01Power Usage Effectiveness (PUE)1.5 default
Conventional AI-facility PUE for Saudi Arabia; a sustainable target would be 1.3. Mode B varies PUE by cooling: air 1.7, hybrid 1.5, liquid 1.3. Alshehri et al. (2025), KAPSARC/ICAIRE, Table B2, p. 62.
02Water Usage Effectiveness (WUE)0.15–1.8 L/kWh IT
By cooling type: air-cooled 1.8, hybrid 0.5, liquid-cooled 0.1–0.15 L/kWh IT. Microsoft Sustainability Report (2024); Google Environmental Report (2025); Spindler et al. (2024).
03GPU TDP & server overhead700–1,800 W · 1.35–1.82×
Per-platform TDP from NVIDIA datasheets. Server overhead (1.39× NVL72 racks, 1.82× DGX H100) covers CPUs, memory, NVLink/NVSwitch fabric, PSU losses and intra-server cooling. NVIDIA datasheets (2024–25); SemiAnalysis (2024), "100K H100 Clusters".
04IT infrastructure overhead1.1×
≈10% on top of compute for top-of-rack networking, storage and cluster-management servers. Industry convention; no single primary source.
05Utilization factor0.80 AI · 0.65 non-AI
75–85% is typical for AI training facilities; 50–70% for enterprise/colocation. Alshehri et al. (2025), Table B2, p. 62; Shehabi et al. (2024).
06Electricity tariff0.18 SAR/kWh
SERA cloud-computing tariff category (≈ $0.048/kWh); requires CST cloud registration and ≥80% annual load factor. The final dedicated data-centre tariff is not yet formally decided. SERA tariff schedule; Alshehri et al. (2025), §4.4, pp. 29–30.
07Land factor2,000–3,500 m²/MW IT
Liquid 2,000 / hybrid 2,500 / air 3,500 m² per MW IT. Real projects span ~1,000–9,000 m²/MW — anchors: Khazna Dammam ≈1,125 (DCD, 2025); Applied Digital Ellendale ≈9,100. Defaults are a hyperscale midpoint.
08Generation capacity factorsCCGT 0.85 · Solar 0.24
Saudi annual solar capacity factor runs 22–26%. Solar PV equivalent is energy-equivalent nameplate, not firm capacity. Solar land assumes ≈500 MW per km². KAPSARC Renewables Tracker (2025).
09T&D losses & reserve margin8% · 1.10×
Typical Saudi transmission & distribution losses are 6–9%; the Saudi Grid Code reports a planning reserve margin of 8–10%. IEA Saudi Arabia profile (2024); SEC Annual Report (2024); Saudi Arabian Grid Code (SERA, May 2024).
10Grid emissions factor0.568 kg CO₂e/kWh
Reflects the Saudi fossil generation mix of 41.2% oil + 58.2% gas. IEA Emissions Factors (2024); Alshehri et al. (2025), Appendix C, p. 63.
11National context anchors80 GW · 365.4 TWh
Saudi peak ≈80 GW (SEC reported 72.9 GW in summer 2024, ≈97% of national supply; ~80 GW reflects the expected 2025–26 peak). 2030 national electricity demand projected at 365.4 TWh. SEC 2024 Financial Results (2025); MEES (2024); Alshehri et al. (2025), Appendix B, p. 61.
12Cooling & conversion constants3.517 kWth/ton
Nearly all electricity consumed becomes heat, so cooling load ≈ total facility power in thermal terms. Refrigeration tons = kWth ÷ 3.517. Year = 8,760 hours. Mode B recommends liquid cooling above 50 kW/rack and hybrid above 25 kW/rack. Uptime Institute (2024); JLL (2024).
13Grid connection voltage33 kV → 380 kV
Indicative thresholds: ≥50 MW facility demand connects at 380 kV; 10–50 MW at 132 kV; below 10 MW at 33–69 kV distribution level.
14Scenario design (2030)2.0 / 4.1 GW IT
Scenario 1 (Moderate): 2,000 MW total IT load — 950 MW AI on GB200-class hardware with hybrid cooling. Scenario 2 (High): 4,100 MW — 3,050 MW AI on B300/GB300-class hardware, liquid-cooled. Both retain a 1,050 MW non-AI baseline at PUE 1.7 and 65% utilization.