Every figure comes with how it was made.
Naturecode is a public record of the environmental impact of AI infrastructure. This is the contract behind it — the trust tiers, the methods, the open sources, and the plain endpoints anyone can pull without a key or an account.
Three ways we know a number.
Modelled from public inputs — the facility's capacity, cooling method, and local grid — using a documented method. The provenance record carries every input, so the number can be replayed and checked.
Disclosed by the operator, in a sustainability report or filing. We keep the source link. Reported figures are more specific than estimates but remain self-reported, not independently metered.
Read from a live meter or feed and carried with a signature, so the reading can be traced to its origin. The strongest tier — and the one most facilities do not yet reach.
Estimates you can replay.
- 01IT loadInstalled capacity × a 60% load factor × 8,760 hours a year.
- 02Total energyIT load × power usage effectiveness (baseline 1.2) for the full facility draw.
- 03On-site waterIT load × water-per-kilowatt-hour for the cooling method (Green Grid WUE).
- 04Off-site waterTotal energy × the grid's water intensity (Ren et al. EWIF).
- 05CarbonTotal energy × the country grid's carbon intensity (Ember).
- 06Embodied, land, e-waste, heatThe fuller footprint, each from capacity and location on a documented per-megawatt basis.
The fuller footprint.
The carbon in building the facility and manufacturing its servers — before it draws a single watt. On a decarbonizing grid this becomes the larger share of lifecycle emissions.
Concrete and steel tonnage in the shell. Cement and steel are among the hardest industries to abate, so the build itself carries weight.
Site area, sealed ground, prior land cover, and how close protected habitat sits. Where a facility lands shapes what it displaces.
Not just how much water is drawn, but whether the basin can spare it. A cubic metre in an arid basin is not the same as one where water is abundant.
Hardware retired on a refresh cycle. Fast GPU turnover concentrates e-waste with recoverable and hazardous materials.
Local peak load and the air emissions of the plants that serve it — plus who ultimately pays for the grid build-out.
Continuous fan and chiller noise at the site boundary, a live grievance for communities next to large facilities.
The share of drawn energy rejected as low-grade heat — a burden today, a resource where it can be reused for district heating.
What every figure traces back to.
Open endpoints, no key.
/api/public/datacenters/api/public/datacenters/{slug}/api/public/datacenters/{slug}/timeseries?kind={kind}&field={field}/api/public/ecosystem/{slug}/timeseriesPublic data is licensed for reuse with attribution. Cite the original sources named above.