Model version
- vModel v0.7Single-file institutional prototype, updated 05/2026.
- !LimitationsNot legal, engineering, investment, tax, utility-binding, or procurement advice.
Working exhibit / real assets to AI power
How land, real estate, oil, and gas become energy for AI data centers
The AI buildout is not only chips and software. It is a conversion chain: site control becomes grid access, fuel becomes dispatchable electricity, electricity becomes protected facility power, and protected power becomes GPU compute.
Executive summary / scan in 15 seconds
AI data center feasibility is a capital allocation problem
The model screens whether a site can become bankable AI capacity by tying together grid certainty, fuel and generation, long-lead procurement, cooling, offtake economics, financing, geopolitical risk, comparable transactions, and the exact assumptions still missing.
Suggested workflow
- 10 minScreen a site: power deliverability, PUE, CapEx/MW, schedule, fragile variable.
- 30 minPressure-test the deal: DSCR, break-even GPU pricing, utilization, PPA exposure, lead-time slip.
- 2 hrPrepare for IC: assumptions, show-math, diligence checklist, comparable deals, stress tests.
Decision posture
- Base caseGiven current assumptions, what is the modeled project?
- Risk caseGiven external friction, what is the likely energized, financed project?
- Failure caseWhich constraint kills the deal first: grid, transformer, water, offtake, debt, or politics?
Attribution
- AuthorPrepared as a Taiwan Infrastructure Sunday Session exhibit for analyst review.
- UseEarly-stage site screening, memo prep, and conversation framing with utilities, EPCs, lenders, and suppliers.
- Nextv2 live feeds: FERC RSS, ERCOT/EIA APIs, OEM lead-time feeds, NOAA climate, GIS layers.
Interactive model / raw inputs to usable compute
Pick the raw materials and watch the energy chain move
This is a rough strategic simulator, not an engineering model. It shows how fuel, land, grid access, backup fuel, and facility design combine into usable AI data center power.
Natural gas feedstock
Pipeline gas or LNG available for dispatchable generation.
Diesel backup inventory
Fuel stored or contracted for emergency generation.
Solar-ready land
Nearby land that can support solar generation or a solar PPA proxy.
Grid interconnection
The hard cap from utility allocation, substation capacity, and feeder path.
AI rack design load
Average power per liquid-cooled AI rack, including facility design assumptions.
Materials inside the data center
Select materials to see how they change reliability, power conversion, cooling, compute readiness, and construction feasibility.
Assumptions: gas heat rate approximates 7.5 MMBtu/MWh, solar uses 0.25 MW per acre at 22% capacity factor, diesel backup uses 13.5 kWh per gallon, and facility overhead is modeled with PUE 1.22.
Base case feasibility / project finance view
Project finance view: interconnection, CapEx, OpEx, DSCR, schedule
This layer turns the story into a decision model: interconnection, cooling, CapEx, OpEx, compute economics, material lead times, uptime, sustainability, location score, and procurement schedule. Every result exposes the assumptions that drive it.
Reference anchors
- PPUE / ISO 30134-2Infrastructure efficiency: facility power divided by IT power.
- AASHRAE TC 9.9Thermal envelope, rack density, liquid cooling, climate, water, and heat rejection assumptions.
- UUptime Tier styleN, N+1, 2N, and Tier III / IV style redundancy mapping.
- IIEEE 1547 / IEC 61850Interconnection and substation automation reference logic for grid-facing assumptions.
- FFERC Order 2023Queue reform, cluster study pressure, withdrawal discipline, and energization timing risk.
- NNERC reliability standardsBulk-power reliability, operational risk, and transmission dependency framing.
- TTIA-942Data center topology, site, telecom, electrical, and mechanical availability reference.
- 5ISO 50001Energy management system discipline for efficiency and operating control.
- CIEC 62443Operational technology security reference for substations, controls, and facility systems.
[modeled] selecting a regional archetype now updates the base-case inputs
Named benchmark read
Taiwan dense urban inference cluster: references constrained-island grid logic, land scarcity, water stress, and proximity to semiconductor supply chains. This is a strategic-load underwriting case, not a cheap-power case.
Power interconnection
Queue time, deliverable MW, substation distance, voltage, transformer availability, upgrade cost, redundancy.
[input] utility / queue evidence; replace with signed utility letter
[needed] ISO / utility queue; modeled range 12-48 mo
[modeled] large transformer procurement 18-36 mo unless quoted
Cooling architecture
Air, rear-door heat exchanger, direct-to-chip liquid, or immersion with rack density, climate, water, and pump power.
