for the data centre industry.”

Indirect water consumption associated with electricity generation is projected to nearly double, rising from 54 billion gallons in 2025 to 91 billion gallons by 2030. This increase reflects the significant challenge facing power companies as they respond to a projected 12% surge in electricity demand. Near-term measures — such as postponing coal plant retirements, recommissioning nuclear facilities, and accelerating natural gas expansion — carry substantial long-term water implications.

By contrast, on-site water consumption for data centre cooling is slowing, as operators transition to more efficient cooling systems and greater use of reclaimed water. On-site consumption is projected to increase more modestly, from 22 billion gallons to 34 billion gallons over the same period.

“The water story has moved upstream, and it is unlikely to reverse,” Walsh noted. “Unlike the existing installed base of data centres, AI gigafactories and hyperscale campuses are being designed from the outset with closed-loop cooling systems, advanced liquid cooling at chip level, and integrated reclaimed water use.”

AI data centre growth is increasingly concentrated in a small number of states, shaped by tax incentives, land availability, and infrastructure capacity. Texas, Virginia and Pennsylvania together represent more than $360bn in planned investment. Expansion is also accelerating in the water-stressed Sun Belt region, where power availability is attracting capital despite limited local water infrastructure.

This imbalance is particularly evident in Arizona, where high water stress coincides with rapid data centre development. In cities such as Phoenix, Atlanta and Columbus, developers are being required to adopt reclaimed water systems and advanced cooling technologies at pace.

The scale of AI infrastructure is redefining water exposure. Hyperscale and AI-focused facilities currently account for 46% of total data centre electricity demand, a proportion that continues to rise rapidly.

Microsoft’s power consumption grew at a compound annual growth rate of 30% between 2020 and 2023. Meanwhile, Google’s indirect water footprint from power generation alone exceeded an estimated five billion gallons in 2024, representing 43% of its total water usage.

A new category of infrastructure is intensifying these pressures further. AI gigafactories — facilities built exclusively for AI workloads with capacities exceeding one gigawatt — carry a substantial water footprint driven predominantly by power demand rather than on-site cooling. Although companies such as Meta, Google and Microsoft are procuring renewable energy to offset carbon emissions, this strategy does little to reduce immediate water pressures on electricity grids.

“We are at an inflection point,” Walsh stated. “The decisions made today regarding the siting and powering of gigafactories will shape US water availability for decades. Increasingly, access to water has become a critical lever — alongside electricity pricing — that communities and local governments are using to determine their engagement with data centre developers.”