Electron·Economics
Primary research · electroneconomics.substack.com ↗
Electron Economics · DC Residual Value Model

Asset Input

Define the data centre asset and its financing structure. The model uses the EE Forecast 2040 demand scenarios as the secondary market backdrop at renewal, then applies asset-specific adjustments to produce a probability-weighted residual value estimate.
⚠ Renewal year capped at 2035 — scenario data extends to 2035. Actual renewal year: —. Model outputs for terms beyond 2035 should be treated as indicative only.
Asset Characteristics
Determines LLT friction, secondary market liquidity, and grid constraint exposure
300 MW
10 yr
Financing Structure
Typical range: $8–14M/MW for AI-grade DC (2025 build cost)
$10,000/kW
60%
25%
IG underwriter minimum — typically 1.20–1.50×
1.30×
Derived — renewal window
Renewal year (model)
—
Total asset cost
—
Debt at origination
—
Debt at renewal
—
EE Forecast Backdrop
Secondary market depth at renewal year — from EE Forecast 2040 model
Constrained
—
Base
—
Accelerated
—
US installed AI DC capacity at renewal year across EE scenarios. Thicker market = stronger residual value support. Bear gap / bull gap from base shown in Residual Value tab.
Cash Flow & Return Assumptions
Inputs to the discounted cash flow in the view. Rent is quoted gross; the NOI margin converts it to net operating income, so the going-in yield is visible rather than assumed. Every field here is an EE analytical default — override before relying on any output.
Turnkey AI-grade capacity, power passed through. Typical quoted band $110–180/kW-month.
$135
Net of opex, property tax, insurance and non-recovered power. Implied going-in yield shown below.
43%
2.5%
0.8%
Incurred once at renewal. Charged to the seller at exit when the hold ends at renewal.
6%
Years held past renewal before exit. Set to 0 to sell at the renewal date.
5 yr
Rent achieved on re-letting relative to the expiring contract rate.
90%
6.50%
The single most powerful input in this model. Everything below turns on it.
9.00%
13.0%
Year-1 NOI
—
Going-in yield
—
Exit year
—
Entry price
—
Model basis: EE Forecast 2040 (v1.9, Aug 2026) provides the secondary market demand backdrop at renewal — base 408.5 GW by 2040 at 15.0% CAGR, FERC-anchored 50 GW baseline, read live from the shared EE_SYNC block. ISO friction sourced from EE Tariff Tracker Aug 2026; LLT severity reflects Oregon Power Act (Jun 2026) and FERC Docket RM26-4-000 (Jun 18, 2026). Replacement cost is denominated per kW of critical IT load, matching the EE Capex Stack basis. Asset-specific adjustments (ISO friction, power contract quality, tenant creditworthiness) are applied on top. This is an analytical framework — not a rating, guarantee, or legal opinion. All inputs are illustrative defaults; override all fields before relying on outputs.  ·  Forecast model ↗  ·  Tariff tracker ↗  · 
Electron Economics · DC Residual Value Model

Residual Value Analysis

Probability-weighted residual value at renewal, conditioned on EE demand scenarios. The residual value is expressed as % of replacement cost and $/kW, across base, accelerated, and constrained scenarios.
Base (40%)
Accelerated (30%) — bull
Constrained (30%) — bear
RV — Base case
—
% of replacement cost
RV — Accelerated
—
% of replacement cost
RV — Constrained
—
% of replacement cost
Prob-weighted RV
—
40% base · 30% accel · 30% cons
Re-let period
—
Months dark (base scenario)
Residual Value by Scenario ($/kW)
Value Drivers — Base Scenario
Market depth from Forecast 2040 ↗ · ISO friction from Tariff Tracker ↗
Scenario Sensitivity — RV as % of Replacement Cost
How residual value changes across the EE demand scenario range. The width of this band is the primary risk variable for IG underwriting. Scenario depth is driven entirely by installed capacity GW at the renewal year, read live from EE Forecast 2040 ↗; the ISO multiplier applied on top is the LLT friction index maintained in the EE Tariff Tracker ↗.
Asset Quality Score
Key Risk Factors
RV methodology: Base residual value = replacement cost × market depth factor × power contract multiplier × ISO multiplier × tenant factor × renewal probability. Market depth factor calibrated to EE Forecast 2040 (v1.9, Aug 2026) ↗ scenario GW at renewal year against a 2025 secondary market baseline (depth = 0.70 + GW/408.5 GW × coefficient; 0.70 = minimum RV fraction recoverable in pre-ramp market; scenario totals base 408.5 GW / accelerated 494.1 GW / constrained 303.3 GW by 2040). Scenario GW are read live from EE_SYNC, not restated here. ISO multipliers from EE Tariff Tracker ↗ (Aug 2026) — the LLT friction index that drives the ISO multiplier is maintained there, updated for Oregon Power Act (WECC), AEP Ohio DCT (MISO SE), and FERC RM26-4-000 context. Power and tenant adjustments are EE analytical estimates — not filed data. Cross-reference all-in capex at EE Capex Stack v20 ↗ ($32.75/W AI-grade all-in, per critical IT watt — the same basis as the $/kW IT-load slider). Full per-parameter provenance in the Methodology view. All multipliers inferred
Electron Economics · DC Residual Value Model

Returns — Discounted Cash Flow

The investment view. Contracted lease cash flows, maintenance and refurbishment capex, a probability-weighted renewal event, re-lease economics, and a terminal value drawn from the residual value analysis — discounted to today. This is where the residual stops being a coverage statistic and starts being a share of the return.