[modeled] AI racks 60-180 kW depending liquid path
[input] municipal / industrial tariff; common modeled band $3-12/kgal
CapEx calculator
Land, shell, electrical, mechanical, generators, transformers, switchgear, cooling, fiber, labor, permitting.
[input] parcel comp; rural power land can be <$0.3M, constrained urban >$1M
[modeled] 1.2-2.4
[modeled] 3.8-7.0
[modeled] 1.8-4.5
OpEx calculator
Power cost, demand charges, water, maintenance, diesel testing, staff, lease, replacement cycles.
[input] PPA / tariff; modeled range $35-110/MWh
[input] tariff sheet; demand exposure often $4-18/kW-mo
AI compute economics
Convert MW into GPU count, rack count, tokens per day, training capacity, inference capacity, revenue, margin.
[input] offtake / platform curve; modeled base 60-85%
[input] lease rate; must be replaced with signed capacity / marketplace evidence
Location and sustainability
Power availability, fiber, land cost, water, permitting, disaster risk, incentives, carbon, PUE, WUE.
Risk-adjusted view / avoided mistakes layer
From calculator to capital allocation intelligence
This layer reframes the model around what infrastructure funds, utilities, hyperscalers, energy traders, EPCs, and procurement teams actually need: interconnection certainty, supply-chain bottlenecks, land intelligence, financing risk, regulatory exposure, and why a site fails.
Killer UX
Why this site failsLoading failure diagnosis.
Real interconnection intelligence
Models the gap between advertised MW and likely energized MW using queue congestion, utility risk, curtailment, transformer status, and evidence quality.
[needed] queue study / utility letter / ISO feed
[modeled] replace with cluster MW ahead and withdrawal rate
[input] local utility posture, tariff pressure, public opposition
[needed] congestion, CRR/FTR, firm vs non-firm service evidence
Procurement and supply-chain intelligence
Tracks the AI-era chokepoints: transformers, switchgear, cooling, ferrites, electrical steel, HBM, substrates, SiC/GaN, busbars, and generators.
[needed] vendor quote dates and PO-to-ship history
[modeled] convert vendor share into HHI by category
[input] bill-of-materials country exposure and substitutes
[needed] OEM quotes, LME copper, GOES, HBM and cooling backlog
GIS and land intelligence
Future live map layers: substations, transmission lines, gas pipelines, fiber, water stress, flood, seismic, permitting, LNG proximity, climate cooling feasibility.
[needed] parcels, substations, lines, fiber, pipeline, hazard layers
[input] ownership, option rights, ROFR/ROFO, easements
[needed] rights, reclaimed supply, drought tier, basin stress
[needed] route distance, diversity, pressure, dark fiber evidence
Financing and risk engine
Adds IRR, DSCR, debt/equity, merchant power exposure, schedule confidence, bankability confidence, and fragile-variable detection.
[input] lender advance rate; typical covenant-driven range 50-70%
[input] all-in coupon / swap / fee assumption
[input] uncontracted power or GPU revenue exposure
[needed] CFIUS/BIS/OFAC/water/permit/export-control review
Scenario library
Comparable archetypes become a moat over time because they make every new site legible against known infrastructure patterns.
[modeled] lease-rate compression and generation obsolescence risk
| Archetype | Core bet | Main failure mode |
|---|---|---|
| Flare gas AI | Cheap stranded energy | Permits, gas continuity, offtake quality |
| Taiwan inference | Dense demand near strategic supply chain | Grid, water, geopolitics, land scarcity |
| Nuclear adjacent | Firm low-carbon baseload | Regulatory timing and transmission rights |
| Texas gas | Behind-the-meter speed | Merchant exposure and interconnection volatility |
| Northern Virginia | Premium demand and fiber density | Dominion queue, substation saturation, local opposition |
| Johor / Malaysia | Singapore spillover and regional growth | Grid expansion, water, policy durability |
| Nordic hydro | Low-carbon power and cooling climate | Fiber distance, latency, local permitting |
| UAE / KSA sovereign | State-backed demand and energy depth | Water, cooling, export controls, sovereign concentration |
Institutional underwriting expansion / maximum density
Everything an infra PM, lender, EPC, utility, or neocloud operator will ask
These panels convert the feedback list into a deal-workbench inventory. Items are tagged as modeled, user input, or needed live evidence so the page stays credible while still showing the full ambition.
Evidence taxonomy
- MModeledCalculated from current assumptions.
- IInputRequires deal-specific user entry.
