Baseload
DemoThe Portfolio
| Project | CapacityIT MW | All-in$ per IT MWh | Annual$ per yr | CODmonth | |
|---|---|---|---|---|---|
Pacific Northwest Non-RTO / BPA | |||||
PJM West PJM | |||||
▸MISO South MISO |
Total committed: 1.60 GW IT · 1.86 GW at the meter. The capacity field is a lever — resize any commitment and the whole instrument re-prices.
The portfolio, over time
energizations · docketed ramp (50/65/80/90%) · spend vs loadThe snapshot above, unrolled onto the calendar: construction capital draws flat across each project's wait to its COD month (the ledger's own draw), load arrives on the docketed hyperscale ramp stepping on contract anniversaries from that month — contracts step with it, so the demand charges do too — and the fixed charges run at full from energization, halls full or not. Portfolio reaches 1.86 GW at full ramp in 2035.
| Year | Load MW | Ramp | TWh | Outlay | Operating | Fixed | Energizes |
|---|---|---|---|---|---|---|---|
| 2026 | 0 | — | 0.0 | $5.38 B | — | — | |
| 2027 | 0 | — | 0.0 | $5.38 B | — | — | |
| 2028 | 295 | 15.9% | 2.1 | $3.35 B | $125 M | $671 M | MISO South · Jul 2028 |
| 2029 | 679 | 36.5% | 4.8 | $1.33 B | $287 M | $1.34 B | |
| 2030 | 856 | 46.0% | 6.0 | $1.33 B | $362 M | $1.34 B | |
| 2031 | 1,343 | 72.2% | 9.4 | — | $637 M | $2.28 B | Pacific Northwest · Jan 2031, PJM West · Jan 2031 |
| 2032 | 1,563 | 84.0% | 11.0 | — | $750 M | $2.28 B | |
| 2033 | 1,724 | 92.7% | 12.1 | — | $839 M | $2.28 B | |
| 2034 | 1,792 | 96.3% | 12.6 | — | $882 M | $2.28 B | |
| 2035 | 1,860 | 100.0% | 13.0 | — | $924 M | $2.28 B | |
| 2036 | 1,860 | 100.0% | 13.0 | — | $924 M | $2.28 B |
Ramp: AEP Schedule DCT Load Ramp Contract Capacity steps (PUCO 24-508-EL-ATA) on contract anniversaries from each project's COD month — a calendar year prints the mean of its twelve monthly shares, ramped load over full-ramp load — and ramp contracts step the demand determinants with the load · outlay drawn flat by month across each project's wait (2026 commit) to its COD month, the ledger's own draw, so §VII's capital block foots to it · fixed charges (annualized capital + O&M) at full from energization · energizes = the COD month the ledger dates (rounded to the month where a slip lands off the grid) · directional.
Lead with MISO South — low all-in cost and quick to power. Weigh PJM West against the leaders — a high power price and a long road to energization — and hold it only for what the cost model doesn't price (fiber density, ecosystem, optionality).
Of the $838M open power position, $244M clears on ISO hubs (PJM Western Hub) — the curve-tradeable slice — while $593M is tariff-priced or bilateral, hedgeable contractually (PPA, or OTC indices like Mid-C) but not on a cleared curve (§IV bands the whole exposure). Locking ~50% of the hub slice narrows its ±20% hub-price swing from ±$49M to ±$24M for a $6M forward premium.
Directional screen off the solved portfolio — reweight the sliders in §II and the call re-derives. Not a trade, not investment advice.
The full brief behind the call — printable — is §XII, The Memo.
The Scenario Desk
The shock study
same book, two worlds — deltas are causal, not correlationalPin the world you believe, then move any lever — the gas rip, the rate move, the walk-down. Both worlds re-solve on the current sites through the same registry, so every difference below is the lever's doing and nothing else's. This is what volatility actually does to the financials, line by line and project by project.
The two worlds are identical — move a lever anywhere (§II is the lever bank; the tape's gas read enters there) and the deltas print here.
The street corroboration
the model's structure, priced by real balance sheets — every figure filedAug 31, 2026: crude jumped 3.4% on the US–Iran strikes. Here is what that day did to two public companies on opposite sides of the power trade — from their SEC filings — and the lever in this instrument that explains each. Both cards load a runnable case into the shock study above.
A gas-levered generator that should love an oil rip — and its stock didn't move, because the book is sold forward through 2027. Their 100/94/72 ladder IS this instrument's contract-cover lever walked through time: the shock only reaches the open rung. Run it at 90% cover, then run the CoreWeave card — same shock, bottom of the ladder — and compare the landings.
An AI-cloud buyer whose 10-Q calls power costs “volatile and unpredictable” — and whose lenders wrote power-cost hedging into the loan covenant. The banks priced this instrument's cover lever before the company hedged. Run their 2030-scale book, lightly covered, through the same gas move.