Unlevered IRR
—
Project return on total asset cost
Levered IRR
—
Equity return after debt service
Equity MOIC
—
Total equity distributions ÷ equity in
PV of residual
—
Terminal value discounted at project WACC
Residual share of value
—
% of gross present value that is terminal, not contracted rent
Unlevered Cash Flow by Year
Operating cash flow net of maintenance and refurbishment capex, with the terminal value stacked at exit. Dark-period carry in the post-renewal years is probability-weighted, not switched on or off.
NPV vs Discount Rate
Where each curve crosses zero is that IRR. The gap between the two crossings is the whole of the leverage contribution — and it widens with LTC, not with asset quality.
Return Summary
Entry Price Test
The price at which this asset clears the target unlevered return, set against where the transaction market actually clears. Bankable band read live from EE Deals Tracker ↗.
Cash Flow Detail
Full annual build. Phase 1 is the contracted term; Phase 2 is the post-renewal hold, where NOI is the probability-weighted blend of a renewing tenant and a dark-then-re-let outcome.
DCF methodology and its one structural trap: renewal probability is applied once. The residual value figures in the Residual Value view are already net of the renewal-risk haircut (— at the modelled renewal year, from EE_SYNC.forecast.renewalRisk). When the hold extends past renewal, that haircut is divided back out of the terminal value and the renewal event is modelled explicitly in the cash flows instead — dark months, carry cost and a marked-to-market re-lease. When the hold ends at renewal, the terminal value keeps the embedded haircut and no post-renewal cash flows are modelled. Either way the tenant-retention risk is charged exactly once. Escalators, NOI margin, mark-to-market, refurbishment and all three discount rates are EE analytical defaults with no filed analogue; the discount rate in particular is an assumption the model previously avoided by not having one. Terminal value scales linearly with replacement cost, so the entry price test is solved on cost with operating cash flow held fixed. All return inputs inferred
Electron Economics · DC Residual Value Model

Underwriting Output

The credit committee view. Coverage ratios, stressed residual value, break-even RV realisation, and the IG underwriting verdict — can this asset support residual value risk without hyperscaler/Oracle intermediation?
Coverage ratio (base)
—
RV / outstanding debt
Coverage ratio (stressed)
—
P10 demand scenario
Break-even RV realisation
—
% of modelled base RV the secondary market must realise for the lender to be made whole — not a physical occupancy rate
Dark period risk
—
Months untenanted (base)
IG capacity needed
—
Residual value guarantee amt
Coverage Ratio Waterfall
All values as % of replacement cost. "Target coverage" = coverage ratio target × debt-at-renewal ÷ replacement cost.
Underwriting Summary Table
Residual Value vs Debt Outstanding — Scenario Range
Green zone = RV > debt (IG supportable without intermediation). Amber = marginal. Red = RV guarantee required.
Systemic vs Idiosyncratic Risk Split
Underwriting note: This model produces analytical estimates for discussion purposes only. It is not a rating, commitment, or guarantee. Coverage ratios below target trigger the IG residual value product requirement — the model quantifies how much guarantee capacity is needed, not the terms on which it would be provided. Renewal risk haircut sourced from EE Forecast 2040 (v1.9, Aug 2026) via EE_SYNC.forecast.renewalRisk: renewal_risk_2030: 0.92; 2029: 1.00 — near-term renewals modelled with minimal non-retention risk. Transaction $/MW context: 47 of 62 tracked transactions (US$219.9B) were Bankable at announcement; Bankable cluster clears $11–12.5M/MW (EE Scarce MW Intelligence v5.8). Cross-reference: Deals Tracker ↗ Analytical estimate
Electron Economics · DC Residual Value Model

Model Methodology

How the residual value and underwriting outputs are computed. Every adjustment factor is named, sourced, and independently challengeable.
