Power as Zoning: Large-Load Compute, Interconnection, and the Repricing of Land

Instrument magnitudes and creditability remain open [VERIFY]. Source dates above cite SB 6, 2025 and Batch Zero [VERIFY].

Where the $100k/MW sits. On 300 MW the stack is $30M — two instruments, not one. The fee is nonrefundable: $15M sunk. The security is a posting, returned, credited against CIAC, or forfeited on milestone performance [VERIFY which]. Both leave the account pre-FID and carry at development-equity cost; only the first must be earned back. At an illustrative 20% hurdle over the 2.6 years to base-case energization, $30M compounds 1.20^2.6 to $48.2M — $161k/MW before construction capital earns anything, and the most junior money in the stack: no lien, no residual, saleable only if queue position is assignable [VERIFY].

What 6-month milestones remove. They convert a near-perpetual option on queue position into dated exercise obligations. That kills queue-position land banking — holding a slot while shopping for a tenant — and moves timing risk onto our balance sheet. Delay is no longer free.

What 100% CIAC moves. Substation and radial work stops being a ratepayer-amortised tariff charge and becomes day-one equity capex that construction debt will not advance against pre-lease [VERIFY]. The asset vests with the transmission provider: no residual, no depreciation shield absent structuring, possibly a tax gross-up [VERIFY]. Magnitude is unknown [VERIFY]; at a $100k/MW placeholder, $30M over 15 years at a 20% hurdle (CRF 0.214) against 2.37m MWh/yr is $2.71/MWh — trivial amortised at zero, not at equity cost.

Node: DFW north load pocket. The node follows the asset: we settle at a load zone, not a generator bus 300 miles west. "Cheap West Texas power" is a generation argument smuggled into a load underwriting — ours only by separately buying transmission or basis. The choice forfeits the negative-priced hours behind western congestion and puts us on the adverse side; a gas-weighted pocket importing renewables over constrained corridors does not inherit that distribution [VERIFY, zonal]. Inputs: mean, P95 and negative hours, five years of hourly load-zone prices [VERIFY].

4CP. Transmission cost is allocated on metered demand in the 15-minute ERCOT peak of each summer month. A flat load has no demand response, so the only lever is storage or on-site generation discharging into those four intervals. Load-factor basis: $6.3–10.1/MWh at 90% LF ($5.7–9.1 at 100%) [VERIFY]. Utilisation compensates for having none.

Congestion is endogenous. The queue we stand in reprices our own pocket: congestion underwritten at signature is not what settles in 2029. Stress at 10 / 25 / 50% of queued in-pocket load energising against a fixed Batch Zero build [VERIFY].

Lead times in weeks — transformer outline-supplied, the rest illustrative, all [VERIFY]: 25–50 MVA transformer 110; gensets 70–100; MV switchgear 40–70; UPS 30–50. Only the first is on the critical path.

Two clocks. Backward from 2029-05-31, less an illustrative 26-week install-and-commission window, less 110 weeks: order by about 2026-10-22 — now, twelve months before the final plan exists. Forward clock + (plan final by ~Nov-2027; later → 2030 Q3). Route certainty costs four quarters, and yields an outcome absent from the table.

The clock we do not control. The transformer is our critical path; the system's is the transmission provider's. The base case implicitly requires the network upgrade built in the ~19 months between autumn 2027 and 2029 Q2 — the most load-bearing [VERIFY] here, and ordering early does nothing to shorten it.

The trade, as a breakeven. Order early if p(orphan) × loss-given-orphan < four quarters pulled forward. Orphaning is not the Downside case — there equipment carries early — but Break plus a route-change variant: p sits nearer 5–15% [illustrative]. LGO is resale haircut plus carry plus spec mismatch, not order value; whether a secondary bid exists at 110-week lead times is [VERIFY] — if not, LGO approaches order value and the trade inverts. At an illustrative 35% LGO and p = 10%, expected loss is ~3.5% of the long-lead package: waiting wins only if four quarters of first-phase margin are worth less than that. Dollarising it needs a unit count and $/MVA we do not have [VERIFY]; the ratio does not.

Queue quality. ~438 GW, ~89% data centre [VERIFY, 2026-06] is an option register, not a forecast — one requirement, several developers filing [VERIFY the duplication factor]. At $100k/MW that implies ~$39B = fee+security on ~390 GW data-centre share ($43.8B full register) [VERIFY collection].

Interruptible designation. Curtailment hours 2021–26 and the duration covering ~95% of events are [VERIFY] against real-time prices and emergency logs. A 24×7 SLA cannot absorb curtailment: the 95th-percentile event is a battery problem, the February-2021 tail a gen-set and contract problem no battery sizes. That battery also shaves 4CP — underwrite both lines.

DA–RT spread. Descriptive only. A must-run load is structurally short real-time, and the spread prices intraday uncertainty. A CFTC-approved exchange listing hourly or binary-settled power adds tenor a bilateral PPA lacks; usability turns on liquidity and basis to our settlement point [VERIFY].

Certainty into exit cap. ERCOT is energy-only [VERIFY any performance-credit mechanism since]: no capacity revenue to capitalise. A PJM capacity ceiling is cross-ISO evidence of repricing speed [VERIFY], not an input — importing it is a category error. Energised with firm interconnection trades tighter than an identical shell; magnitude awaits a brokerage cap-rate survey [VERIFY].