The governing question is not what the tariff rate is. It is what the delivered cost of power is for a 100 MW, 500 MW, or 1 GW facility - and what contractual risk the utility imposes to deliver it. This page answers both. Markets are ranked. LLT severity is scored. Upcoming regulatory decisions are flagged. The map shows rates. This page shows decisions.
Breaking Jun 2026
Oregon Power Act: PGE files +29% data center rate hike
Portland General Electric has filed a 29% rate increase for data centers under Oregon's new Power Act, which creates a legally distinct DC customer class and requires the industry to cover the cost of new generation and transmission built to serve it. Residential customers get a 1.3% cut - confirming regulators' finding that DCs had been cross-subsidized by households. Hillsboro city council votes on a development moratorium Jun 9. PacifiCorp faces the same requirement by 2028, with OPUC initial ruling expected this summer. PGE Oregon LLT severity updated from 14 → 48. Market score revised from Tier 2 → Tier 3.
Source: The Oregonian, Jun 2026 · OPUC filing pending
▲ Top 5 Cheapest Markets
All-in at 85% LF · best available schedule
▼ Most Restrictive LLTs
Scored on take-or-pay, term, collateral, cost allocation
⚡ Regulatory Calendar
Upcoming decisions · latest: Oregon Power Act Jun 2026
Market Ranking · DC Site Selection Score
Composite score 0-100 across: all-in power cost (40%), LLT severity (25%), regulatory stability (20%), grid buildout (15%). Higher = better for data center operators. Click any row to open Tariff Tracker.
Methodology: Rate score inverted from ¢/kWh (lower rate = higher score). LLT severity from Risk Ledger protection scores. Regulatory stability from tension scores and recent rulings. Grid buildout from IRP filings and interconnection queue data. Scores are analytical estimates - verify independently before site selection decisions.
Estimated total annual power cost ($/yr) at 85% load factor for three facility sizes. Does not include capital costs, interconnection, or LLT collateral.
Facility size
Annual cost = all-in ¢/kWh × MW × 8,760 hrs × 0.85 LF. Does not include LLT minimum billing shortfall, collateral, or site-specific infrastructure charges. Indicative only - verify with utility.
Cost Stack Decomposition · How the Rate Is Built
For practitioners who evaluate utilities, a single ¢/kWh figure is insufficient. This shows what drives it: generation, transmission, distribution, demand charges amortized at load factor, and riders. Select a market to drill in.
Generation = base energy charge. Transmission = separate T charge or OATT adder where applicable. Distribution = D component or estimated split. Demand amortized = $/kW-mo × 12 ÷ (LF × 8,760) × 100. Riders = fuel cost recovery, environmental, and other mandatory adders. Sources: filed tariff PDFs, FERC Form 1, IRP filings.
Regulatory Pipeline · Upcoming Decisions
Forward-looking calendar of utility proceedings, rate cases, and LLT rulings expected through 2027. Each creates a binary risk event for DC developers in that territory.
Dates are best estimates from filed dockets, utility communications, and PUC scheduling orders. Regulatory proceedings routinely slip. Treat as signal direction, not a hard calendar. Track PUCO, NCUC, CPUC, and relevant PUC docket systems for current status.
Power markets intelligence · US Data Center Power Cost Map
Every data center deal is a power delivery deal.
The governing question: what is the all-in effective cost of grid power for a data center operator in each US utility territory - and what contractual restrictions does the utility impose to deliver it? This map answers the first question. The Tariff Tracker answers the second. The Risk Ledger answers both simultaneously.
Recent regulatory developments · Jun 2026: AEP Ohio DCT approved Jul 2025 (85% take-or-pay, 12-yr term) · Georgia PSC base rate freeze through 2028, 9,985 MW new gen certified Dec 2025 · Xcel Energy filed LLT Apr 2026: 100% new gen+tx cost allocation, ≥50 MW (PUC pending) · Duke Carolinas rate case Jan 2026 formalizing large-load tariff terms · PPL Electric $275M rate case settlement Mar 2026 includes DC tariff provisions · Wisconsin VLC tariff approved Apr 24 2026: ≥100 MW (lowered from 500 MW proposed), 100% cost allocation, 15-yr; Oracle challenge dropped Aug 17 2026.
Load factor85%
Color by
All-in rate ¢/kWh
2.5¢5¢7.5¢11¢+
Green = cheapest. Red = most expensive. Dark = not tracked.