Step 1 — Secondary Market Depth (EE Forecast 2040)
The EE Forecast 2040 model (v1.9, Aug 2026) produces US installed AI DC capacity by scenario at the renewal year, anchored to FERC's confirmed 50 GW 2025 baseline (24% CAGR 2020–2025). Base 408.5 GW by 2040 at 15.0% CAGR. Accelerated 494.1 GW at 16.5% CAGR. Constrained 303.3 GW at 12.8% CAGR. Renewal window scenarios: base 139.2 GW · accelerated 169.1 GW · constrained 102.2 GW at 2031. Scenario probabilities: 40% base, 30% accelerated, 30% constrained. These figures are read live from the shared EE_SYNC block (EE_SYNC.forecast.gw), not restated in this file — a superseded constrained run was previously hard-coded here and had drifted up to 5% high. The base case embeds Jevons rebound (efficiency drives more adoption, not less — calibrated to RAND RRA3572-1), OEM-capped BTM gas (~70 GW cumulative by 2040), ~24 GW nuclear (T1 contracted + T2 probability-weighted), and an LLT friction suppression of ~14% at 2040. Cross-reference: EE Forecast 2040 ↗
Market depth factor: Parameterised as depth = 0.70 + (GW / 2040 base total) × coefficient, with no clamp. The 0.70 intercept represents the minimum recoverable RV fraction in a thin pre-ramp secondary market (approximately 2025 conditions). The GW term scales linearly with installed capacity relative to the 2040 base total. Removed Aug 2026: the formula previously carried min(1.05), min(1.10) and max(0.55) guards. None of them bound anywhere in the 2029–2035 domain — computed ranges are depthBase 0.782–0.920, depthAccel 0.810–0.989, depthCons 0.757–0.835. The floor would require constrained GW below ~68 against a domain minimum of 74.6 GW; the caps would require base GW above 408.5 against a domain maximum of 257.1 GW. They were presented as live constraints and were not. Coefficients differ by scenario (base 0.35, accel 0.38, cons 0.30) to reflect asymmetric demand elasticity — constrained scenarios exhibit thinner depth per installed GW than accelerated. At 2031 base (139.2 GW): depth ≈ 0.82. Accelerated scenario (+22% vs base) supports stronger secondary demand and shorter re-let. Constrained scenario (−23% vs base) implies thinner secondary market and longer dark period. Efficiency breakthrough scenario excluded — at 103.4 GW in 2031 it implies demand well below installed base, representing a structural technology shift rather than a cyclical softening.
Step 2 — Asset-Specific Adjustments
ISO / LLT friction multiplier: Georgia Power PLL territory (1.08×) — near-zero friction, 11 GW contracted, PSC base rate frozen through 2028, 9,985 MW new gen certified Dec 2025. ERCOT (1.04×) — low friction, merchant exposed, BYONG pathway operational. MISO Southeast (0.96×) — AEP Ohio DCT approved Jul 2025 (85% take-or-pay, 12-yr term) signals moderate friction rising. PJM (0.92×) — 4+ year average queue, high friction; BYONG partial escape valve reduces effective friction for large loads bringing own generation. Dominion Virginia (0.85×) — $1.5M/MW collateral, 14-year minimum-take, highest friction, queue processing declining. WECC (0.84×) — Oregon Power Act Jun 2026: PGE filed +29% DC rate increase, LLT severity revised Tier 2→Tier 3; relet_adj increased to +8 months. Source: EE Tariff Tracker Aug 2026, llt_friction_per10pp=0.018 (EE_SYNC.tariff.lltFrictionPer10pp). FERC Jun 18, 2026: Six tailored orders issued requiring all regional grid operators to justify or reform large-load tariff structures within 60 days (Docket RM26-4-000). If utilities reform under this mandate, PJM/Dominion/WECC friction may loosen materially for 2027–2030 — this is upside to the constrained RV scenario.
