AI and Water White Paper

Clarifying the discussion around AI's water footprint

The conversation around AI and water consumption lacks precision. Direct on-site water (consumed for cooling at the data centre) and embodied off-site water (consumed in energy generation, semiconductor fabrication, and the broader supply chain) are fundamentally different measurements with different levels of certainty, but they are routinely conflated.

For direct water, there is reasonable certainty. Water Usage Effectiveness (WUE) exists as a metric and operators can measure it. The challenges are real but tractable. For embodied water, the picture is less clear. Calculating embodied water requires attribution down to specific power grids and generation sources, and the methodologies to do that reliably are still developing. A significant portion of embodied water sits in areas where uncertainty remains high.

This assembly will produce a white paper that maps both sides clearly: what can be measured now, what cannot, where the data gaps are, and what work would be needed to close them. The goal is to provide the industry with a shared reference point that distinguishes what is known from what is assumed, and to bring clarity rather than add further noise to the discussion.

The white paper will also serve as the evidence base for the Green Software Foundation's planned Software Water Intensity (SWI) specification, informing its scope and identifying which measurement areas are ready for standardisation and which require further research.

If you work in data centre operations, cooling systems, water stewardship, energy grid analysis, or environmental reporting, this assembly needs your expertise.