Large-load tariff
Active / approved
Pending review
Proposed / filed
None enacted
Market type
Regulated IOU
ERCOT deregulated
PJM T&D only
Data confidence
⬤ Filed tariff PDF
◎ Regulatory filing
○ FERC Form 1 / inferred
Cheapest market
-
Most expensive
-
Rate gap
-
cheap → expensive
LLTs active / pending
12 / 6
of 60 markets tracked
All-in rate: base energy + demand ÷ (LF × 730 hrs) + riders. ERCOT: TDSP delivery (PUCT confirmed) + ERCOT 2024 RT avg $26/MWh + REP margin. Sources: filed utility tariff PDFs, PUCT, IUC annual report, ERCOT IMM. Cross-link to Deals Tracker for power risk by transaction.
Power markets intelligence · Tariff Tracker
US Data Center Tariff Tracker
60 utility service territories · Rate schedules · LLT status · Effective $/kWh model · Click utility to navigate · Click schedule row to expand.
Utilities
60
top DC markets
Fully modeled
2
VA · GA filed tariff
Filed tariff data
8
VA·GA·IA·OR·NV·WI·TX×2
Rate range
2.6-11¢
all-in best schedule
LLTs active
15
approved or in force
Cheapest market
IA/VA
MidAmerican · Dominion
Ratepayer protection & DC expectations
Risk Ledger
Two-sided analysis of every major US DC power market. Left: mechanisms utilities use to protect ratepayers. Right: what data centers actually get in return. The gap between the two columns is the market signal.
⚠ Utility protection mechanisms
Instruments utilities use to prevent stranded cost exposure when large-load customers reduce or exit. Scored by stringency - how much financial and operational risk is transferred to the data center operator. Higher = more protection for ratepayers, more risk for operators.
◎ Data center expectations
What operators typically receive in exchange: rate certainty, interconnection speed, renewable content, and operational flexibility. Scored 1-5. Higher = more favorable to the data center operator. Tension score = mismatch between protection stringency and DC-friendliness.
Sort by
Market / utility
⚠ Ratepayer protection mechanisms
◎ DC expectations delivered
Tension
Min billing %
Term (yr)
Collateral
Cost alloc
LF floor
DR req
Rate certainty
Interconnect
Renewable
Flexibility
Protection stringency: ○ none · ● low · ●● moderate · ●●● high · ●●●● severe. DC scores 1-5 (higher = better for operator). Tension = protection stringency minus DC score average. Sources: filed utility tariffs, PUCO/SCC/PSC orders, FERC dockets.
Electron Economics · Primary Research Analytics
Power Market Analytics
The structural finding: the US power market for data centers has bifurcated into two regimes - low-rate, low-protection markets (Iowa, Virginia, ERCOT) where utilities want the load, and high-protection, constrained markets (AEP Ohio, Wisconsin, California) where utilities are managing the load they already have. The LLT adoption wave is not uniform: it tracks exactly the utilities that reached capacity constraints first.
Cumulative count of large-load tariffs filed, approved, or proposed. Inflection tracks capacity constraint saturation in key ISO territories.
Sources: utility rate case filings, PUC orders, press releases. Pending and proposed counted from filing date, not approval. 2026 count reflects filings through Jun 2026.
Protection vs. DC-Friendliness Quadrant
Each bubble = one utility. X = DC-friendliness score (1-5). Y = protection stringency (0-4). Bubble area ∝ all-in rate. Color = tension level. Ideal for DC operators: bottom-right.
Scores from Risk Ledger primary research. Protection: average of 6 mechanisms (min billing, term, collateral, cost alloc, LF floor, DR). DC score: average of rate certainty, interconnect, renewable, flexibility.
Rate Component Decomposition · Selected Markets
Stacked bars showing base energy, demand amortized at 85% LF, and riders. Demand component is the primary driver of market differentiation.
Markets selected for analytical contrast: cheapest regulated, ERCOT, highest protection, and highest rate. Full dataset in Tariff Tracker.
Rate Sensitivity to Load Factor · Key Markets
How much does all-in rate change across 50-98% load factor? High demand-charge utilities (Georgia, Wisconsin) are most sensitive. Iowa and Virginia least sensitive.
Sensitivity = (all-in at 50% LF) − (all-in at 98% LF). Higher sensitivity = demand charge dominates. Lower sensitivity = energy charge dominates. Demand-heavy utilities penalize low-utilization operators most.
Market Regime Matrix
60 utility markets classified on two axes: rate competitiveness vs. LLT stringency. Quadrant placement indicates operator risk profile.