Power contract multiplier: Prime grid with 20-yr LLT (1.0×, baseline). Nuclear PPA direct (1.10×) — clean, predictable, IG counterparty; nuclear PPAs command 3× the $/MW of grid-dependent deals (EE Scarce MW Intelligence v5.8). BTM RICE (0.88×) — on-site gas, peak-economics clearing, ESG exposure; 14.9 GW firm + 5.6 GW framework = 20.5 GW across 21 projects (EE Modular Gas Tracker Aug 2026). Corrected Aug 2026: this line previously read "11.9 GW firm + 3.8 GW framework" — the 3.8 GW was the tracker's INNIO/VoltaGrid single-OEM contracted total read as an all-market framework figure. BTM heavy GT (0.82×) — supply-constrained replacement; GE Vernova backlog 116 GW under contract incl. service agreements (Q2 2026; ≥125 GW year-end target), heavy-frame lead times 24–60 months. The prior "100 GW / 40+ months" was the Q1 2026 backlog (44 firm + 56 SRA) and a lead time the tracker does not hold; the lead-time band is now the heavy-frame class from EE_SYNC.gas.leadTimeMonths. BTM grey market (0.72×) — CF6-80C2 pool depleting, no replacement pathway. Source: EE BTM analysis, $0.06–0.08/kWh prime vs $0.25–0.30/kWh BTM gas. OEM capex context: powered shell $4.50/W of $32.75/W AI-grade all-in cost (EE Capex Stack v20, Aug 2026). Basis: $32.75/W is stated per critical IT watt, not per facility watt. Cross-reference: EE Capex Tracker ↗ · BTM OEM Tracker ↗
Tenant factor: Hyperscaler (1.0×). Oracle/Blue Owl IG-wrapped (0.98×). GPU cloud — non-IG but contracted (0.88×). AI lab non-IG standalone (0.78×). Colocation/multi-tenant (0.92×). The tenant factor captures the correlation between tenant credit quality and residual value — a non-renewing non-IG tenant signals broader compute economics stress, not just idiosyncratic default. Source: EE Who Wears the Risk Apr 2026, Crusoe-Oracle-OpenAI capital stack analysis. Deals context: Core Scientific shareholders rejected CoreWeave's $9B bid (Oct 2025) — sets floor at >$7M/MW for energised US HPC capacity; hyperscaler-backed assets clear $11–12.5M/MW (EE Scarce MW Intelligence v5.8 Jun 2026). Cross-reference: EE Deals Tracker ↗
Renewal probability: EE Forecast 2040 renewal_risk parameter at renewal year. At 2029: 1.00 (near-term renewal; minimal non-retention risk modelled). At 2030–2033: 0.92 (8% non-renewal). At 2034–2035: 0.88 (12% non-renewal). Non-renewal is not a credit default — it is a product strategy decision. The 8% haircut reflects Meta/OpenAI lease optionality observed in current deal structures.
Step 3 — Residual Value Calculation
Base RV ($/kW) = replacement_cost × market_depth_factor × power_multiplier × iso_multiplier × tenant_factor × renewal_probability
Re-let period (months dark): base 12 mo · constrained 24 mo · accelerated 6 mo. Represents the expected time to find a secondary tenant at market rate. Adjusted for ISO territory — PJM/Dominion add 6–8 months, Georgia/ERCOT subtract 3 months.
Probability-weighted RV = 0.40 × base + 0.30 × accelerated + 0.30 × constrained. The 40/30/30 split reflects EE's view that the base scenario is most likely but the distribution is not symmetric — the constrained scenario is as probable as the accelerated given capital discipline risk.
Break-even RV realisation = debt at renewal ÷ total base RV. This is the fraction of modelled base residual value the secondary market must realise for the lender to be made whole — not a physical occupancy rate. It was labelled "break-even occupancy" prior to Aug 2026; the label was wrong and the calculation was not. A break-even of 75% means the lender is covered if secondary market realisations are at least 75% of modelled base RV.
Step 4 — Underwriting Outputs
Coverage ratio = prob-weighted RV / debt outstanding at renewal. Debt outstanding = origination debt × (1 − amortisation %). Target: 1.20–1.50× per IG underwriter convention.
Stressed coverage = constrained-scenario RV / debt. This is the number the IG risk-taker is writing against — coverage in the scenario where secondary demand is thinnest.
IG capacity needed = max(0, target_coverage × debt − prob_weighted_RV). This is the guarantee amount — the shortfall between the RV the asset supports and the coverage the credit committee requires. This is the size of the IG residual value product.
Coverage waterfall chart: All bars expressed as % of replacement cost. "Target coverage" bar = target ratio × (debt at renewal / replacement cost) — it is the coverage-adjusted debt target, which may exceed 100% of replacement cost for assets with high LTC and low amortisation at aggressive coverage targets.
Systemic vs idiosyncratic split: Renewal non-retention is modelled as 60% systemic (correlated with AI demand scenario) and 40% idiosyncratic (asset/tenant specific). The systemic component cannot be diversified away across a portfolio — it requires scenario-conditional underwriting, not actuarial pooling. This is the core argument for IG risk-taking capacity rather than pooled insurance.
Renewal year cap: Scenario data is available through 2035. For lease vintages and terms producing a renewal beyond 2035, the model uses 2035 scenario values and displays a warning. This is a model boundary, not a view on marketability.
Step 5 — Structured Provenance
Every numeric constant in this model, with the evidence it rests on. Confidence is stated per source: Filed the document is on file with a regulator or is a published EE dataset with its own provenance · Filing an active proceeding, the number may still move · Inferred an EE analytical estimate, not an observed value. The multiplier values are EE analytical estimates throughout — the entry-level pill says so rather than leaving the concession in the Limitations prose.