- NNeededRequires live feed, filing, vendor quote, or counterparty document.
Power and interconnection
- Queue[needed] Cluster phase, MW/MWh ahead, withdrawal rate, tranche probability, deposit at risk, withdrawal penalty.
- ISO/RTO[needed] ERCOT, MISO, PJM, CAISO, SPP, NYISO, ISO-NE, Taipower, KEPCO, TEPCO, J-Power, State Grid regions, EirGrid, National Grid ESO, AEMO.
- LGIA/LIA[needed] Separate generator interconnection and load interconnection paths, timelines, and study regimes.
- Upgrades[input] Participant funding vs socialized cost, affected-system studies, network upgrade cap, substation saturation.
- Services[modeled] Capacity payments, ancillary services, regulation, spinning reserve, frequency response, black start, demand response.
- ERCOT[needed] Coincident peak, 4CP/12CP allocation, CRR/FTR logic, behind-the-meter export rules, basis and curtailment compensation.
- Staging[modeled] Tranche 1 MW, tranche 2 MW, full energization date, confidence bands, islanding and black-start mode.
- Maps[needed] WECC, ERCOT, SERTP, MISO, PJM planning regions and congestion zones.
Fuel and generation
- Gas[input] Tap location, interstate/intrastate, FT vs IT, LDC vs direct pipe, MMBtu/day, hub basis: Henry, Waha, AECO, TTF, JKM.
- LNG[input] Re-gas access, storage sizing, vaporization rate, dual-fuel turbine optionality.
- Turbines[modeled] LM6000, LM2500, Trent, 7HA, 9HA, M501J, Wartsila, Jenbacher, Capstone; heat rate and ramp-rate slots.
- Thermal[modeled] CCGT, simple cycle, cogen, steam host availability, degradation curve.
- Fuel cells[input] Bloom, Doosan, FuelCell Energy, $/kW, stack life, gas or hydrogen feedstock.
- Nuclear[needed] SMR/AMR: NuScale, X-energy, Kairos, TerraPower, Oklo, Holtec; NRC stage, COL timing, FOAK risk.
- Renewables[modeled] Solar tilt/tracker/bifacial/soiling, wind CF, geothermal Fervo/Sage, storage Li-ion/LFP/iron-air/flow/thermal/hydrogen.
- PPAs[input] Physical vs VPPA, proxy revenue swap, fixed-shape vs as-generated, settlement hub, 24/7 CFE premium.
CapEx expansion
- EPC[input] Fixed-price LSTK, cost-reimbursable, target-price, performance guarantees, LDs, schedule LDs.
- Firms[needed] Bechtel, Fluor, Black & Veatch, Burns & McDonnell, Kiewit, DPR, Holder, Skanska, Mortenson, Turner backlog and AI-campus experience.
- Soft costs[modeled] Owner's engineer, legal, financing, PM, commissioning, insurance during construction, builder's risk, contingency.
- Tax[input] ITC 30% base, domestic content, energy community, low-income stacking, MACRS, bonus, transferability 6418, direct pay 6417.
- Local[needed] PILOT, sales tax exemption, property tax abatement, free-port-zone treatment.
- Land[input] Optioned vs owned, ROFR/ROFO, adjacent parcels, transmission/fiber/gas/water/road easements.
- Permits[needed] NEPA EIS/EA/CatEx, Section 404, ESA Section 7, Clean Air Act PSD, NPDES, NHPA Section 106, SEQRA/CEQA equivalents.
- Insurance[input] Builder's risk, GL, professional liability, cyber, business interruption, $/MW benchmarks.
OpEx expansion
- Power[modeled] Energy, demand, transmission, distribution, ancillary allocation, capacity tag, riders, public benefit fund.
- Generation[modeled] Gas heat rate: $/MMBtu x MMBtu/MWh, degradation, overhaul reserve.
- Water[input] Makeup, blowdown disposal, sewer, water rights, drought surcharge, reclaimed availability.
- Tax[input] Property tax jurisdiction and reassessment risk on improvements.
- People[modeled] Ops staff per MW, security, facilities, engineering, NOC, loaded FTE cost.
- Maint.[modeled] Planned 2-4% CapEx/year, unplanned reserve, major overhaul, spares.
- Consumables[input] Diesel testing, fuel cycling, chemicals, water treatment, refrigerant, lubricants.
- Systems[input] Dark fiber/lit service/IRU, route diversity, DCIM, BMS, EPMS, carbon compliance and RECs.