Rate tier: Low (<5¢), Mid (5-7.5¢), High (>7.5¢). LLT stringency: None / Low (no LLT or minimal), Moderate (pending/proposed), High (active, ≥85% min billing or full cost alloc). Quadrant = operator siting signal.
Energy vs Demand Decomposition · All 33 Markets
Stacked bars showing what fraction of each market's all-in rate comes from base energy (dark green), demand charges amortized at 85% LF (mid green), and riders (amber). The white % label shows demand share when it exceeds 50% — these markets punish low utilization hardest.
All 33 US markets. Sorted low → high all-in rate at 85% LF. Georgia Power and Oregon PGE are outliers: energy charge is trivially small but demand + riders dominate. Markets where demand > 50% of total are most sensitive to load factor variation.
Load Factor Penalty · Rate Swing 98% → 50% LF
How much does your all-in rate increase if load factor drops from 98% to 50%? Demand-heavy markets (Georgia, Wisconsin, Oregon) are severely punished. Iowa and Virginia are demand-light — rate barely moves.
Swing = all-in at 50% LF minus all-in at 98% LF. Markets with high demand charges ($/kW-mo) have large swings because demand amortization is inversely proportional to LF. A 500 MW DC underperforming at 50% LF pays an extra $40–90M/yr in demand-heavy markets.
LLT Severity Fingerprint · Component Breakdown
Each bar shows which of 6 protection mechanisms drives a utility's LLT severity score. Some markets are restrictive on one dimension (Wisconsin: demand charge), others on multiple (AEP Ohio: take-or-pay + term + collateral).
6 components: min billing % (dark red), contract term (amber), collateral requirement (orange), generation/tx cost allocation (dark), LF floor (brown), demand response (light green). Top 16 markets by severity score. Total = LLT Severity Index 0–100.
Every US market. X = all-in rate ¢/kWh. Y = LLT severity 0–100. Bubble size = market score (larger = better). Color = tension level. Bottom-left is ideal; top-right should be avoided.
Bubble size proportional to DC site selection score (0–100). Tension: green = low, amber = moderate, red = high, dark red = extreme. All tracked US markets. Iowa and Grant PUD anchor the bottom-left ideal zone. Wisconsin and AEP Ohio occupy the top-right danger zone.
Market Treemap · All 33 Markets by Score
Area = site selection score (larger = better). Color = tension level. Rate label shown inside each cell. A quick visual of the entire US DC power market landscape.
Cell area proportional to site selection composite score. Color: green = low tension (DC-friendly), amber = moderate, red = high, dark red = extreme. Rate shown in ¢/kWh at 85% LF. Grant PUD and Iowa dominate by area — highest scores, lowest tension.
Rate vs Protection Stringency · All 33 Markets
X = all-in rate ¢/kWh. Y = average protection stringency (0–4 from Risk Ledger). Color = LLT status. Bubble size = site selection score. Bottom-left quadrant = ideal (cheap + low protection).
Protection stringency = average of 6 Risk Ledger protection mechanism scores (0–4 each). All 33 US markets plotted. Labeled markets: notable anchors of each quadrant. LLT color: green = active, amber = pending, teal = proposed, grey = none.
Axes normalized to 0–1. Rate score: inverted ¢/kWh (higher = cheaper). LLT: inverted severity (higher = less punitive). Regulatory: 0–20 component of site selection score. Grid: 0–15. DC-friendliness: Risk Ledger DC expectations score. Iowa polygon anchors all five axes near maximum.
Power Cost Trajectory · 2023–2030
Projected all-in ¢/kWh for key markets based on filed rate cases, approved increases, and announced trajectories. Dashed = projected. Solid = confirmed.
Projections: Grant PUD +9.5%/yr (commission policy through 2036). Georgia Power base frozen 2028, FCR floats. AEP Ohio escalator from DCT. PGE Oregon Power Act +29% base + annual reassessment. Iowa MidAmerican flat (EAC adjusts). Sources: filed rate cases, commission orders, utility IRPs.
Cross-Subsidy Signal · Rate Shift at DC Tariff Adoption
When a utility adopts a DC-specific tariff, residential rates move in the opposite direction. The spread is the cross-subsidy that existed before the tariff. Wider bar = larger prior subsidy.
Operator Risk Exposure Matrix · Annual Power Cost by Market × Facility Size
Each cell = estimated annual power cost ($M/yr) at 85% load factor. Color intensity = cost magnitude. Use as a first-pass site selection filter — red cells are markets where power cost dominates the P&L.