Filed sources · VST: Q2 2026 release (Aug 7 '26; hedge as-of Aug 3) + FY2025 10-K Item 7A (VaR; no $/MMBtu sensitivity disclosed), closes via stockanalysis.com · CRWV: Q2 2026 10-Q (Aug 12 '26) — power inside cost of revenue, MD&A +$87M utilities/power YoY, PPA derivative marks $0, DDTL 4.0 “includes certain power cost hedging requirements” · These cases map each company's STRUCTURE onto this instrument's book — they are structural analogies at portfolio scale, not reconstructions of either company's actual sites or contracts · directional.
The assumptions register
every given the engine solves from — 60 inputs, each with its sourceNothing saved yet — set the levers to a case worth keeping and save it under a name; the comparison appears once one case is saved. Every case carries a share link and an API workpaper URL.
Strategy, Utility and Market Desks
The flows the tape trades on: this portfolio alone adds 0.30% to national electricity demand, implies 0.15 Bcf/d of incremental gas burn at regional marginal mixes, and puts $18.79 B of construction financing into the market. Below, the live grid those flows land on — national first, then your committed markets, then the detail. Scale any of it by the number of portfolios being built at once.
Live market data · GridStatus · directional, monitoring only — describes flows, not securities. Not investment advice.
Loading live market data — prices, the national grid, generation mix, and curtailment…
The market files
the full dossier on all twelve — committed or notEverything the instrument knows about a market, before you commit a megawatt to it: delivered-price basis, grid headroom, the tariff framework a large load actually signs, where a hedge would settle, and the structural signals behind the time-to-power figure.
Site of Meta's largest announced AI campus — originally 2 GW, expanded to 5 GW in 2026; a regulated utility building dedicated generation compresses time-to-power.
no liquid hub — Thin hub liquidity in MISO South — hedging is largely bilateral (PPA).
MISO gen queue ~382 GW footprint (Dec '25) · ~5 yr to power
HeadroomA regulated utility building dedicated generation compresses timelines; MISO South is a relatively uncongested capacity area.
LPSC U-37425 · minimum bill covers 100% of new generation cost · exit: termination fee
Tariff basis · LPSC U-37425 (approved Aug 20 2025): Entergy/Meta Richland Parish settlement — ~2.3 GW dedicated CCGTs, 15-yr minimum bill covering 100% of new generation cost. Bundled utility contract inside MISO — priced here as tariff proxy, not hub + PRA.
The Risk Desk
| Project | Counterparty | Cover | Strike $/MWh | From | To | Covered $M/yr |
|---|---|---|---|---|---|---|
| Fully open at the levers: no contract to derive. Set a cover, or edit as instruments and add one. | ||||||
Fully open: no volume under contract, every megawatt-hour rides the market. Take a preset or set a cover — the whole instrument re-prices, and the exposure tiles above shrink to the open remainder.
One portfolio-wide book, financially settled — the modeled primitive behind every 2025–26 headline deal shape (nuclear PPA, solar VPPA, utility special contract). Strikes illustrative; a licensed curve prices the real book. Tenor flows into the NPV as a two-annuity split — inspect any energy figure for the algebra. A counterparty is a label; it never moves a number.
Fully floating: every dollar of the power bill rides the market. Of the $838 M/yr that floats, $244 M/yr clears on ISO hubs — the only slice a forward can lock; the rest is tariff or bilateral. Drag to hedge the hub slice and watch the range tighten.
Pacific Northwest → Mid-C (bilateral index). Hedge at Mid-Columbia; basis tracks BPA transmission and hydro conditions.
PJM West → PJM Western Hub. Liquid PJM hub; basis is node-to-hub within PJM.
MISO South → no liquid hub. Thin hub liquidity in MISO South — hedging is largely bilateral (PPA).
Forward priced as spot +5% (illustrative contango); a live forward curve is a licensed feed. Structuring analytics only — Baseload does not execute trades or provide investment advice.
The risk framework
four dimensions · one engine · every figure re-solves liveThe exposures a risk committee walks through, quantified from the same solve as every other number on the page — and the section where each one is held, priced, and managed. Move any lever and this table re-prices with the book.
Hedged in §IV (contract book, hedgeable share, basis by hub) · gas lever §II · rate legs §VI
Terms as data in the §III market files · collateral quantified from the §V bill engine's own minimums
Priced in §V (walk-down lever, dead weight, exit fees) — floors are engine equations, not footnotes
Slip lever §II · peak levers §V · cooling design §V · bridge option (Speed vs Wait, §VI)
Market swings from the §IX tornado (re-solved per driver) · volumetric at a 70% walk-down through the §V floors · collateral from the anchor tariffs' own minimum-charge schedules · operational at the construction financing leg · directional, sources on each underlying panel.
The clean book
hub-settled vPPA · as-generated P50 · bundled RECs retiredA contract-for-differences against the same hub the delivered price is built from: you pay the strike, you receive the captured value of what the project actually generates, and the RECs come bundled. Two numbers decide it — the capture rate, which is falling wherever solar is winning, and the cost per ton of the claim it buys.
The cheapest strike in the country — and the steepest cannibalization: capture fell 0.93 → 0.65 → 0.57 in two years as solar penetration passed 13%.
