Background — The Water Footprint of Software
Background — The Water Footprint of Software
In short: software doesn't use water the way a factory does, but it causes water use all the same — through the electricity it draws, the data centres it runs in, and the hardware it's built from. Unlike carbon, where a tonne of CO₂ matters the same everywhere, water matters differently depending on exactly where it was used — which is the one idea on this page worth sitting with longest.
Why water, and why now
Software does not use water directly in any way most practitioners are used to thinking about. It uses electricity, and it runs on hardware — and both of those things are surprisingly water-intensive. Think of it as software's water footprint: the water use that follows, indirectly but measurably, from running and building the systems software depends on.
That water footprint reaches software through three separate pathways:
- Direct water use at data centres. The facilities where software actually executes use water for cooling, humidification, and other operations. This scales with the energy the workload consumes.
- Indirect water use through electricity generation. Power plants use water — often a great deal of it — to generate the electricity a software system consumes. How much depends on both the software's energy intensity and the water intensity of whatever is generating its electricity.
- Embodied water in hardware manufacturing. The servers, network equipment, and other infrastructure software runs on required water throughout their own lifecycle: extracting raw materials, manufacturing components and devices, and transporting them.
Four concepts you need before the details
Water intensity measurement introduces ideas that carbon intensity measurement does not need, because the impact of water use is highly dependent on context. Before going further into any specific component of the SWI formula, it helps to have these four in place:
- Consumption vs. withdrawal — two different things happen to water when it's used, and they matter differently. See Water Accounting Basis.
- Stress adjustment — the same volume of water means something very different depending on how scarce water already is where it happened. See Stress Characterisation.
- Geographic attribution — where a water use event happens is not incidental detail; it's central to what the number means (see below).
- Operational vs. embodied — water used while software runs, versus water used building the hardware it runs on. See Operational Water and Embodied Water.
The one thing that has no SCI analogue: location matters
If you're coming to SWI from the Software Carbon Intensity (SCI) specification, this is the idea to sit with, because SCI has nothing quite like it. Carbon dioxide is a global pollutant: a tonne of CO₂ has essentially the same climate effect no matter which power plant emitted it. Water is not like that. Water stress is regional. A cubic metre of water consumed in a basin that is already under severe stress represents a fundamentally more significant impact than the same cubic metre consumed somewhere water-abundant — and water saved in one basin does nothing to help a different basin under stress. There is no "netting out" across geography the way there implicitly can be with carbon.
This has a direct, practical consequence that recurs throughout this guidance: the location where software runs is not necessarily the location where its water impact happens. A software system might run in a facility in one place, while the electricity it consumes was generated at a power plant somewhere else entirely, under very different water conditions. The specification is explicit that the software's operating location is never a valid substitute for the location where the underlying water use actually occurred — direct operational water is attributed to the facility's watershed, indirect operational water to wherever the electricity was generated, and embodied water to wherever the relevant lifecycle process took place. Getting this geographic attribution right is arguably the single most consequential correctness question in an SWI assessment.
A sense of scale
Public water-use disclosures from major cloud providers — for example, hyperscaler reporting that breaks down per-customer water use, including both consumption and withdrawal — give a useful, if illustrative, sense of just how much water sits behind everyday cloud computing. This guidance does not treat any specific disclosed figure as an SWI reference value; they are examples of the kind of transparency that exists in the industry today, not inputs to plug into a calculation.
Please submit any comments you have here.