We've been building out CarbonLayer's facility-level water stress metrics, and one pattern keeps showing up in the underlying research: the biggest variable in water stress isn't the operator, or even the infrastructure standard a facility is built to. It's the site.
The problem with fleet-level numbers
Most water usage reporting for data centers happens at the company or portfolio level — one WUE (Water Usage Effectiveness) figure representing an entire fleet. That number is easy to report and easy to compare. It's also structurally incapable of showing you the thing that actually matters: whether any specific facility sits in a water-stressed region.
Two facilities built to the identical infrastructure standard, run by the identical operator, can have completely different water risk profiles — because water stress is a function of local supply and demand, not build spec. A facility in a water-scarce region can carry a deficit for most of the year while a sister facility, averaged into the same company-wide number, sits in a low-stress basin. Blend them into one figure and the site carrying the real risk disappears into the average.
This isn't a hypothetical edge case — it's the norm. Water availability is regional and seasonal; data center portfolios are geographically distributed by design. The mismatch between how the risk actually varies and how it typically gets reported is the gap CarbonLayer exists to close.
Modeled vs. measured — a separate axis entirely
The other place aggregation causes false confidence: conflating modeled stress signals (built from frameworks like WRI Aqueduct) with audited or measured facility data. These are different confidence levels. A modeled estimate is a reasonable starting signal. A third-party audited figure is a verified claim. Reporting them side by side as if they're interchangeable is how a portfolio ends up looking safer on paper than it is on the ground.
Why we measure at the facility level
This is the core design decision behind CarbonLayer: metrics at the facility level, not the company level, with modeled and measured signals kept explicitly separate rather than blended into one sustainability score. If you're an operator, regulator, or customer trying to assess real exposure, the resolution you measure at determines whether the risk is visible at all.
Where this is going
We're expanding audited facility coverage (Toronto and Singapore next) and working on a standard for disclosing modeled vs. measured water metrics side by side, instead of as one blended number.
If you work in data center ops, ESG reporting, or hydrology/geospatial data and have thoughts on how facility-level disclosure should work — I'd like to hear them.
Top comments (0)