Limitations: This model does not produce a legal guarantee, credit rating, or investment recommendation. All multipliers are EE analytical estimates. The renewal risk parameter is calibrated to observed 2025–2026 deal structures — it may not reflect future market conventions. The systemic/idiosyncratic split is an assumption, not an empirical estimate. Scenario GW figures read live from EE Forecast 2040 (v1.9, Aug 2026) via EE_SYNC; ISO multipliers from EE Tariff Tracker (Aug 2026). Every multiplier carries a structured sources array — see Step 5 below and the Changelog. LLT friction landscape is actively evolving: FERC Docket RM26-4-000 (Jun 2026) may reduce friction in constrained ISOs faster than base case assumes. Community opposition (36+ projects delayed, $162B in investment — Data Center Watch Jun 2026) and Oregon Power Act are emerging friction factors not yet fully parameterised. This tool is a structured framework for practitioner discussion.  ·  electroneconomics.substack.com  ·  Analytical framework · Not a rating
Electron Economics · DC Residual Value Model

Changelog — What Changed and Why

Every model update, data revision and source addition. No version removes data — superseded figures are stated alongside their replacement so a reader can reproduce the delta.
New First time this data appears
Update Existing figure replaced with newer source
Fix Error corrected
Structure UI / tab / layout change
v1.6 Sep 12 2026 Update
A second mark-to-market print, from a different asset layer, and it lands the same way as the first: this model’s 0.90x re-letting assumption is now contradicted twice.
  • Oracle disclosed on 10 September that all GPU capacity coming up for renewal was renewed or resold at a 20 per cent premium to prior contracts, with most of those accelerators at least four years old. That is a compute-layer mark-to-market of 1.20x on the asset class supposed to decay fastest.
  • It sits alongside the real-estate print this model already carried: Digital Realty renewed at 1.254x cash portfolio-wide. Two independent prints, two different layers, both above 1.20x, against a modelled 0.90x. That assumption is the most consequential unevidenced input in the model and it is now the one with the most evidence against it.
  • Read the boundary carefully. Oracle’s figure is a rental-rate premium on renewal, not a hardware resale price. It says the revenue-generating capacity of a four-year-old accelerator re-contracted above its prior rate. It says nothing directly about what the chip would sell for, and it is not a residual value.
  • Useful-life assumptions: no company changed one between 3 and 12 September — a verified null, searched specifically. Nebius’s extension from four years to five, effective 2026, remains the most recent and stands unrevised.
  • CoreWeave discloses no useful-life assumption at all, in any filing found. Given its leverage and the volume of commentary on its depreciation, the absence is itself the finding.
The secondary market is structurally incapable of producing the comp everyone wants
Compute Exchange opened a secondary marketplace for used H100s and A100s in July and is the obvious venue for residual-value evidence. It runs on request-matching, not auction, and publishes no clearing prices by design. Anyone waiting for auction comps on used accelerators is waiting for something this market is not built to produce. Silicon Data launched monthly GPU residual-value benchmarks on 24 August; its coverage, methodology and values could not be obtained and that remains the largest open item in this model.
SiliconANGLE, 17 Jul 2026; Silicon Data launch, 24 Aug 2026
v1.5 Sep 07 2026 UpdateFix
Four assumptions are contradicted by the only 2026 prints available. None has been changed, and all four now carry the contradiction on the face of the model.
  • Re-lease mark-to-market is modelled at 0.90× prior rent. Digital Realty’s Q2 2026 renewals repriced at +25.4 per cent cash and +32.0 per cent GAAP — 1.254×. That is a 39 per cent gap in going-out rent and it compounds through the residual.
  • The rent escalator is modelled at 2.5 per cent. The one disclosed 2026 escalator on stabilised hyperscale Northern Virginia stock is 3.6 per cent, on 15-year leases to three investment-grade tenants at a blended AA−.
  • Going-in rent is modelled at US$135 per critical IT kW-month. Digital Realty signed new leases above 1 MW at US$157/kW in Q2 2026. US$135 is a 2024 wholesale number.
  • Refurbishment capex is modelled at 6 per cent of cost. A liquid-cooling retrofit runs about US$2m/MW, which against this model’s own US$10,000/kW replacement cost is 20 per cent. Retrofit only becomes the relevant intervention above about 40 kW per rack and AI roadmaps run past 200, so some stock is not retrofittable at that price at all.
  • Nothing was re-fitted, and the reason matters. Every one of those prints is portfolio-level in a 1.4 per cent vacancy market. This model values end-of-life, air-cooled stock at expiry, and no published re-letting outcome for that asset exists from anyone at any date. The assumptions may still be right for what they describe; they are no longer right as descriptions of the market.
  • Six further assumptions are now labelled as having no published evidence at all: re-let downtime, NOI margin, maintenance capex, dark-carry cost, and both discount rates. They are judgement, and the model says so rather than implying a market basis.
  • Replacement cost carries a disclosed dispute: US$10,000/kW here against US$17,600/kW in Cushman & Wakefield’s September 2026 cost guide, a 76 per cent spread that cannot be reconciled because neither source states its denominator.
The only clean cap rate in the period
Digital Realty’s buyout of Blackstone’s interest in three Northern Virginia facilities: 288 MW critical IT, US$7.8bn gross, initial stabilised cap rate above 6.5 per cent, 100 per cent leased, 15-year terms, blended AA−, 3.6 per cent escalators.
