AI Data Centers' Hidden Water Footprint Revealed in New Report
Indirect water use from power generation is 12 times higher than direct cooling
A new investigation has revealed a massive gap in how AI hyperscalers report their water consumption, highlighting that indirect usage from power generation far outweighs on-site cooling. This disclosure shift matters because it reveals the true scale of AI's resource strain on local water infrastructure and the limitations of current sustainability reporting.
Key details
Research from Lawrence Berkeley National Laboratory estimates that indirect water use for U.S. data centers—water consumed at power plants to generate electricity—is roughly 12 times greater than direct on-site cooling. In 2023, total data center water consumption reached approximately 228 billion gallons, with only 17 billion gallons used for on-site cooling while 211 billion gallons were tied to electricity generation.
By 2028, total data center water use is projected to reach between 469 billion and 844 billion gallons, depending on cooling technology and the grid's fuel mix. While Google reported a 34% rise in direct operational water use to 10.9 billion gallons in 2025, independent analysis suggests its total footprint is several times higher when accounting for power supply.
Why this matters
The reporting gap obscures the actual environmental cost of AI scaling. Most hyperscalers, including Microsoft, Google, and Amazon, primarily disclose direct cooling water, leaving the much larger upstream impact uncounted in headline sustainability figures. This lack of transparency makes it difficult for investors, regulators, and communities to assess the material liabilities of water-intensive AI infrastructure.
Context
This trend is colliding with local water scarcity in the American Southwest, where two-thirds of new U.S. data center construction is concentrated. In Phoenix, data centers currently account for 3% of city water use, but this is projected to exceed 20% by 2031, rivaling the water used for all residential lawns. While companies use Renewable Energy Credits (RECs) to offset carbon, these credits do not restore water to the specific drought-stressed watersheds where the power is actually drawn.
Risks and open questions
Regulatory pushback is mounting, with 16 state attorneys general challenging whether RECs can legitimately substitute for actual resource reductions. Furthermore, approximately $170 billion of AI data center capacity has been stalled or canceled since 2024 due to local opposition over water rights and community impacts. It remains an open question whether retrofitting existing facilities with lower-water cooling will be economically feasible at the necessary scale.
What happens next
Tech giants are accelerating transitions to more efficient cooling architectures to mitigate these risks. Nvidia is touting closed-loop systems, while Microsoft has committed to zero-water cooling for all new data centers starting in 2027. Watch for increased regulatory pressure for standardized site-level water reporting and more stringent municipal water permits for new AI infrastructure projects in arid regions.
Source: WhatJobs News Published on AI Usage Global, author: AUG Bot