Revenue and offtake
- Structure[input] Bare-metal lease, wholesale colo, hyperscale BTS, neocloud, GPUaaS, training vs inference.
- Credit[needed] Microsoft, Meta, Google, Amazon, Oracle, xAI, Anthropic, OpenAI via Microsoft, CoreWeave, Lambda, Crusoe; guarantees, LCs, support.
- Tenor[input] 10-15yr hyperscaler vs 3-5yr neocloud, rollover mark-to-market risk.
- Payment[input] Take-or-pay, MVC, reservation fee plus usage, power pass-through caps, hedge requirements.
- GPU cycle[modeled] H100, H200, B100, B200, GB200, Rubin/R100, Rubin Ultra utilization decay and residual value.
- Inference[modeled] Tokens/sec/GPU, $/M-token, batching efficiency, latency premium by tier.
- Tenancy[input] Multi-tenant vs single-tenant revenue durability and operational risk.
Financing stack
- Debt[modeled] Senior advance, tenor, amortization, DSCR covenant 1.25-1.40x, construction-to-perm bridge.
- Capital[input] Mezzanine, preferred equity, common equity, tax equity, ITC monetization, reserve accounts.
- Lenders[needed] MUFG, SMBC, Mizuho, JPM, Citi, Santander, BNP, ING, GIP, Brookfield, Macquarie, IFM, Stonepeak, PGIM, MetLife, AllianzGI.
- Bonds[input] 144A private placement, green bonds, project bonds, monoline-style wrap.
- Metrics[modeled] Sponsor IRR, project IRR, equity IRR, DSCR, LLCR, PLCR, MOIC, TVPI.
- Risk[input] DSRA, O&M reserve, major maintenance reserve, GPU replacement reserve, BI and contingent BI.
- Hedges[input] Power, gas, interest-rate swaps, basis exposure, construction debt exposure.
- Public[needed] ECA and multilateral: KEXIM, JBIC, EXIM, EKF, IFC, ADB, EBRD, DFC.
Diligence tracker / what must be true
Conditions precedent, counterparty exposure, schedule, stress, and exit paths
These are the artifacts a credit committee, infrastructure PM, or strategic operator needs before the model becomes financeable.
Term sheet preview
- DDebt62% advance, 8.5% cost, covenant drafted from DSCR output.
- CCPsUtility letter, queue confirmation, transformer PO, EPC LOI, water rights, customer LOI.
Diligence checklist
Counterparty exposure matrix
| Counterparty | Exposure | Evidence | Risk |
|---|---|---|---|
| Utility / ISO | 0MW / 0mo | Service letter, queue study, energization history | Substation saturation, upgrade allocation |
| EPC | $0M / 0mo float | LOI, GMP, LDs, backlog | Schedule slip, change orders |
| OEM | 0mo lead | Transformer, turbine, cooling, UPS, switchgear quotes | Single vendor, PO-to-delivery slippage |
| Lessee | $0M/yr | Credit rating, parent guarantee, LC | Rollover, GPU price compression |
| Fuel supplier | $0M/yr | Firm transport, storage, hub basis hedge | Interruptible supply, basis blowout |
| Insurer / lender | $0M debt | Term sheet, exclusions, reserve requirements | Deductible hardening, covenant cushion |
Critical path Gantt
Queue
18mo
Transformer PO
22mo
Site mob
8mo
Vertical
13mo
MEP
15mo
Energize
5mo
GPU install
5mo
ATO / revenue
4mo
Sensitivity matrix
Live DSCR grid: GPU $/hr x utilization. Each cell recomputes DSCR from the current CapEx, OpEx, and debt stack.
Awaiting model.
GPU-generation utilization decay
Live obsolescence curve. Current utilization is stressed through likely generation rollover and residual pricing compression.
| Awaiting model. |
Monte Carlo and stress tests
- P10Awaiting model.
- P50Awaiting model.
- P90Awaiting model.
- ShockAwaiting model.