Size a tranche to price the book — nameplate above zero.
Strikes: LevelTen P25 offer index, Q4-2025 (ERCOT solar $49, PJM solar $81.03) and continental wind $74.69 (Q3-25) → $79.40 (Q1-26), directional · capture rates: Potomac ERCOT SOM 2025 (solar 0.57, wind ~0.51), LBNL value factors (PJM solar 1.12 incl. capacity value, 1.0 energy-only) · CFs: ERCOT SOM Table 2, PJM IMM Table 5-35, LBNL fleet · settlement hub-priced, seller holds node-to-hub basis inside the strike (Norton Rose Fulbright; Pexapark 2025) · negative hours floor at $0, premium inside the strike (Pexapark 2024) · hourly-equivalent ≈70% of annual per Princeton ZERO Lab (Google reported 66% hourly at 100% annual, CY2024) · market-based Scope 2 per GHG Protocol, annual matching · directional.
The Demand Charges & The Bills
No ERCOT site committed — priced on the next 200 MW in ERCOT North (commit it in §I). ERCOT's energy looks cheap; this is the line the energy price doesn't show, and the one a flexible load can manage down.
ERCOT's flexible loads curtail through forecast peak windows at scale — commercial forecasts hit all four 2025 peaks. The cost is a few dozen curtailed hours a summer; the prize is $15 M/yr of transmission bill. The lever re-prices the whole instrument — the §VI allocation table included. Same lever lives in §II.
The single highest ERCOT-wide 15-minute interval of each month sets the year. Intervals are only official in December — avoidance means curtailing through every plausible window.
ERCOT load unavailable — the market feed is offline.
Rate $67/kW-yr per GridStatus & NRG's PUCT remarks; cross-check: 2025 TCOS $5.56B ÷ avg 4CP peak 80.9 GW ≈ $69 · billed the following year · rule risk: PUCT's proposed 12CP + minimum-demand-charge methodology (due Dec 2026) would narrow the avoidance play · directional.
The committed 400 MW in PJM West pays capacity on its PLC — set by its draw through the five highest summer RTO peaks — and NITS on its NSPL, the zonal single annual peak. Six 1-hour windows set two charge determinants for the whole following year — same physics as the 4CP, and the same lever works.
Harder than ERCOT's four: the best published backtest (Amperon, 2012–22) called all 55 PJM 5CP days — at the cost of ~46 false alerts, so ~10 curtailment days a summer to catch five. PJM's own stakeholder FAQ concedes peak shaving lowers the PLC; the caveat is you can't double-dip — load reductions claimed as DR on a 5CP day don't reduce the PLC. The ceiling here is 80%, not ERCOT's 95%.
PLC: avg draw through the five highest PJM RTO daily peaks, Jun–Sep, published mid-Oct, effective next Jun 1 · NSPL: zonal 1CP, resets Jan 1 · at the $329/MW-day FERC cap a shaved MW is worth ~$120k/yr of capacity alone · sources: PJM SCRSTF FAQ; Amperon CP backtest (May 2023); tariff Attachment DD / OATT · directional.
The bill, built
actual tariff sheets, line by line — then reconciled to the engineThe engine's wires and delivery lines are directional; a utility's bill is not. Here two anchor tariffs are computed the way the utility computes them — and the difference against the engine's basis is printed, not smoothed. Where the gap is material, the bill wins.
| Line | Rate | Billed on | $ / yr |
|---|---|---|---|
| Basic Transmission Cost Rider — demand ⓘ | $6.74 /kW-mo | 456k kW | $37 M |
| Basic Transmission Cost Rider — energy ⓘ | 0.0422 ¢/kWh | 3,196k MWh | $1 M |
| Ohio kWh excise tax (tail block) ⓘ | 0.3630 ¢/kWh | 3,196k MWh | $12 M |
| PIPP universal-service rider (tail block) ⓘ | 0.0176 ¢/kWh | 3,196k MWh | $561 K |
| Customer charge — transmission ⓘ | $6,800 /mo | 12 mo | $82 K |
| Supplemental customer charge — New Load ⓘ | $20,000 /mo | 12 mo | $240 K |
| Percent-of-distribution riders (ESRR + DIR + EDCRR, ~13%) ⓘ | $884 /mo | 12 mo | $11 K |
| The wires bill | $51 M | ||
Minimum billingBilling demand = highest 30-min peak; post-ramp it is floored at the greater of 85% of the trailing-11-month peak (a ratchet) or the Minimum Demand table — capped at 85% of Contract Capacity. During ramp: ≥85% of Load Ramp Contract Capacity (which steps 50/65/80/90% over 4 years). A customer that never energizes is billed at 100% of Contract Capacity.
The readEnergy + capacity are shopped (CRES supply or SSO) — the market lines stay with the engine's PJM build-up. gridSMART ($22.68/mo) and storm rider ($1.02/mo) omitted as de minimis; reactive assumed unity-PF.