Annual cost = all-in ¢/kWh × MW × 8,760 hrs × 0.85 LF. Does not include LLT minimum billing shortfall, collateral, or site-specific infrastructure. LLT collateral can add $30–150M upfront for 500 MW facility depending on market. All-in rates at 85% LF per main tracker data.
Open Rate Case & LLT Docket Gantt
Active regulatory proceedings affecting DC power cost. Each bar = filing date → expected ruling. Color = risk direction for DC operators.
Dates from filed dockets and commission scheduling orders. Regulatory proceedings routinely slip 3–6 months. "Expected" column = analyst estimate, not commission commitment. Track individual PUC docket systems for live updates.
Market Momentum · Regulatory Direction Score
Each market scored -5 (rapidly tightening) to +5 (actively DC-friendly) based on the last 18 months of regulatory activity. Arrow = direction of travel.
Momentum score = weighted sum of: new LLT filings (-), LLT approvals with punitive terms (-), rate freezes (+), economic development riders (+), no LLT despite DC growth (+), moratorium discussions (-). 18-month window Jan 2025 – Jun 2026.
Regulatory Signal Timeline · 2024-2026
Key utility regulatory actions affecting data center power cost and access. Color = direction of change for DC operators.
Sources: utility press releases, PUC orders, state legislative filings. Green = DC-favorable outcome. Red = DC-restrictive. Amber = neutral/pending. All filings primary research - no third-party data subscriptions.
Electron Economics · International Markets
International DC Power Markets
5 Tier 1 markets: Ireland, Netherlands, Sweden, UK, Singapore. All-in effective rates converted to US¢/kWh at current FX. Market structure, LLT status, and regulatory risk assessed on the same framework as US utilities.
The structural finding: European markets are 2–4× more expensive than the best US markets on absolute power cost. Sweden (SE1 north) is the only international market competitive with Virginia or Iowa. Singapore is expensive but constrained by island geography — there is no alternative. Ireland and the Netherlands are price-restricted and physically constrained. The UK is gas-linked and politically uncertain. For greenfield hyperscale economics, the US remains dominant.
FX rates used: EUR/USD 1.09 · GBP/USD 1.27 · SGD/USD 0.74 · SEK/USD 0.096 · Updated Jun 2026. All rates converted to US¢/kWh for comparability with US tracker.
International vs US Markets · All-in Rate Comparison
Same methodology as US tracker: all-in effective ¢/kWh at 85% load factor. International rates converted from local currency at Jun 2026 FX.
US rates from main tracker. International: best available large industrial or DC-specific rate where distinct. Ireland/NL/UK: wholesale + network + levies for large industrial customer. Sweden SE1: hydro-advantaged northern zone. Singapore: EMA regulated non-household tariff Q2 2026. Rates are indicative — exact large-load contracts are negotiated and confidential. FX: EUR/USD 1.09, GBP/USD 1.27, SGD/USD 0.74, SEK/USD 0.096.
International Regulatory Pipeline
Active proceedings and policy changes affecting DC power costs and access in Tier 1 international markets.
Electron Economics · Long-range Fundamentals Model · Mirror
US Data Center Capacity Forecast to 2040
Mirror of EE Forecast 2040 v1.9 — the model runs there, not here. This view is read-only. Every capacity, CAGR, PUE and supply-split figure below is read at render time from the canonical EE_SYNC.forecast block, which is the executed output of eeforecast2040.html. This file previously carried its own stale v3.1 copy of the model whose constrained case printed 338 GW against the canonical 303 GW; that copy has been deleted rather than re-ported. Change a forecast number in the canonical model, not here.
WHAT THIS VIEW IS
A read-only mirror of the canonical series. computeForecast2040() in this file is a shim over EE_SYNC.forecast.gw and EE_SYNC.forecast.supply — it does not re-run the model, and it has no weights of its own.
WHAT WAS REMOVED
A stale local copy of the model (v3.1: FM40_WEIGHTS plus nine helper functions) with a reversed LLT-friction exponent that the canonical model has since fixed. Divergence at 2040, stale → canonical:
Base 411.8 → 408.5 GW · Accelerated 497.3 → 494.1 GW · Constrained 337.6 → 303.3 GW · High Efficiency 125.7 → 134.7 GW
The constrained case was the material error — an 11% overstatement of the downside.