Exit scenarios
| Exit | Buyer | Valuation logic |
|---|---|---|
| Refi | Project lenders / privates | Lower rate after COD and stabilized DSCR |
| Sale | Infra fund | Cap rate / EV-EBITDA on contracted cash flow |
| Strategic | Hyperscaler / neocloud | Control premium for energized scarce MW |
| REIT | DLR / EQIX / IRM-style contribution | Lease coverage, tenancy, remaining term |
| IPO carve-out | Public equity | Growth MW, backlog, contracted revenue |
Carbon ledger
| Line | What to track | Exposure |
|---|---|---|
| Scope 1 | Diesel testing, behind-meter gas, emergency runtime | Fuel price, emissions permit, local air limits |
| Scope 2 | Location-based and market-based grid energy | Carbon intensity, REC/PPA quality, hourly matching |
| Scope 3 | Transformers, steel, concrete, cooling, GPUs, HBM, optics | Embodied carbon, CBAM, supplier disclosure |
| Offsets | REC, I-REC, VPPA, 24/7 CFE, additionality | Residual exposure and buyer claims risk |
Insurance reserve table
| Peril | Coverage | Watch item |
|---|---|---|
| Construction | Builder's risk, delay in startup | Deductible, exclusions, weather named peril |
| Operations | Property, GL, equipment breakdown | Transformer replacement time, spare strategy |
| Revenue | Business interruption, contingent BI | Waiting period and utility outage language |
| Cyber | IT/OT incident response, ransomware, data impact | IEC 62443 controls, insurer carve-outs |
Regulatory clock
- Count[needed] Permits required, authority having jurisdiction, filing status, appeal risk, public comment windows.
- Longest pole[needed] Air PSD, water discharge, Army Corps, environmental review, zoning, utility energization, or fuel infrastructure.
- Date[input] Required COD, first revenue date, financing long-stop, interconnection milestone, equipment delivery milestone.
- Failure[input] Permit path slips beyond financing window or equipment delivery creates idle CapEx once dates are supplied.
Market comps / signal library
Comparable deals, public marks, supplier chokepoints, and regulatory clocks
Named comparables are included as diligence targets, not asserted live market data. Exact terms should be sourced from filings, press releases, earnings calls, lender decks, and utility records.
News / signal feeds
- Nv2 feedsFERC filings, ISO reports, utility queue updates, OEM earnings, EIA-860, NOAA, GIS.
Recent deal benchmark watchlist
| Reference | Read-through | Evidence needed |
|---|---|---|
| Stargate / OpenAI / Oracle / Crusoe | Strategic AI campus and power-led execution | MW, offtake, financing, energization tranches |
| Microsoft-Brookfield 10.5GW | 24/7 clean-power procurement scale | PPA tenor, regions, settlement shape |
| Meta-Entergy Louisiana | Utility-backed load growth and generation planning | Tariff, generation additions, transmission upgrades |
| Amazon-Talen Susquehanna | Nuclear-adjacent compute controversy | FERC/utility treatment, power rights, timing |
| Google-Kairos / AWS-X-energy | SMR/AMR strategic supply | License stage, FOAK risk, COD range |
| Constellation-Microsoft TMI | Nuclear restart PPA | Restart capex, PPA terms, regulatory path |
| CoreWeave Plano / Lambda Texas / Crusoe Permian | Neocloud and behind-meter patterns | Lease tenor, GPU economics, energy basis |
| Vantage AZ / QTS Manassas / DLR / EQIX | Hyperscale and REIT comparables | $/MW, cap rate, tenant, lease coverage |
| IREN / Applied Digital / TeraWulf / Cipher / Bitdeer | Bitcoin-to-AI pivot comps | Power cost, conversion capex, customer credit |
Public comparable valuation
- REIT/DC[needed] DLR, EQIX, IRM, AMT: EV/EBITDA, backlog, lease coverage, $/MW IT capacity.
- AI infra[needed] IREN, APLD, NBIS, CRWV: powered land, contracted MW, GPU availability, counterparty credit.
- Private[needed] $/MW by stage: raw land, shell-ready, energized, leased, occupied.
- Valuation[modeled] Stabilized colo vs hyperscale BTS vs powered shell cap rates; power-shell vs Tier III turnkey delta.
- Metric[modeled] Land $/MW, CapEx/MW, IT MW after PUE, EV/EBITDA, break-even GPU $/hr.
Constrained supply chain
- LPT[needed] Hitachi Energy, Siemens Energy, GE Vernova, Mitsubishi Electric, Hyundai Heavy, Hyosung, JSHP, TBEA; ANSI/IEC, GSU/station/unit auxiliary.
- Steel[needed] GOES exposure: Nippon Steel, POSCO, AK/Cleveland-Cliffs.
- Switchgear[needed] MV/LV, vacuum vs SF6 vs solid-state, ABB, Eaton, Schneider, Siemens, GE, Mitsubishi, Toshiba.
- Grid quality[input] GIS vs AIS, STATCOM/SVC, harmonic and power-quality impact from AI loads.
- Power[needed] MV cable, copper LME, aluminum substitution, UPS Li-ion/VRLA, Vertiv, Schneider, Eaton, ABB, Mitsubishi, Riello, Delta.
