Carbon and water

In May 2026 the Equator Principles Association published guidance telling project-finance lenders how to assess the environmental risk of a data centre. It covers energy, greenhouse gases, water, waste, land and labour. On two of those, carbon and water, it asks the lender to do very different work.

The analysis follows the electricity off the site and into the power system.

“Scope 2 indirect energy related emissions are largely driven by electricity consumption and therefore depend on the carbon intensity of the grid supplying the Project.”

It continues past the grid into the supply chain. Embodied carbon in construction materials, mechanical and electrical plant, and IT hardware is named, and including it is called best practice. Whole Life Carbon Assessment is defined in the glossary. Carbon Usage Effectiveness is offered as a companion metric to PUE and WUE. Operational energy, the guidance notes, “can account for up to 80% of a Project’s total whole life emissions.”

And on timing, the guidance states the industry standard plainly:

Industry best practice: “Matching renewable electricity consumption on an hourly basis.”

An hour is the unit. Not a year, not an annual average, not a contract. The lender is told to ask whether the load was drawing clean power in the hours it ran.

Section 3.5 is the Water section. These are its questions for lenders, in full:

• Does the data centre utilise evaporative or adiabatic cooling? If so, has the water scarcity within the region and local catchment area been assessed?
• Does the facility have an estimated annualised and peak Water Usage Effectiveness (WUE) calculation? Are these in line with industry and regional expectations or targets?
• Are the PUE and WUE considered holistically?
• If utilising significant amount of water for cooling purposes, have alternative sources to a potable supply been reviewed?
• If utilising significant amount of water for cooling purposes, have the downstream impacts of wastewater been evaluated?

Every question stops at the property line. WUE is defined in the same document as annual water consumption over IT energy: a meter reading on the building, divided by a year.

There is no water equivalent of grid carbon intensity anywhere in the guidance. No Scope 2 for water. No hour.

The water a data centre draws through its own pipes is not most of the water it spends. The rest evaporates at the power plants that run the servers, in the hours they run, and never crosses a meter the basin owns.

Harvard researchers mapped both pathways across 472 U.S. hyperscale facilities in June 2026. Of roughly 300 billion litres consumed per year, 226 billion, about three quarters, is consumed in the power system rather than at the buildings. The finding survives the most contested assumption in the method: setting reservoir evaporation from hydropower to zero still leaves the electricity side at 63% of the total.

Guidi and Dominici, Harvard University, “The Hidden Water Geography of U.S. Hyperscale Data Centers in the AI Era,” arXiv:2607.02531, 5 June 2026.

The same authors state their own limit: the study is “annual-average and location-based; it does not resolve seasonal drought coincidence, dispatch-level electricity, exact source-water provenance, or facility-specific measured WUE.”

So the guidance asks lenders about the quarter of the water that has a meter. The three quarters that does not has been measured once, annually, at national scale, by a university.

For carbon, a lender financing a data centre is told to trace the resource to the grid and to ask about the hour. For water, on the same building, in the same document, the lender is told to ask the building.

What does an independent, hourly, attributed record of the second one change about what you require, price, or reserve against?

On MISO-Central, the water behind a megawatt-hour is about 313 gallons (1.18 L/kWh), measured on the region’s own cooling fleet and graded hour by hour. Consumption basis, never withdrawal.

The method·The record

1. Equator Principles Association, Guidance to Support Effective Application of the Equator Principles for Data Centre Projects, May 2026 (developed with Arup). Sections 2.4.3, 3.3, 3.4, 3.5, glossary, Appendix B.

2. Guidi, G. and Dominici, F., The Hidden Water Geography of U.S. Hyperscale Data Centers in the AI Era, arXiv:2607.02531v1, 5 June 2026. Baseline totals 74.2 GL direct and 225.7 GL electricity-related of 299.9 GL/yr; no-hydro sensitivity 128.4 GL of 202.5 GL/yr.