The lever runs through the whole engine: energy and the avoidable 4CP fall with the load everywhere, while floored demand lines fall only to their docket minimums — portfolio dead weight $0/yr at this setting. Rent (§VI) never falls: the lease bills reserved kW — the landlord holds the same floor the utility does.
Aug 28, 2026 · https://www.aepohio.com/lib/docs/ratesandtariffs/Ohio/September_2026_AEP_Ohio_Tariff_Book.pdf · https://puco.ohio.gov/news/puco-orders-aep-ohio-to-create-data-center-specific-tariff
The water line
on-site cooling only · the constraint before the costCooling design sets the draw: the hyperscale fleet spans Meta's ~0.2 L/kWh (air-heavy) to Google's ~1.0 (evaporative-heavy) — a strategic choice, not noise. The dollars are small; the gallons are what a city council, a drought clause, or a permit hearing prices.
Per-site cooling designs live in the §I tune panel — dry-cool the desert site, run towers where water is cheap.
WUE bands: Introl 2025, Equinix 2024 (evaporative 1.99–4.0 · hybrid 0.3–1.5 · closed ≈0.1) · fleet: Meta ~0.20, Microsoft ~0.30, Google ~0.84–1.1 implied (Koomey) · prices: Dallas $5.11, Mesa $3.24, Loudoun $3.53–6.07 (reclaimed is the cheap tail), Atlanta $5.57; $4.50/kgal elsewhere · on-site (Scope 1) only — water embedded in purchased electricity (~3–4.5 L/kWh US avg, LBNL 2024) excluded · directional.
The Capital Stack
Discounted through a real capital structure: a weighted cost of capital of — equity at 11.0% on a 40% layer, debt at 6.8% (60%) after the 25% tax shield, both priced off the live 10-year below. The flat-rate CRF screen is the sticker; the after-tax number is what the build actually costs the balance sheet.
CRF is the pre-tax sticker at a flat rate; the DCF puts every cost on one after-tax basis — capital at the WACC net of the depreciation shield, operating cost at (1−τ), since opex is deductible too. Straight-line over the tax life — the MACRS toggle in the cashflow table below would front-load the shield. The NPV is the closed form's screening figure: capital carried at the half-draw K′, the ramp flat, tax on the operating clock; the calendar-dated ledger of §VII walks from it to the period model's PV one convention at a time, and the bridge under the pro forma pins the residual at $0. Directional, not tax advice.
CFADS proxy = the cost-recovering rent (the CRF sticker's capital charge) — a cost instrument's honest stand-in; a revenue object replaces it. Debt at the live 10-yr plus the bucket's observed DC-paper spread (the financing conditions below; Hyperion priced ~225 bps over, 24-yr amortizing, A+); data-center ABS prints 150–200 bps. Covenant 1.30× sits in the underwritten 1.20–1.50× band — realized DSCR in outstanding DC ABS runs a median ~1.93× (CRA, Dec 2025). Coverage is flat across projects that share a debt share and tenor, and improves as rates rise — structural, since the CFADS proxy collapses DSCR to CRF over the mortgage constant; a revenue object would differentiate projects and restore the lender's sign. Coverage by year is §VII's lender view (integrated CFADS — the book pays its own power bill); this bucket is NNN — lease EBITDA once a rent is set, the cost-recovering rent before — a different basis, and both print, never reconciled silently. The tranche here is d × K′, the screening figure §VI sizes and buckets on; the term loan the construction facility actually terms out (d·K plus the interest it capitalized) prints beside it in §VII with the difference. Directional.
The revenue line
rent on reserved kW · NNN “+E” · power passes through at costEverything above prices the site as a cost. Set a market rent and the same solve reads as a landlord's P&L — rent bills the reserved critical kW, every month, utilization notwithstanding; the tenant pays the engine's power stack at cost. The breakeven rent works even in cost-only mode: it is the all-in sticker, re-quoted in the unit the leasing market prices in.
CBRE H2-2025: ~$196/kW-mo average asking for hyperscale build-to-suit; the institutional band runs $120–220. Per-site rents live in the §I tune panel — a real portfolio doesn't lease uniformly.
Cost-only mode. Market asking (~$196) above the breakeven says the market currently pays a development margin on this portfolio; set the lever to read EBITDA, the covenant DSCR, and yield on cost.
Structure: triple-net “+E” (Orrick Jan 2026; datacenterHawk 2026) — rent covers shell, power distribution, cooling; metered power at cost to the tenant · rates quoted per kW critical IT (CBRE H2-2025 ~$196 asking; band $120–220) · benchmarks: YoC ~8% turnkey (RCLCO, DgtlInfra), stabilized cap ~5.5% ≈ 10-yr + 100–150 bps (Dominus 2026) · directional.