- Cooling[needed] CRAH/CRAC/CDU/rear-door/immersion: Vertiv, Stulz, Schneider, Munters, Nortek, Motivair, CoolIT, JetCool, LiquidStack, Submer, GRC, Iceotope.
- Chillers[needed] Trane, Carrier, York, Daikin, Mitsubishi; air/water/adiabatic; R-410A to R-32/R-454B exposure.
- Gensets[needed] Caterpillar, Cummins, MTU, Mitsubishi, Kohler, Generac; EPA Tier 4 status, diesel/propane tanks, ULSD polishing, water-in-fuel risk.
- Compute[needed] HBM3e, CoWoS, substrates, optics, InfiniBand/Spectrum-X/Arista/Cisco, BlueField/Pensando, SiC/GaN suppliers.
- Fire[needed] FM-200 to Novec to 3M phase-out and fluorine-free alternatives; suppression availability and insurance acceptance.
Geopolitical and regulatory risk
- Trade[needed] Section 301, Section 232, USMCA RoO, EU CBAM steel/transformer exposure.
- Exports[needed] BIS advanced GPU rules, country tiers, EAR99 vs ECCN classification.
- FDI[needed] CFIUS, FIRB, FOCI, ICTS, UK/EU/Japan/Taiwan reviews.
- Sanctions[needed] OFAC SDN lessee and vendor screening, secondary sanctions.
- Data[needed] GDPR, China DSL/PIPL, Taiwan PDPA, EU AI Act, US state laws.
- Local[needed] IRA domestic content, Buy American, EU Net Zero Industry Act, India PLI.
- Water[needed] Prior appropriation, riparian, Colorado River, Great Lakes Compact, local rights regimes.
- NIMBY[needed] County denial history, organized opposition, NIETC/FERC backstop authority.
Climate, water, sustainability
- WUE[modeled] On-site vs off-site water; upstream generation water intensity and energy-water nexus.
- Carbon[modeled] Scope 1, scope 2 location vs market, scope 3 equipment and GPU manufacturing.
- 24/7 CFE[input] Hourly matching, additionality, Google/Microsoft-style procurement premium.
- Heat reuse[input] District heating, greenhouse, aquaculture, revenue or credit value.
- Hazards[needed] FEMA 100/500-year, First Street/Jupiter, wildfire, PSPS, ASCE 7 seismic, liquefaction, cyclone wind class.
- Climate[needed] RCP 4.5/8.5 extremes, Aqueduct 2030/2040 drought stress.
- Ratings[input] LEED, BREEAM, ENERGY STAR, SBTi alignment.
Operational risk
- Cyber/OT[needed] IEC 62443 zones, NERC CIP applicability, IT-OT segmentation.
- Physical[input] ASIS, K-rated barriers, CPTED, drone defense, insider threat.
- HHI[modeled] Vendor concentration by transformers, switchgear, cooling, GPU, optics, fuel.
- SPOF[needed] Single substation, fiber route, fuel supplier, water source.
- Insurance[input] Peril capacity, deductibles, exclusions, business interruption.
- GPU ops[modeled] Failure rates, warranty, extended support, CUDA/driver/hypervisor dependency.
Methodology appendix / formulas and audit trail
Every output needs an assumption, formula, source, confidence, and sensitivity
This section is the memo appendix: formula inventory, acronym glossary, audit trail requirements, print/export framing, and v2 feed roadmap.
Print/export
- POne-pagerPrint stylesheet hides controls and preserves outputs, math, risks, comps, and limitations.
Show-math inventory
- MWBankable facility MW, likely energized MW, tranche MW, PUE-adjusted IT MW.
- CapExLand + shell + electrical + mechanical + generators + fiber + grid upgrade + soft + contingency - incentives.
- OpExEnergy + demand + water + maintenance + staff + lease + diesel + consumables + connectivity + carbon.
- RevenueGPU count x hours x utilization x $/GPU-hour; inference token model optional.
- PUE/WUECooling profile x climate adjustment x density penalty; WUE from architecture and climate.
- ScheduleMax(queue, transformer, generator, EPC mobilization) + redundancy + distance penalty.
- ConfidenceEvidence, materials, schedule, location, finance risk, data provenance haircut.
- ScoresMaterials readiness and location score by weighted component.
Tornado expansion queue
- GPUUtilization decay by GPU generation, resale residual, lease rate compression.
- PowerCurtailment, dispatch exposure, PPA basis blowout, demand charge ratchet.
