How every number here is produced

METHODOLOGY, FORMULAS, ASSUMPTIONS AND LIMITS

1. DATA ARCHITECTURE (layers, in build order)

L0 Raw preserved - the supplied workbook, byte-for-byte, sheets 70-73. Never modified, never overwritten.

L1 Entity resolution - individual facilities identified from operator listings and market directories,

deduplicated by (operator, facility name, area). Multi-building campuses are carried as a

CAMPUS parent row plus child FACILITY rows; the parent is excluded from every sum so that

capacity is never double counted.

L2 Geospatial - each facility mapped to one of 283 reference AREAS (industrial estate / business park /

suburb) carrying coordinates, land-use class, resident density band, basin water-stress

score and mean annual temperature. Sheet 52.

L3 Capacity - reported where published; otherwise allocated from a documented market total. Section 3.

L4 Indicators - the six pillars' underlying quantities, derived from L3 and the benchmark parameters

in sheet 51. Sheets 11-16.

L5 Normalisation - indicators rescaled 0-1 across the whole facility population. Section 4.

L6 Pillar and composite scores, risk classes, sensitivity. Sheets 20-21.

L7 Clusters and aggregates - every aggregate is a SUM or MEDIAN of L3/L4 facility values. Sheets 30-34.

L8 Time series - the 2026 fleet back-cast to 2020. Section 6. Sheets 40-41.

L9 Provenance and quality - sheets 50, 60, 61, 62.

2. EVIDENCE TAXONOMY (inherited from the supplied workbook, extended)

Reported - taken directly from an operator disclosure, regulator, government dataset or facility specification.

Derived - arithmetic on other values in this workbook (e.g. waste heat = annual energy / 8,760).

Modelled - produced by an explicit documented formula anchored to at least one real, cited data point.

Allocated - a documented market or portfolio total distributed across facilities by a stated weighting rule.

Proxy - a real observation from a different scale (area, city, country, operator fleet) used because no

site-specific figure exists. The scale mismatch is the limitation.

Not Available - no value could be established. Left blank, and flagged in 61_Missingness_Matrix.

CONFIDENCE SCALE (as specified in the brief)

5 Directly reported or measured by the operator or a government body FOR THAT FACILITY.

4 Highly credible secondary source, or an operator fleet figure applied to that operator's own site.

3 Reliable estimate from a documented methodology.

2 Weak or incomplete evidence - typically an area-level or class-level band.

1 Inference or highly uncertain estimate.

0 No evidence.

Record_Confidence is the mean of the field confidences for capacity, PUE, WUE, coordinates, status, year and the

two context fields. It is a summary, not a substitute for the per-field confidence columns.

3. CAPACITY METHOD (the single most consequential assumption in this workbook)

Facilities with a published capacity keep it (Evidence_Type = Reported, Confidence 4).

For every other facility in a market:

Capacity_Ultimate_MW = (Market_Anchor_MW - reported capacity in that market)

x Class_Weight(facility) / SUM over the market of [Class_Weight x Energisation_Factor]

Capacity_2026_MW = Capacity_Ultimate_MW x Energisation_Factor(status)

Class weights: Hyperscale 30, Wholesale/purpose-built colocation 12, Mid-tier colocation 6, Small/edge/enterprise 1.5.

Energisation factors: Operational 1.00, Under construction 0.35, Announced 0.00, Planned 0.00.

Market anchors and their sources are in 51_Benchmarks_Reference. This construction guarantees that the facility

records sum back to the published market total, which is exactly what makes the aggregates auditable - and it

also means an individual facility's MW figure is an ALLOCATION and should not be quoted as that site's capacity.

4. INDICATOR FORMULAS

Energy_MWh = Capacity_2026_MW x 8,760 h x Load_Factor(class) x PUE

Energy_Intensity = 8,760 x Load_Factor x PUE (MWh per MW of IT capacity per year)

Water_m3 = Energy_MWh x WUE (1 L/kWh == 1 m3/MWh)

WS_Weighted_Water = Water_m3 x (basin water-stress score / 5)

CO2_Location_t = Energy_MWh x grid emission factor

CO2_Market_t = Energy_MWh x grid factor x (1 - operator matched-renewable %) [only where the operator publishes one]

Waste_Heat_MW_th = Energy_MWh / 8,760 (energy-balance proxy: essentially all electricity leaves as heat)

Heat_Island_Proxy = Waste_Heat_MW_th x resident density / 1,000

Noise_dB(A) @ ~50 m = 50 + 10 x log10(Capacity_MW), clipped to [45, 85]

Engineering proxy only. Anchored so that a ~30 MW site returns ~65 dB(A), consistent with the

60 dB(A) county ordinance threshold that Prince William County (Virginia) residents report

data-centre noise routinely exceeding.