Digital Realty press release, 29 Jun 2026
What the model values has never been observed
There is no published re-letting outcome for legacy air-cooled stock at lease expiry. Every re-leasing figure available is a portfolio renewal spread in the tightest market on record. That absence is the single largest uncertainty in this model and it is not reducible by searching harder.
Searched 1 Aug to 7 Sep 2026; nothing found
v1.4Aug 23 2026Update
The Texas pause is recorded against ERCOT, and the multiplier is deliberately left where it is.
  • Texas stopped data centers advancing through ERCOT's interconnection process on August 3 2026, pending a statewide audit of a queue totalling roughly 474 GW, about 90 percent of it attributed to data centers.
  • The ERCOT multiplier stays at 1.04. The pause cuts both ways on residual value and the two effects cannot be separated with the evidence available: it raises development risk, which argues the number down, and it freezes new supply while demand persists, which makes already-energized Texas capacity scarcer and argues the number up.
  • A lease renewal in 2033 to 2035 sits well past a nine-month audit either way. Moving the multiplier would assert a direction the evidence does not support, so the argument is written into the model instead, with three conditions that would reopen it: the audit producing permanent admission criteria, the April 2027 deadline moving materially, or Texas transaction premiums diverging from comparable markets.
  • ERCOT's label changes from low friction to low on terms, paused on access.
Update
ISO_ADJ.ercot multiplier HELD at 1.04, and the reason is now in the file. Label moves from Low (merchant) to Low on terms, paused on access. Two sources added: the Aug 3 2026 directive (conf filed) and an explicit calibration note recording that the pause raises development risk and simultaneously makes already-energised Texas capacity scarcer, that the two are not separable on current evidence, and that a 2033–2035 renewal sits past a nine-month audit regardless. Deliberate non-change, documented as one rather than left silent.
EE RV Model calibration · Abbott letter Aug 3 2026
v1.3 Aug 20 2026 New Fix Structure
The model now discounts, so it answers what an asset is worth today rather than only whether it covers its debt.
  • A new Returns view runs the full investment case: purchase, contracted rent, maintenance and refurbishment spend, the renewal decision, re-letting, and sale value, all discounted to today. It reports unlevered and levered return, equity multiple and the present value of the residual.
  • On the default asset, 30 percent of the value is a bet on what the building is worth at renewal rather than on the lease itself. That is the number worth arguing about.
  • A new entry price test solves what the asset can be bought for and still clear a 9 percent unlevered return. The answer is US$7.28 million per MW, against a market that has been clearing at US$11 to 12.5 million per MW.
  • Tenant renewal risk is charged once, not twice. Depending on whether the asset is sold at renewal or held past it, the risk sits either in the sale value or in the cash flows, and the model states which.
  • The discount rate is now the most powerful input in the model, and it is an assumption. The previous version avoided that by having no discount rate, which hid the same assumption at zero.
New
Discounted cash flow engine and a Returns view. Until this version the model answered a coverage question — what fraction of today's asset value survives at renewal, and does it clear the debt. That is not the investment question. The new engine runs a full annual build: acquisition, contracted lease cash flow with an escalator, maintenance capex, a probability-weighted renewal event with dark months and standby carry, a marked-to-market re-lease, re-tenanting refurbishment, and a terminal value drawn from the residual value analysis. Outputs are unlevered IRR, levered IRR, equity MOIC, PV of residual, unlevered and equity NPV, year-1 DSCR, and the share of gross present value that is terminal rather than contracted rent. On the shipped defaults that last figure is 30.1% — the reason a discount rate could not be avoided any longer.
EE RV Model v1.3 · all return inputs are EE analytical estimates
Fix
Renewal risk would have been charged twice, and now is charged once. computeRV() already multiplies the retention haircut into rvPct. A DCF that also models the renewal event in its cash flows and then uses that same rvPct as terminal value double-counts tenant-retention risk and understates every return in the view. The engine branches explicitly: when the hold runs past renewal the haircut is divided back out of terminal value and the renewal event is modelled in the cash flows; when the hold ends at renewal the haircut stays in terminal value and no post-renewal cash flows exist. Which branch ran is reported in the Return Summary as renewalRiskChargedIn and asserted in the verification suite.
EE RV Model v1.3 · verification: 94 assertions across 6 parameter cases
Fix
Entry-price solver was pinning terminal value to the slider instead of the price being tested. Terminal value is a percentage of replacement cost, so it must move with the cost the bisection is trying. The first build read inp.rc inside the cash flow constructor rather than the tested cost, which meant the solver reported a root of an equation it was not solving. Caught by an independent linearity assertion, not by inspection. NPV-neutral entry price on the shipped defaults moved $7.86M/MW → $7.28M/MW, and unlevered IRR is unaffected.