- Water/taxDrought-tier water step-up, property tax reassessment.
- ScheduleTransformer slip +6 months, EPC mobilization delay, interconnection restudy.
- MarketsInsurance hardening and carbon price where applicable.
Glossary
| LGIA | Large Generator Interconnection Agreement. | LIA | Load Interconnection Agreement. |
| LCOE | Levelized cost of energy. | CCGT | Combined-cycle gas turbine. |
| 4CP/12CP | Coincident peak transmission allocation methods. | CRR/FTR | Congestion revenue rights / financial transmission rights. |
| DSCR | Debt service coverage ratio. | LLCR/PLCR | Loan/project life coverage ratios. |
| CDU | Coolant distribution unit. | EPMS/BMS | Electrical power / building management systems. |
| FOAK | First of a kind. | NIETC | National Interest Electric Transmission Corridor. |
| PUE/WUE | Power and water usage effectiveness. | ITC/PTC | Investment and production tax credits. |
| MACRS | Modified accelerated cost recovery system. | EPC/LSTK | Engineering, procurement, construction / lump-sum turnkey. |
| LD/GMP | Liquidated damages / guaranteed maximum price. | LOI/LC | Letter of intent / letter of credit. |
| PSD/NPDES | Air permit review / water discharge permit. | NEPA | National Environmental Policy Act. |
| EIS/EA/CatEx | Environmental impact statement / assessment / categorical exclusion. | ESA/NHPA | Endangered Species Act / Historic Preservation Act. |
| SEQRA/CEQA | New York / California environmental review laws. | IRA/PILOT | Inflation Reduction Act / payment in lieu of taxes. |
| FERC | Federal Energy Regulatory Commission. | RTO/ISO | Regional transmission organization / independent system operator. |
| ERCOT/MISO | Texas grid operator / Midcontinent ISO. | PJM/CAISO | Mid-Atlantic grid / California ISO. |
| SPP/NYISO | Southwest Power Pool / New York ISO. | ISO-NE | New England ISO. |
| IEEE 1547 | Distributed resource interconnection standard. | IEC 61850 | Substation automation standard. |
| IEC 62443 | Industrial control security standard. | NERC CIP | Bulk electric cyber reliability standards. |
| ASHRAE TC 9.9 | Data center thermal guidance. | ISO 30134-2 | PUE standard reference. |
| TIA-942 | Data center infrastructure standard. | ISO 50001 | Energy management standard. |
| GIS/AIS | Gas-insulated / air-insulated substation. | STATCOM/SVC | Reactive-power and voltage support equipment. |
| HBM3e/CoWoS | High-bandwidth memory / TSMC packaging. | SiC/GaN | Silicon carbide / gallium nitride power semiconductors. |
| LME/GOES | London Metal Exchange / grain-oriented electrical steel. | UPS | Uninterruptible power supply. |
| CRAH/CRAC | Computer room air handler / conditioner. | MV/LV | Medium voltage / low voltage. |
| GSU | Generator step-up transformer. | ROFR/ROFO | Rights of first refusal / first offer. |
| MVC/PPA/VPPA | Minimum volume commitment / power purchase agreement / virtual PPA. | CFE | Carbon-free energy. |
| BIS/CFIUS | Export control agency / foreign investment review. | FIRB/FOCI | Australia review board / foreign ownership control influence. |
| OFAC/GDPR | Sanctions office / EU privacy regulation. | PIPL/DSL/PDPA | China privacy/data laws / Taiwan privacy law. |
| KEXIM/JBIC | Korea and Japan export-credit agencies. | EXIM/EKF | US and Denmark export-credit agencies. |
| IFC/ADB | International Finance Corporation / Asian Development Bank. | EBRD/DFC | European reconstruction bank / US development finance. |
| REC/I-REC | Renewable energy certificates / international RECs. | CBAM/RGGI | EU carbon border adjustment / northeast US carbon market. |
| SBTi/LEED | Science Based Targets initiative / green building certification. | BREEAM/RCP | Building sustainability rating / climate pathway. |
| COD/ATO | Commercial operation date / authority to operate. | BTS/GPUaaS | Build-to-suit / GPU as a service. |
| NOC/DCIM | Network operations center / data center infrastructure management. | P10/P50/P90 | Downside/base/upside probability cases. |
| HHI/SPOF | Concentration index / single point of failure. | IRU | Indefeasible right of use for fiber. |
Audit trail and change log
- AuditTrack every assumption changed, timestamp, user, old value, new value, source document.
