- University of Michigan researchers estimate that indirect water consumption could add 14 bn gallons annually to the water footprint of three planned AI data centers in the state.
- Meta reported consuming 21 bn gallons of water in 2024, nearly 120% more than in 2021, with approximately 96% attributed to indirect consumption.
- Estimates of annual indirect water use by US data centers range from 211 bn to 4 trillion gallons, as electricity demand and concerns about regional water supplies increase.
Artificial intelligence data centers are placing growing pressure on US water resources, with electricity generation accounting for a substantial share of their water consumption even as operators introduce more efficient cooling technologies.
Data center developers are promoting systems designed to reduce on-site water consumption, particularly in drought-prone states such as Arizona. Yet these improvements do not directly address the water used by power plants to generate electricity for AI servers and their cooling infrastructure, Bloomberg noted.
Research from the University of Michigan estimates that indirect water consumption would add approximately 14 bn gallons annually to the combined water footprint of three large AI data centers planned in the state.
The findings draw attention to water consumption occurring beyond data center facilities, particularly at fossil fuel and nuclear power plants that use water to generate steam for electricity production.
There has been much less attention to this indirect water usage, even though that oftentimes is significantly higher than the direct water usage and has impacts on watersheds and regional water resources.
Shaolei Ren, a professor at the University of California, Riverside
Meta Platforms is among the few major technology companies to disclose its indirect water consumption. According to its latest sustainability report, the company consumed approximately 21 bn gallons (80 bn liters) of water in 2024, an increase of nearly 120% compared with 2021.
Indirect consumption accounted for approximately 96% of that total. The remaining 4% was consumed on-site, primarily at data centers, which also account for nearly all of Meta’s electricity consumption.
The scale of indirect water use reflects the electricity requirements of AI infrastructure. According to researchers, computing-intensive AI data centers can consume 10 times as much electricity as conventional facilities, with most energy used to operate servers.
The rapid expansion of these facilities could increase pressure on water supplies already affected by climate-driven droughts and heat waves.
Estimates of the industry’s indirect water consumption vary considerably. A 2024 Lawrence Berkeley National Laboratory study estimated annual indirect water use by US data centers at 211 bn gallons (800 bn liters).
Demand for data centers is just skyrocketing and the truth of the matter is that the power and water facilities won’t be able to keep up
Kerri Hickenbottom, an associate professor of engineering at the University of Arizona
A more recent analysis by environmental nonprofit Ceres suggested consumption could reach as much as 4 trillion gallons annually.
Those estimates correspond to the yearly household water requirements of approximately 2 mn to nearly 37 mn average US households.
The expansion of data centers has also generated opposition over environmental impacts and rising electricity bills.
According to research firm Data Center Watch, at least 45 US data center projects valued at nearly $68 bn were blocked or delayed between April and June 2026. Local resistance has resulted in moratoriums and other development restrictions.
Measuring indirect water consumption remains difficult because the industry lacks a standardized reporting methodology. Developers cite this absence of common accounting standards as one reason for limited disclosure.
The difference between on-site cooling requirements and water consumed through electricity generation presents a significant measurement challenge as AI infrastructure expands, particularly in regions where water availability is already constrained.