Severe = power price +20 pts on the curve · capacity at the zonal-separation print (1.35× — Dominion 25/26 vs the cap) · energization +1.5 yr · rates +100 bps (the 10-yr and the annualization rate), spread re-rated on the stressed coverage and never tighter than the base's. Both ends re-solved through the same equations — deterministic and traceable, not a percentile. Plan against the band, not the point.
| Yr | Capital | Operating (1−τ) | Depr. shield | Net cost | DF @ 7.4% | PV | |
|---|---|---|---|---|---|---|---|
| 0 | $11.00 B | — | — | $11.00 B | 1.000 | $11.00 B | |
| 1 | — | $602 M | −$183 M | $418 M | 0.931 | $389 M | |
| 2 | — | $602 M | −$183 M | $418 M | 0.866 | $363 M | |
| 3 | — | $602 M | −$183 M | $418 M | 0.806 | $337 M | |
| ⋮ 15 | — | $602 M | −$183 M | $418 M | 0.341 | $143 M | |
| 16 | — | $602 M | — | $602 M | 0.317 | $191 M | |
| ⋮ 20 | — | $602 M | — | $602 M | 0.238 | $143 M | |
| Σ PV | $15.53 B | foots ✓ | |||||
Level nominal cashflows make the annuity form exact — the closed form is this table, not a stand-in for it. One after-tax basis throughout: opex at (1−τ), capital net of the straight-line shield over 15 yr (MACRS would front-load); IDC carried in K′ at year 0; costs, not revenues. Zero opex escalation against a nominal WACC of 7.4% understates lifetime operating cost — read the NPV as a floor.
Reads the selected site's solved values — pick a different site in §I and the table re-derives. Shield runs to year 15, operating cost to year 20.
The next gigawatt of capital is cheapest in ERCOT North at $215/MWh after tax and quick to power (2.0 yr) — where the marginal dollar belongs.
| # | Market | After-tax $/MWh | To power | Cost of a year's slip |
|---|---|---|---|---|
| 1 | ERCOT NorthERCOT | 2.0 yr (1–3) | +$308 M/GW | |
| 2 | MISO South· committedMISO | 2.5 yr (2–4) | +$298 M/GW | |
| 3 | MISO WestMISO | 3.5 yr (2–4) | +$305 M/GW | |
| 4 | Southeast regulatedSouthern Co. | 3.5 yr (3–5) | +$318 M/GW | |
| 5 | CarolinasDuke Energy | 3.5 yr (3–5) | +$315 M/GW | |
| 6 | Pacific Northwest· committedNon-RTO / BPA | 5.0 yr (4–7) | +$331 M/GW | |
| 7 | Desert SouthwestNon-RTO / SRP-APS | 4.5 yr (3–6) | +$324 M/GW | |
| 8 | NYISO UpstateNYISO | 5.5 yr (4–7) | +$341 M/GW | |
| 9 | PJM West· committedPJM | 5.0 yr (4–7) | +$324 M/GW | |
| 10 | New EnglandISO-NE | 5.0 yr (4–6) | +$373 M/GW | |
| 11 | PJM DominionPJM | 5.0 yr (4–7) | +$363 M/GW | |
| 12 | CAISO Silicon ValleyCAISO | 6.5 yr (5–8) | +$438 M/GW |
Marginal cost of a 1 GW increment, after-tax (WACC + depreciation shield). Value of speed = the NPV a single year of energization slip adds, per GW — the price of a long queue. Directional; re-derives with §II.
The model prices the cost of speed exactly and refuses to price the benefit — a compute-year's value is the buyer's number, not the grid's. For scale: at the breakeven, $63 M buys a year of 1000 MW of accelerated capacity. Set a rent in §VI and this panel prices the market floor of that year. The tons come with it: the bridge emits ≈6.56 Mt across the window at on-site gas intensity (~529 kg/MWh, simple-cycle) — above every grid in the book.
An option study — nothing here reprices the portfolio. Bridge cost directional ($80–160/MWh spans rental turbines to fuel-cell contracts; the xAI Memphis and Bloom-era deals live in this band) · grid comparator = the site's delivered power cost incl. demand charges, assuming islanded operation through the window (no grid demand charges accrue until interconnection) · carry at the live cost of debt, half-drawdown convention.