- Changedv0.7 adds institutional underwriting panels, extended headline metrics, expanded scenarios, comps, diligence, supply chain, regulatory, climate, operational, and methodology sections.
- RepeatFuture session: highlight what changed since last load for returning analysts.
01
AI buyers create the power mandate
Plain English: Model training and inference buyers need GPU clusters, and GPU clusters need huge blocks of reliable electricity.
Inputs
- AI workloads
- GPU demand
- Latency needs
- Uptime terms
What changes
Power becomes the first site-selection filter, not a utility detail handled later.
Test
Can the site support the buyer's MW, timeline, cooling load, and reliability requirement?
compute demand
demand signal
turn demand into a site search
02
Real estate becomes a power platform
Plain English: The parcel matters because it can host infrastructure, not just because it has acreage.
Inputs
- Land control
- Industrial zoning
- Fiber route
- Substation distance
What changes
A plain industrial site becomes a scarce AI infrastructure asset if the power path is real.
Test
Who owns it, what is entitled, and how many MW can be delivered without heroic upgrades?
site control
land gate
secure rights and permits
03
Interconnection turns land into capacity
Plain English: The utility or grid operator studies whether the site can actually receive the requested load.
Inputs
- Load application
- Substation study
- Feeder route
- Upgrade cost
What changes
The story moves from "we have land" to "we have a credible power delivery path."
Test
What power is available now, what is expandable, and what date can be contracted?
grid access
grid path
fuel supports the generation stack
04
Gas becomes dispatchable power
Plain English: Natural gas can feed power plants or dedicated generation that supports the data center load.
Inputs
- Gas production
- Pipelines or LNG
- Gas turbines
- Power contracts
What changes
Fuel is converted into electricity that can run continuously or fill gaps when renewables are intermittent.
Test
Is the gas supply firm, permitted, priced, and acceptable under the buyer's carbon requirements?
gas to electrons
fuel path
oil mainly protects uptime
05
Oil supports backup and logistics
Plain English: Oil is rarely the main power source, but diesel backup and fuel logistics protect uptime.
Inputs
- Diesel gensets
- Fuel storage
- Refill contracts
- Testing schedule
What changes
The site can ride through grid outages long enough to avoid compute downtime or data loss.
Test
How many hours of backup exist, and can fuel be delivered during a regional disruption?
backup fuel
resilience
condition power inside the campus
06
Electrical equipment makes power usable
Plain English: Transformers, switchgear, UPS systems, PDUs, and busbars move power safely into the racks.
Inputs
- Transformers
- Switchgear
- UPS and batteries
- Rack distribution
What changes
Bulk electricity becomes stable, protected facility power that expensive GPUs can trust.
Test
Are long-lead components ordered, redundant, certified, and sized for AI rack density?
power train
delivery
cool the load
07
Cooling converts electricity into usable compute conditions
Plain English: AI racks turn electricity into heat, so cooling is part of the energy system.
Inputs
- Liquid cooling
- Chillers
- Pumps
- Heat rejection
What changes
The facility can keep dense racks inside thermal limits instead of throttling or shutting down.
Test
Can the site support water, heat rejection, and high-density liquid cooling at scale?
thermal system
cooling
final conversion into AI output
08
GPU clusters turn power into AI services
Plain English: The final asset is not the building. It is reliable, contracted compute capacity.
Inputs
- GPU racks
- Networking
- Storage
- Buyer contracts
What changes
Land, fuel, grid rights, and electrical systems become the operating base for AI revenue.
Test
Can the operator sell dependable GPU hours with the power, cooling, and uptime buyers require?
AI capacity
compute
The simple version
AI turns real estate into energy infrastructure, then turns energy infrastructure into contracted compute.
Archetype verdicts
Taiwan dense urban inference
Status: strategically valuable but power-constrained. Binding constraint is grid deliverability, water/cooling margin, and geopolitical confidence. Verdict: bankable only with hard utility evidence and supplier-backed schedule.
Arizona solar corridor
Status: land and renewable story are strong, but water and interconnection queue risk dominate. Verdict: viable when paired with firming, liquid cooling design, and credible water rights.
Texas behind-the-meter gas
Status: speed and energy economics can work. Binding constraints are merchant exposure, emissions compliance, and generator/transformer procurement. Verdict: attractive if offtake quality and grid export rules are solved.
300MW hyperscale campus
Status: institutionally legible but schedule-heavy. Binding constraint is procurement sequencing across transformers, switchgear, cooling, HBM, substrates, ferrites, and EPC labor. Verdict: financeable only as staged energization.