EE RV Model v1.3 · spec/verify_dcf.js
New
Entry price test, set against the transaction market. The model now solves the replacement cost at which unlevered NPV is zero at the target return, and prints it beside the Bankable clearing band read live from EE_SYNC.deals.bankableClearingPerMwM. On the shipped defaults the asset supports $7.28M/MW at a 9.00% unlevered return against a Bankable cluster of $11.0–12.5M/MW. The gap is the part of the clearing price that contracted cash flow and residual value do not explain at that discount rate — it is being paid for scarcity, for the interconnect, or at a lower required return than the one set.
EE Deals Tracker v5.9 · via EE_SYNC.deals
Structure
EE_SYNC v1.2 → v1.3; rv.dcf added and owned here. Eleven return-model defaults — going-in rent, NOI margin, escalator, maintenance and refurbishment capex, dark carry, mark-to-market, cost of debt, unlevered and equity discount rates, post-renewal hold — are published in the shared block rather than living in this file's slider attributes. The sliders are assigned from the block on load; the HTML value= attributes are a no-JS fallback only. Any other model quoting a data-centre return now reads the same discount rate rather than inventing one. Propagated byte-identical to all seven models.
EE_SYNC v1.3 · md5 80b54d03576a9c61fe91c384671050c0
Structure
Tab order now runs Asset Input → Residual Value → Returns → Underwriting Output → Methodology → Changelog. Returns sits between the residual analysis that feeds its terminal value and the credit view that consumes its coverage. The Asset Input page gains a Cash Flow & Return Assumptions card and four derived readouts — year-1 NOI, going-in yield, exit year, and entry price per MW — so the yield implied by the rent and margin sliders is visible at the point of entry rather than inferred three tabs later.
EE RV Model v1.3
New
What this version still does not do. Residual value is credited only at exit — mid-life resale is not modelled, and modelling it would move the hold-period answer. Rent is a single going-in rate rather than a lease-by-lease rent roll. The discount rate is now the most powerful input in the model, and it is an assumption; the previous version avoided that discomfort by having no discount rate at all, which was worse. Nothing in the return stack has a filed analogue.
EE RV Model v1.3 · stated limitations
v1.1 Aug 18 2026 Fix Update New Structure
Renewal-year capacity and the behind-the-meter pipeline now read from the published models, and a misleading KPI label was corrected.
Fix
Constrained scenario was tracking a superseded Forecast 2040 run. The hard-coded EE_SCENARIOS.cons row sat 3–5% above the live model at every renewal year — 2029 78 → 74.6, 2030 91 → 87.2, 2031 107 → 102.2, 2032 126 → 119.7, 2033 147 → 140.5, 2034 164 → 156.9, 2035 184 → 175.2 GW. Base and accelerated rows already matched to rounding (2035 base 257 → 257.1; accelerated 310 → 310.4). The literal object is deleted: all three scenarios now read live from EE_SYNC.forecast.gw so they cannot drift again. Stressed coverage and constrained RV move; base and probability-weighted figures move only at the third decimal.
EE Forecast 2040 v1.9 · via EE_SYNC.forecast.gw
Fix
2040 scenario totals and CAGRs corrected. Base 408 → 408.5 GW, accelerated 493 → 494.1 GW, constrained 318 → 303.3 GW; CAGR constrained 13.1% → 12.8% (base 15.0% and accelerated 16.5% unchanged). The 318 GW figure came from the Forecast 2040 changelog, which was itself stale relative to its own live model. The 408 denominator in the market-depth formula now points at EE_SYNC.forecast.gw.base[2040], and the missing-year fallbacks (139 / 169 / 107) now read from EE_SYNC rather than being restated.
EE Forecast 2040 v1.9 · Aug 2026
Fix
Gas Tracker citation was a single-OEM figure read as a market total. The BTM RICE multiplier note cited 11.9 GW firm + 3.8 GW framework (Jun 2026); the 3.8 GW was the tracker's INNIO/VoltaGrid single-OEM contracted total, not an all-market framework number. Replaced with the tracker's own project-array computation for Aug 2026: 14.9 GW firm + 5.6 GW framework = 20.5 GW across 21 projects (firm = Operational + Ordered + Permitted; framework = Framework + Framework signed + Testing), read from EE_SYNC.gas.pipelineGW.
EE Modular Gas Tracker · Aug 2026 project array
Update
GE Vernova backlog was a quarter stale; lead time was not a tracked value. 100 → 116 GW under contract — 100 GW was the Q1 2026 figure (44 firm + 56 SRA); Q2 2026 is 116 GW against a ≥125 GW year-end target, read from EE_SYNC.gas.oemBacklogGW.geVernova. Lead times "40+ months" → 24–60 months, heavy-frame class from EE_SYNC.gas.leadTimeMonths.heavyFrame; the tracker holds no "40+ months" value.