Residential_Distance_m = land-use class band (Dense urban 60, Urban industrial 150, Suburban IT park 300,

Peri-urban 800, Rural 2,000). The dense-urban band is anchored to the Virginia JLARC 2024

finding that ~1/3 of Virginia data centres sit within 200 ft (61 m) of residential zoning.

Population_1km = resident density x pi x 1 km^2

Population_5km = resident density x pi x 25 km^2 x 0.60 (0.60 = radial density-decay factor)

Jobs_Estimate = Capacity_MW x 5 permanent roles per MW (industry rule of thumb, Confidence 1)

Infrastructure_Pressure = facility capacity as a percentage of its cluster's total capacity

5. SIX-PILLAR INDEX

Direction: for every indicator, HIGHER = GREATER environmental or community impact.

Normalisation: volume indicators (energy, water, CO2, waste heat, population, capacity) are min-max scaled on

log10(1+x) because they span five orders of magnitude; ratio and score indicators (PUE, WUE, water stress,

temperature, noise, land-use sensitivity, regulatory flags) are min-max scaled on the raw value. Scaling is

computed across the whole facility population so scores are comparable between countries.

Pillar score = weighted sum of its normalised sub-indicators x 100. Sub-indicator weights are printed in full

on 20_SixPillar_Scores. Composite = weighted mean of the six pillars; the default scheme is equal weights (1/6).

Risk class = quartiles of the composite across all 1098 facilities: Low <54.1, Medium <59.6, High <65.7, Very High above.

KNOWN PROPERTY OF THIS INDEX, STATED PLAINLY: four of the six pillars are dominated by LOCATION CONTEXT (water

stress, climate, density, land use, regulation) and only two by SCALE. A small facility in a dense, water-stressed

city therefore outranks a very large facility in a cool, low-stress rural location. That is a deliberate design

choice for a community-impact index, but it means the composite must NOT be read as 'biggest environmental

footprint'. For footprint, use the absolute columns in sheets 11-16 and the capacity-weighted aggregates.

6. TIME SERIES BACK-CAST (2020-2026)

Operational facilities are ordered by a vintage proxy (hyperscale greenfield later; lower facility numbering

earlier) and assigned a first-power year such that the cumulative fleet capacity reproduces the national capacity

growth curve that the SUPPLIED workbook already sources. Only the SHAPE of that curve is used, never its level,

because this workbook covers one flagship cluster per non-Indian country rather than the whole nation.

Under-construction, announced and planned facilities are given indicative first-power years of 2027, 2028 and 2029.

Every back-cast year carries Confidence 1. Where a real commissioning year was found it overrides the model.

7. SOURCE HIERARCHY APPLIED (highest first)

1 Government datasets and regulatory filings 2 Operator sustainability reports and official disclosures

3 Planning applications and environmental impact assessments 4 Utility and grid-operator information

5 Official company announcements 6 Peer-reviewed literature 7 Established industry datasets and market

trackers 8 Credible institutional research 9 News reporting (labelled wherever used).

Where sources conflict, the higher tier wins and the conflict is recorded in 62_QC_Reconciliation rather than

silently resolved. The Kolkata capacity conflict is the worked example.

8. WHAT THIS WORKBOOK STILL CANNOT TELL YOU

* Any individual facility's actual metered electricity or water consumption.

* Any individual facility's measured noise emission or its measured setback to the nearest dwelling.

* Actual population counts within 1 km (this needs a GHSL/WorldPop raster zonal statistic against real coordinates).

* Whether a community has in fact objected to a specific Indian, Brazilian, Singaporean or German facility.

* On-site generation, storage, heat reuse contracts or water-recycling rates at any site.

The single highest-value next step is to replace the AREA centroids with rooftop coordinates and run a real

GHSL population zonal statistic and a real WRI Aqueduct basin lookup; that alone would move Pillars 2, 5 and 6

from Confidence 1-2 to Confidence 3-4.