The Pro Forma
| Calendar year | build 2026 | build 2027 | build 2028 | build 2029 | build 2030 | ramp 2031 | ramp 2032 | ramp 2033 | ramp 2034 | steady 2035 | steady 2036 | steady 2037 | steady 2038 | steady 2039 | steady 2040 | steady 2041 | steady 2042 | steady 2043 | steady 2044 | steady 2045 | steady 2046 | steady 2047 | steady 2048 | steady 2049 | steady 2050 | Σ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Operating | ||||||||||||||||||||||||||
| Purchased energy — contracted | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |
| — | — | (124.0) | (285.3) | (359.7) | (592.6) | (693.5) | (769.6) | (803.8) | (838.0) | (838.0) | (838.0) | (838.0) | (838.0) | (838.0) | (838.0) | (838.0) | (838.0) | (838.0) | (838.0) | (838.0) | (838.0) | (589.9) | (341.8) | (341.8) | (15,795.6) | |
| — | — | — | — | — | (41.1) | (53.4) | (65.7) | (74.0) | (82.2) | (82.2) | (82.2) | (82.2) | (82.2) | (82.2) | (82.2) | (82.2) | (82.2) | (82.2) | (82.2) | (82.2) | (82.2) | (82.2) | (82.2) | (82.2) | (1,548.9) | |
| — | — | (151.8) | (303.6) | (303.6) | (502.9) | (502.9) | (502.9) | (502.9) | (502.9) | (502.9) | (502.9) | (502.9) | (502.9) | (502.9) | (502.9) | (502.9) | (502.9) | (502.9) | (502.9) | (502.9) | (502.9) | (351.1) | (199.3) | (199.3) | (10,058.4) | |
| — | — | (0.6) | (1.4) | (1.8) | (2.9) | (3.3) | (3.7) | (3.8) | (4.0) | (4.0) | (4.0) | (4.0) | (4.0) | (4.0) | (4.0) | (4.0) | (4.0) | (4.0) | (4.0) | (4.0) | (4.0) | (2.7) | (1.5) | (1.5) | (75.4) | |
| EBITDA | — | — | (276.5) | (590.3) | (665.1) | (1,139.5) | (1,253.2) | (1,342.0) | (1,384.5) | (1,427.0) | (1,427.0) | (1,427.0) | (1,427.0) | (1,427.0) | (1,427.0) | (1,427.0) | (1,427.0) | (1,427.0) | (1,427.0) | (1,427.0) | (1,427.0) | (1,427.0) | (1,025.9) | (624.8) | (624.8) | (27,478.2) |
| Capital | ||||||||||||||||||||||||||
| Construction draw | (5,376.8) | (5,376.8) | (3,352.8) | (1,328.8) | (1,328.8) | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | (16,764.0) |
| Interest during construction | (104.3) | (342.2) | (347.5) | (215.1) | (287.4) | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | (1,296.5) |
| of which the screen carries: K′ − K | (580.0) | (580.0) | (405.0) | (229.9) | (229.9) | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | (2,024.8) |
| Financing | ||||||||||||||||||||||||||
| Construction facility drawn | 3,330.4 | 3,568.3 | 2,359.2 | 1,012.4 | 1,084.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 11,354.9 |
| Construction facility repaid at COD | — | — | (6,608.1) | — | — | (4,746.8) | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | (11,354.9) |
| Permanent facility drawn | — | — | 6,608.1 | — | — | 4,746.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 11,354.9 |
| — | — | (223.0) | (437.0) | (418.3) | (718.7) | (684.4) | (647.8) | (608.7) | (566.9) | (522.4) | (474.8) | (424.0) | (369.8) | (312.0) | (250.2) | (184.3) | (113.9) | (62.8) | (32.4) | — | — | — | — | — | (7,051.5) | |
| — | — | (134.0) | (277.1) | (295.8) | (508.4) | (542.7) | (579.3) | (618.4) | (660.2) | (704.7) | (752.3) | (803.1) | (857.3) | (915.1) | (976.9) | (1,042.8) | (756.2) | (450.2) | (480.5) | — | — | — | — | — | (11,354.9) | |
| Tax | ||||||||||||||||||||||||||
| — | — | 302.5 | 434.4 | 448.5 | 765.6 | 785.4 | 798.4 | 799.3 | 799.5 | 788.4 | 776.5 | 763.8 | 750.2 | 735.8 | 720.3 | 703.8 | 508.6 | 495.9 | 488.3 | 356.8 | 356.8 | 256.5 | 156.2 | 156.2 | 13,147.6 | |
| Net cash flow (equity) | (2,150.7) | (2,150.7) | (1,672.2) | (1,401.5) | (1,462.3) | (1,601.0) | (1,694.9) | (1,770.6) | (1,812.3) | (1,854.6) | (1,865.8) | (1,877.7) | (1,890.4) | (1,903.9) | (1,918.4) | (1,933.8) | (1,950.3) | (1,788.5) | (1,444.1) | (1,451.7) | (1,070.3) | (1,070.3) | (769.4) | (468.6) | (468.6) | (39,442.7) |
| Cumulative | (2,150.7) | (4,301.4) | (5,973.6) | (7,375.2) | (8,837.5) | (10,438.5) | (12,133.4) | (13,904.0) | (15,716.3) | (17,570.9) | (19,436.7) | (21,314.4) | (23,204.8) | (25,108.7) | (27,027.1) | (28,960.9) | (30,911.2) | (32,699.6) | (34,143.8) | (35,595.5) | (36,665.8) | (37,736.1) | (38,505.5) | (38,974.1) | (39,442.7) | — |
| Memo — non-cash | ||||||||||||||||||||||||||
| — | — | (710.4) | (710.4) | (710.4) | (1,204.0) | (1,204.0) | (1,204.0) | (1,204.0) | (1,204.0) | (1,204.0) | (1,204.0) | (1,204.0) | (1,204.0) | (1,204.0) | (1,204.0) | (1,204.0) | (493.6) | (493.6) | (493.6) | — | — | — | — | — | (18,060.5) | |
