Masterplan
8. Basis of design
Class 4 design-build study.
8.1Design basis
Vintage 2026-08-31. AACE Class 4 (-30% / +50%). Capex excludes GPUs.
Why the campus is this size
Design basis is Atlas Phase 2: ~500 MW gross / ~460 MW firm, compiled as 350 MW IT.
The Atlas plan is the program. This run compiles that campus. 350 MW IT is the IT load that puts firm generation on the plan's 460 MW N+2 claim, using the same firm-to-IT ratio as modelled first power. Phase 3 is the 345 kV cut-in of that campus, not a larger build.
The pad is the Atlas Phase 1+2 envelope, 3,261.0 ac inside the 23,200 ac vest. Campus vertical (halls + MEP, GPUs out) is compiled Class 4 on top of the Atlas site stack; those are two different checks.
P2 is compiled: every transformer, PCU, conductor and check in this document is that case. P1 is modelled at 100 MW IT. P3 is the same IT load as P2, grid-connected. Plan
Campus lot 3,254 acres (1,541 acres used). Generation is 16 × GE Vernova LM2500XPRESS+G4 DLE at 26.42 MW site rating, with rental as the first-power bridge. BESS 140 MW / 75 MWh. Hardware NVIDIA VR200 NVL72 (Vera Rubin). IT 350 MW (100 MW first power). 3 collector sections. Heat rejection dry for 95.9% of hours. 345 kV is on the northeast corner.
8.2How the design was assembled
Land and GIS live in the site file (layers, sited elements, the West Big Field lot). Electrical design sizes racks, fleets, faults, arc-flash, and coordination. Cost and schedule are Class 4 estimates with a stated accuracy band.
8.3What is fixed, what is calculated, what is handed off
- Fixed. 350 MW IT campus, 3 × 117 MW zones, 415 V, N+1 radial, island source, NVIDIA VR200 NVL72 (Vera Rubin) at 190 kW, 140/75 BESS, 3 × 34.5 kV sections, code editions.
- Calculated. Rack count, load roll-up, the hourly cooling plant and both halves of PUE, turbine count at site conditions, fleets, measured routes, time-to-power percentiles, tornado, alternative sketches.
- Not searched. Layout is placed, not optimized. Alternatives A–F are enumerated.
- Handed to licensed specialists. Everything in §6.6 and §7.5. This study is a design-build study, not a seal.
How each item is classified
A = calculated from the design basis. G = geometric. B = scenario. D = documentary. X = a licensed study we still have to buy.
| Item | Class |
|---|---|
| Grading cut/fill | A |
| Water–energy (WUE, cost) | A |
| Acoustics (ISO 9613-2) | A |
| Availability (RBD) | A |
| Generation portfolio and dispatch | A |
| Emissions (CUE) | A |
| Transmission import (exploration) | A |
| Transmission import (filing) | X |
| Ground grid (C26) | A inputs, PS check |
| Buildable area buildable area | G |
| Inundation, WOTUS, cultural, well, pipeline masks | G |
| Facility layout | G |
| Telecom ring routing | G/D |
| Permitting critical path | B |
| Groundwater, air, water permits | B |
| Species / SB6 / WOTUS jurisprudence | B |
| Estimate class and CPM | B/D |
| Phase I ESA, wetlands, cultural, geotech | X |
| Surface use agreements | X/D |
| Buildings and BIM | D/G |
| Community and economic | D/B |
| Code-edition field | D |
8.4What this study produced, and what people still add
From the design basis
- Rack count, load roll-up, PUE from the cooling mix, transformer and PCU fleets, N-1 transfers, fault / arc-flash / coordination checks, IEC 60909 on the worst section, 2024 climate at the centroid, measured routes, time-to-power percentiles, tornado, and the six alternatives.
- Campus layout on the West Big Field lot.
Judgment and records
- Site selection narrative and the 2026-08-28 due-diligence memo.
- The freezes: 3,254 ac campus lot, NVIDIA VR200 NVL72 (Vera Rubin), adiabatic-assisted dry plant, 16-unit aero-equivalent, 140/75 BESS, 3 collector sections.