GE Vernova Q2 2026 earnings Jul 22 · EE Modular Gas Tracker
Update
Capex citation re-versioned and its basis stated. EE Capex Tracker v17 (Jun 2026) → EE Capex Stack v20 (Aug 2026); the all-in figure is read from EE_SYNC.capex.allInPerW and the "$32.75M/MW blended, 2025" phrasing is dropped. Basis now stated explicitly: $32.75/W is per critical IT watt, not per facility watt. This matters because the replacement-cost slider is denominated $/kW of IT load — the two are consistent, but that consistency was previously accidental and undocumented. Powered shell stated as $4.50/W from EE_SYNC.capex.layersPerW rather than the "~$4–5/W" range.
EE Capex Stack v20 · via EE_SYNC.capex
Fix
KPI was mislabelled in a way that would mislead an underwriter. "Break-even occupancy" → "Break-even RV realisation" (default 60.4%). The code comment above the calculation already said it was not a physical occupancy rate; the tile did not. The definition — the % of modelled base RV the secondary market must realise for the lender to be made whole — is now in the sub-label, not only in the methodology prose.
EE RV Model · label correction
Fix
Dead clamps removed from the depth formulas. Math.min(1.05, …), Math.min(1.10, …) and Math.max(0.55, …) never bound anywhere in the 2029–2035 domain — computed ranges are depthBase 0.782–0.920, depthAccel 0.810–0.989, depthCons 0.757–0.835. The floor would require constrained GW below ~68 against a domain minimum of 74.6 GW; the caps would require base GW above 408.5 against a domain maximum of 257.1 GW. A reader reasonably assumed they were active constraints. Numerically a no-op; the methodology now records why.
EE RV Model · formula audit
New
Structured provenance added — the largest gap in this model. Every entry in ISO_ADJ, POWER_ADJ and TENANT_ADJ, plus the depth coefficients, renewal retention, re-let periods and the systemic/idiosyncratic split, now carries a sources array of {t, u, d, conf} records drawn from citations already in the methodology (Dominion GS-5, Georgia Power PLL, AEP Ohio DCT, Oregon POWER Act / PGE filing, PJM queue, FERC RM26-4-000, EE Tariff Tracker, EE Modular Gas Tracker, EE Capex Stack, EE Deals Tracker, Data Center Watch). No source was invented. Where the file already conceded a value is an EE analytical estimate, the record is conf:'inferred' and that concession is now a visible confidence pill rather than a line in the Limitations prose. Rendered as Step 5 in the Methodology view.
EE RV Model · provenance pass Aug 18 2026
Update
Renewal risk and scenario weights now read from EE_SYNC. RENEWAL_RISK → EE_SYNC.forecast.renewalRisk and SCENARIO_PROBS → EE_SYNC.forecast.scenarioProbs (keys remapped accelerated/constrained → accel/cons). Values verified identical before the swap — 1.00 / 0.92×4 / 0.88×2 and 40/30/30 — so no output moves. RELET_BASE likewise reads EE_SYNC.rv.reletBaseMonths (12 / 6 / 24 unchanged).
EE_SYNC v1.0 · Aug 18 2026
Fix
Nav date showed today's date over Jun 2026 constants. A prior revision assigned new Date().toLocaleDateString() to the nav date under a comment reading "use live date instead of hardcoded string" — which inverted the intent: the nav date is the data vintage, not the current month. The leftover assignment is removed so it can no longer override EE_SYNC.vintage. Model basis updated EE Forecast 2040 v3, Jun 2026 → v1.9, Aug 2026.
EE_SYNC.vintage · Aug 2026
Structure
Changelog tab added; Residual Value view cross-linked. This tab. The Residual Value view now links inline to EE Forecast 2040 where scenario depth is explained and to the EE Tariff Tracker where the ISO friction multiplier is explained, so a reader can walk upstream from any number to the model that owns it.
EE house pattern · Aug 18 2026
v1.0 Jun 2026 New
First release: residual value at lease renewal, conditioned on demand scenario, power contract, tenant credit and grid friction.
New
Initial release. Asset Input, Residual Value and Underwriting Output views. Residual value as replacement cost × market depth × ISO multiplier × power contract multiplier × tenant factor × renewal retention, with coverage ratios, re-let periods, IG guarantee sizing and the systemic/idiosyncratic split.
EE RV Model v1.0
Electron Economics · Track record

Accuracy Ledger

Every figure this platform has published and later changed, across all seven models, with the size and direction of each change and what caused it. Research products ask you to trust their numbers. This is the evidence for how often those numbers have moved and by how much.
Revisions by model
Split between our own errors and figures that moved because new information arrived. A model with more revisions is not a worse model; it is usually a more heavily used one.
Our errorNew information
What actually goes wrong
Cause of our own errors only. New-information updates are excluded, because they are not mistakes.
Every revision
Rows marked with a green edge belong to the model you are looking at. Sorted by size of change.