| — | — | (1,209.9) | (1,737.7) | (1,793.8) | (3,062.3) | (3,141.7) | (3,193.8) | (3,197.2) | (3,198.0) | (3,153.5) | (3,105.9) | (3,055.1) | (3,000.9) | (2,943.1) | (2,881.3) | (2,815.3) | (2,034.5) | (1,983.5) | (1,953.1) | (1,427.0) | (1,427.0) | (1,025.9) | (624.8) | (624.8) | (52,590.3) | |
| — | — | 302.5 | 434.4 | 448.5 | 765.6 | 785.4 | 798.4 | 799.3 | 799.5 | 788.4 | 776.5 | 763.8 | 750.2 | 735.8 | 720.3 | 703.8 | 508.6 | 495.9 | 488.3 | 356.8 | 356.8 | 256.5 | 156.2 | 156.2 | 13,147.6 | |
| Construction facility drawn, end | 3,330.4 | 6,898.7 | 2,649.8 | 3,662.1 | 4,746.8 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |
| Permanent facility balance, end | — | — | 6,474.1 | 6,197.0 | 5,901.2 | 10,139.6 | 9,596.9 | 9,017.6 | 8,399.2 | 7,739.1 | 7,034.3 | 6,282.1 | 5,479.0 | 4,621.7 | 3,706.6 | 2,729.7 | 1,686.9 | 930.7 | 480.5 | — | — | — | — | — | — | — |
| Coverage | lender ledger | |
|---|---|---|
| Minimum DSCR | — | x |
| Average DSCR | — | x |
| 1.46 | x | |
| CFADS basis | integrated |
| Returns | equity ledger | |
|---|---|---|
| Equity IRR | — | %/yr |
| MOIC | — | x |
| Payback year | — | year |
| 39,445.4 | $m | |
| 2.7 | $m |
| Funding | equity ledger | |
|---|---|---|
| 39,442.7 | $m | |
| Peak equity month | 2050-12 | |
| 8,777.3 | $m | |
| 11,354.9 | $m | |
| 18,060.5 | $m | |
| 18,060.5 | $m |
cost-only book: coverage is §VI's proxy · cost-only book: no owner return to measure — the equity funds a cost center
The Filings
| DUK · FY2025 basis | as filed FY2025 | restated · hedged 0% 2026 · Δ | restated · hedged 0% 2027 · Δ | restated · hedged 0% 2028 · Δ | |||
|---|---|---|---|---|---|---|---|
| Income statement · as filed, restated | |||||||
| 32,237.0 | — | — | — | ||||
| (9,041.0) | — | — | — | ||||
| (8,246.0) | — | — | — | ||||
| (6,324.0) | — | — | — | ||||
| 8,626.0 | — | — | — | ||||
| (3,634.0) | — | — | — | ||||
| 720.0 | — | — | — | ||||
| 5,712.0 | — | — | — | ||||
| (642.0) | — | — | — | ||||
| (102.0) | — | — | — | ||||
| 4,968.0 | — | — | — | ||||
| Memo — ratios and the ladder | |||||||
| 14,950.0 | — | — | — | ||||
| 6.32 | — | — | — | ||||
| 4.11× | — | — | — | ||||
| 5.36× | — | — | — | ||||
| Hedged share | — | — | — | — | |||
| Ratepayer — the utility desk's read | |||||||
| — | — | — | — | ||||
Composition & Sensitivity
All-in cost by site$ per IT MWh by component
What moves the blended costone driver at a time
The buildout, energizedannual spend as sites energize
The Derivation
Hover any quantity to light its full ancestry and see the equation that produced it. Plain cards are given inputs with sources; accent-edged cards are derived by the engine.
The proof
the engine's equation vs the public recordTrust in a model is earned one checkable number at a time. Here the engine's own capacity equation — C_cap = L × r_cap × 365, the line that prices every PJM site above — is run on the inputs PJM published, against the total PJM published. Both auctions cleared at the FERC cap across the whole footprint: one uniform price, so the reproduction is a single line anyone can do by hand.
“The cleared supply procured in the auction times the clearing price totals $16.4 billion.”
PJM news release, July 14, 2026 · stated by PJM's own release
PJM published the inputs (134,311 MW at $329.17); EVA and Enel North America state the $16.1B product.
PJM 2026/27 BRA report, July 22, 2025 · stated by EVA · Enel (from PJM's inputs)
The 2026/27 clear — $329.17/MW-day — is the very capacity rate this instrument carries for its PJM zones, so the reproduction exercises the registry's own given, not a lookalike. In the demand-charge unit: $325 × 365 ÷ 1000 ≈ $119/kW-yr — a PJM capacity obligation now costs ~1.8× an ERCOT 4CP transmission obligation per kW-yr, which is the arbitrage §V prices.
Model vs bill
the reconciliation — the loop the forecast tools never closeThe proof above checks the engine against the public record; this checks it against your invoice. Enter a month's actual bill and the variance decomposes into a volume effect (usage vs the model's assumption, at the model's rate) and a rate effect (the utility's rate vs the model's, on your actual usage) — the two foot to the total exactly. Reconciliations save locally, so the model builds a track record month by month.
Enter the month's metered MWh and invoice total to reconcile — billed demand sharpens the read but isn't required.