- Rural factor, AI premium, hybrid liquid adder, 20% contingency.
- Anything a PE must stamp.
8.5Which results are authoritative
| Result | Authoritative? |
|---|---|
| NOAA Atlas 14 point precip | Yes. NOAA Atlas 14 at the centroid |
| 345 kV on-site, Palo Duro certificate, MAG 49, queue freeze | Yes. Public record, DD memo 2026-08-28 |
| Load roll-up, rack count, fleet counts | Yes inside the frozen gates; replace when a tenant SKU arrives |
| Hourly cooling plant (52,608 ERA5 pairs), CDU and cooler counts, mechanical PUE, WUE | Yes for this loop and this profile. A vendor coil curve and a selected cooler replace the planning module rating. |
| Turbine site rating and unit count | Screening correlation on 104.4 °F and 610 m. An OEM site-rating run replaces it before the PO. |
| Fault, IEC 60909, arc-flash | Island estimator plus IEC 60909 on the worst 34.5 kV section. A PE-sealed ETAP/SKM file is still the stamp. |
| Capex, time to power, tornado, WUE | Class 4 / Class B models. Quotes and CPM replace them. |
| Site climate and ASHRAE design conditions | Open-Meteo ERA5 2019-2024 at the centroid, 52,608 hours. NSRDB TMY3 replaces it. |
| Due-diligence memo versus older notes | The 2026-08-28 memo wins |
Documentary checks D01–D07
| ID | Check | Status | Finding |
|---|---|---|---|
D01 | Every completeness-fixture heading populated or outstanding Appendix I | pass | 13 populated, 18 partial, 7 outstanding of 38 headings. |
D02 | Quantity reconciliation: one run_id, no divergent restatements R-IMMUTABLE | pass | No superseded quantity appears in any component. Every displayed number resolves through report.json. |
D03 | Permit-matrix coverage: agency, statute, lead, status Appendix J | pass | 18 rows. Lead times are placeholders until agency-sourced. |
D04 | code_editions set and register clauses re-keyed Part 0 / C-register | pass | NEC/IBC/IFC/IMC/IPC/IECC/TAS are a manifest field. |
D05 | Every exclusion in Ω_build traces to a source polygon and layer site_geometry | partial | Exclusion list is sourced. Acoustic buffer is now a modeled contour, not a blank. Polygons remain Atlas, not ALTA. |
D06 | Every non-core item carries a fit-class and a source with access note Appendix M | partial | Kind chips on quantities. Fit-class A/G/B/D/X is named in the fixture, not on every WBS line. |
D07 | Every time-sensitive number carries an as-of date and refresh flag Appendices G, H, N | pass | as_of 2026-08-31 on headline quantities. |
What we calculated: IEC 60909 on the worst 34.5 kV section, and 52,608 hours of ERA5 at the centroid through the cooling loop. What a PE still has to buy and seal: ETAP/SKM, CDEGS, Civil 3D. Those sit before the turbine purchase order, not before this study.
8.6Known limits
- 3 × 117 MW IT zones on 3 independent 34.5 kV collector sections, radial 415 V. No gas, water, or fiber hydraulic model inside this study.
- Campus-wide 5000 A busway takeoff is not used; N+1 is per section. Do not quote unrepaired conductor totals.
- Faults are first-cycle island subtransient plus IEC 60909 on the worst section. A PE still stamps ETAP before the turbine purchase order.
- Time-to-power is a triangular CPM. Ranking of paths is the result we will defend; the decimal is not.
- The tornado is one lever at a time. Visualizations are not a survey. No IFC / BIM. No sealed ETAP. GPU capex excluded.
- The cooling plant is an hourly energy and water balance, not a CFD or a selected-equipment submittal. Cooler modules carry a planning rating; the CDU rating is published.
- Turbine derate is a screening correlation on inlet temperature and elevation. It sets a fleet count of 16, and an OEM performance deck can move that.