BASIS
Facility MW (post-PUE). Anchor: FERC 2025 State of Markets = 50.0 GW confirmed at end-2025. Pre-2025 history on the charts below is a back-cast (24% CAGR 2023–25, 18% 2018–23, 7% before) carried over unchanged from the retired local model so the chart domain stays 2020–2040 — it is illustrative history, not canonical output.
OPEN FORMULA, CLOSED WEIGHTS
The formula disclosure, weight calibration, backtest and parameter sensitivity all live in the canonical model. They are not restated here, because restating them is exactly how this file drifted.
US 2025 (FERC)
50 GW
actual baseline
Base 2030
—
canonical series
Base 2035
—
canonical series
Base 2040
—
canonical series
2040 range
—
constrained → accel.
CAGR 2025–40
—
emergent, not assumed
PUE 2040
—
facility ÷ IT watt
BTM 2040
—
behind-the-meter share
Scenario
US DC Installed Capacity · 2020–2040 · Canonical Series (mirror)
Four scenarios as published by EE Forecast 2040 v1.9. Shaded band = Accelerated to Constrained. Facility MW, post-PUE, anchored to FERC 50.0 GW at 2025. Pre-2025 is back-cast history, not model output.
2040 outcomes read from EE_SYNC.forecast.gw: Base 408 GW (CAGR 15.0%) · Accelerated 494 GW (16.5%) · Constrained 303 GW (12.8%) · High Efficiency 135 GW (6.8%). The High Efficiency case is the only one where hardware efficiency gains outrun compute growth and facility demand plateaus before 2040 — it is a genuine tail, not the central case. The retired local v3.1 copy printed 412 / 497 / 338 / 126 GW respectively; the constrained figure was the material divergence.
Supply Decomposition · Front-of-meter vs Nuclear vs BTM
Canonical base-case split from EE_SYNC.forecast.supply, where ftm + nuc + btm equals the base total exactly. Nuclear is a classification of grid-connected supply, not an additive pathway, so it is drawn as an inner band rather than stacked.
The supply decomposition is published for the base case only. When a non-base scenario is selected, the bands are scaled pro-rata by that scenario’s total (results[y] ÷ base[y]) — the split itself is not re-solved, and the chart footer says so. BTM reaches 58 GW by 2040 in the base case, roughly 14% of total installed capacity; against the Gas Tracker’s firm pipeline of 14.9 GW (20.5 GW including framework agreements across 21 projects) that implies the BTM pathway is mostly still unbuilt. OEM backlogs — GE Vernova 116 GW, Siemens Energy 87 GW — are the binding constraint on how fast it fills.
Canonical Scenario Table · EE_SYNC
Every cell is read from EE_SYNC at render time. Scenario probabilities are the canonical model’s, not this file’s. The active scenario is highlighted.
This table replaces two charts that the retired local model used to draw: a driver-trajectory series and a parameter sensitivity tornado. Both required mutating the local weight vector and re-running the model, which is no longer possible here and should not be — parameter sensitivity is an attribute of the model, and the model runs in EE Forecast 2040. Probabilities sum to 1.00 across base/accelerated/constrained; High Efficiency is an unweighted tail case.
Historical Anchors · Canonical Series vs Published Actuals 2014–2028
Canonical series plotted against independent external anchors. Grey band = LBNL 2024 plausible range. Orange dots = FERC/LBNL confirmed actuals. Pre-2025 model line is the back-cast described above.
Anchors: LBNL 2024 US DC Energy Usage Report (TWh→GW via utilization factors); FERC 2025 State of Markets (50 GW end-2025, 24% CAGR 2020–2025). The 2014–2018 residual reflects behind-the-meter enterprise load that LBNL includes and grid-connection data does not. The canonical series prints 65.0 GW at 2027 against Goldman Sachs’ ~76 GW operational estimate; the gap is definitional — Goldman includes BTM enterprise, and the canonical model tracks BTM separately (1.3 GW at 2027 in the base case). Calibration and backtest scoring are performed in the canonical model, not here.
Electron Economics · Developer Intelligence
LLT Collateral Requirements
Upfront capital commitments, financial assurance requirements, and minimum billing structures across all tracked utilities. The collateral requirement is often larger than the rate differential — it belongs on the balance sheet, not in the operating model.
The $750M question: Dominion Virginia GS-5 requires $1.5M/MW upfront — $750M for a 500 MW facility at contract execution. Virginia also enacted a $0.011/kWh DC consumption tax effective Jul 1 2026 (~$41M/yr at 500 MW). ERCOT is not zero — SB6 financial security and site control obligations apply to ≥75 MW loads. Iowa and the PNW PUDs require zero. Not every requirement scales with MW: Xcel Colorado's Schedule TL is a flat $600K upfront regardless of facility size, and is shown as such.