Compares the utility-billed power lines only (energy + capacity + wires + coincident-peak) — never your own capital or O&M · model month = annual solve ÷ 12 (seasonality is a known simplification; a hot-month bill reads high against it) · saved in your browser, synced nowhere.
Every solved value with its equation, every given with its source, the scenario, and a link that reproduces this exact case — as one hash-stamped file a workpaper or memo can carry. Reconstructable by hand. The CSV files open in a spreadsheet: the inputs as priced, and the assumptions with every source — each stamped with this case, run and model.
The Discovery
The Memo
- To
- Infrastructure Leadership
- From
- Portfolio Strategy — Data Center Energy
- Date
- —
- Re
- Portfolio energy outlook — 1.60 GW across 3 projects, 3 regions, 3 markets
The portfolio under review comprises 3 projects in 3 regions across 3 markets (Non-RTO / BPA, PJM, MISO), totaling 1.60 GW of IT capacity — 1.86 GW of grid load once cooling and distribution overhead are carried. At 80% utilization the portfolio purchases 13.0 TWh annually, for an all-in cost of $3.20 B per year against $16.76 B of construction capital. The blended all-in cost is $285.44 per IT megawatt-hour under base assumptions.
Cost structure is led by annualized capital at 55% of annual spend ($1.77 B). Financing terms and construction efficiency currently outweigh the power bill itself — the balance sheet is the lever.
Sensitivity analysis ranks build cost (±20%) as the dominant driver: a move of that size shifts blended cost by ±$40.60/MWh. Scenario work should concentrate there before refining second-order inputs.
Planning should run against the band, not the point: under the named severe case (power price +20 pts on the curve · capacity at the zonal-separation print (1.35× — Dominion 25/26 vs the cap) · energization +1.5 yr · rates +100 bps (the 10-yr and the annualization rate), spread re-rated on the stressed coverage and never tighter than the base's), blended cost moves from $285.44 to $326.30 per IT MWh and lifetime cost of ownership to $28.90 B — both ends solved through the same equations, deterministic and traceable.
— Concentration. 63% of IT capacity sits in MISO: that share answers to one market's rate cases, capacity auctions, and queue reforms. This is a concentration measure — the model prices no correlation between markets and no probability of loss — and diversification across market structures is the structural response to it, priced case by case in §II.
— Capacity market. 400 MW of capacity is exposed to PJM, where the auction has cleared at the FERC cap three years running. The model carries capacity explicitly at the 2026/27 clear ($55 M/yr portfolio-wide); the stress case is zonal separation — the Dominion zone printed $444/MW-day in 2025/26, ~35% above the cap-bound RTO price.
— Unpriced carbon. The portfolio emits 5.01 Mt CO₂ annually at a carbon price of zero. At $50/t this is a latent $250 M/yr exposure — material to siting if policy or voluntary commitments harden.
— Time to power. Full portfolio energization completes in 2031. Every year of wait is carried in the figures above as construction-period interest on the capital drawn (§VII's facility; the screen's K′ carry) without the compute the capital was raised for — the model's largest timing exposure.
On current assumptions the portfolio's cheapest project is MISO South at $262.58 per IT MWh and its dearest is PJM West at $337.76 — a spread of $75.18/MWh on this scenario's assumptions. The spread is a screening figure: which project should take incremental megawatts depends on the objective — the §I scorecard weighs cost with speed to power, grid headroom, carbon and price basis — and on the scenarios compared in §II, not on the cost ranking alone. Full derivation of every figure above is available in the instrument, §X.
- Case
- Current case
- Case hash
- ba70228c
- Run ID
- 08b8d2c45b799220
- Model version
- m-35d7daaaab75e1da
- Data vintage
- Figures approximate, vintages stamped per source: wholesale markets built up from EIA STEO hub strips + PJM BRA / capacity clears + NITS formula rates (2026) · bundled-tariff markets on EIA industrial retail (2024) with docket-pinned large-load terms · EPA eGRID emission rates · utility interconnection processes for time-to-power. Directional by design.
- Basis
- 10-yr 4.70% (calibration figure) · baked vintage prices · curve tilt 1.03%
- Generated
- A directional strategy screen, not utility-grade dispatch: every figure is approximate and vintage-stamped; the model's equations, sources and conventions are the audit pack's.
- One docketed load ramp (AEP Ohio Schedule DCT, 50/65/80/90 % on contract anniversaries) is applied to every project.
- Fixed-price cover is the only contract instrument the ledger prices; a vPPA settles in the clean book (§IV), not as a ledger row.
- The power-price shock moves every market by the same fraction — there is no per-market shock.
- Two coverage bases print and neither is adjusted to the other: §VI's screening DSCR (cost-side proxy on d × K′; lease EBITDA when a rent is set) and §VII's integrated lender DSCR (CFADS over the year's debt service).
- Calculated monthly, presented annually, under the period conventions of docs/period-model.md; working-capital lags are 0/0 unless set, and the mid-quarter MACRS convention is disclosed, not applied.