Capital Impact at Selected MW · Upfront Collateral Requirement
Hard-dollar requirements only. Most are $/MW and scale with facility size; Xcel Colorado's is a flat amount and is labelled "flat, not MW-scaled". Negotiated structures (Ameren, Evergy, AEP Ohio, Consumers MI) use 2-year minimum bill projections as collateral — typically $50–200M+ for 250 MW facilities.
Sources: VA SCC final order Nov 25 2025 (GS-5 $1.5M/MW); Virginia budget act (DC tax $0.011/kWh Jul 1 2026); CPUC Advice 2018-E Apr 2026 (Xcel CO — $600K flat, being a $120K study deposit plus $600K total upfront before service begins, not $600K per MW). Treating the Xcel figure as $/MW printed $150M at the 250 MW default against a true $600K — a 250× overstatement at 250 MW and 1,000× at 1 GW — and because it set the maximum, it also compressed the bar scale for every other utility in this chart. Negotiated: Ameren/Evergy/Consumers = 2yr projected min bills.
Collateral Matrix · Tracked Utilities
Full collateral, minimum billing, term, and exit fee structures. Capital impact at selected facility size.
Utility / Tariff
LLT
Collateral Type
$/MW
Capital Impact
Min Billing
Term
Notes
Oregon PGE: no upfront cash — 100% distribution upgrade costs + 1¢/kWh surcharge for >100 MW loads + 30-yr term for ≥220 MW. · Wisconsin We Energies: VLC approved Apr 24 2026 at a ≥100 MW threshold (lowered from the 500 MW proposed Mar 2025), 100% cost coverage, 15-yr. Oracle dropped its challenge Aug 17 2026, so the financial security requirement now stands unchallenged — the amount is still TBD in the revised tariff. Oracle had faced a >$7B letter of credit (~$100M/yr); an A-/A3 rating avoids posting. · ERCOT: SB6 applies to ≥75 MW — real capital obligation. PUCT centralized evaluation approved Jul 2026. FERC Jun 18 show-cause orders excluded ERCOT.
Electron Economics · DC Power Intelligence Tracker · Last updated Aug 18 2026
Changelog — What Changed and Why
Every data revision, corrected number and structural change to this tracker. No version silently removes data — superseded figures are named alongside their replacement.
This pass was a cross-model reconciliation, not a redesign. Where this file disagreed with the model that owns a number, the owning model won and this file became a reader. Three of the entries below are arithmetic errors that were visible on screen: a 250× collateral overstatement, an 11% overstatement of the constrained 2040 downside, and a nav date that was being written twice with two different answers into an element that had already been destroyed.
New First time this appears
Update Figure replaced with a newer source
Fix Error corrected
Structure Code / layout change
v2.3Sep 12 2026NewUpdate
A federal layer, a NERC registration regime that turns a data centre into a registered entity, and the contract terms that actually determine what a large-load tariff costs.
The six FERC section 206 show-cause proceedings are added — PJM, SPP, NYISO, MISO, CAISO and ISO-NE — and the status correction matters: they are IN ABEYANCE. Responses were due 17 August and are now due 16 November (20 November for SPP). As of 12 September no RTO has filed. Coverage through August routinely implies otherwise.
NERC Project 2026-02 creates Computational Load Owner and Computational Load Operator as registered entity types, with three new standards (CLO-001/002/003). Draft registration thresholds are 1 MW of IT load AND 20 MW aggregate at a single point of interconnection, at 60 kV or above. A data centre is about to acquire reliability obligations, which is a category change rather than a tariff term.
Contract terms are consolidated by utility for the first time: minimum take, minimum term, ramp, exit fees, collateral and load-factor screens across nine to eleven utilities each. The median minimum contract term moved from 5 years pre-2025 to 12 years post-2025 — the single largest structural change in these tariffs, and one that directly lengthens the term a lender can underwrite.
Delaware signed four bills on 26 August, effective immediately. HB 445 requires large energy-use facilities to produce or secure 100 per cent of their own power, reported as the first such state legislative mandate in the US. HB 233 creates a separate rate class paying full infrastructure cost and prioritises disconnection during emergencies.
TVA approved a new large-customer rate class on 21 August, about 10 per cent phased over three years, effective 1 October. TVA is not PUC-regulated so there is no docket, and no MW threshold, term or minimum take has been published.
CAISO proposes a 50 MW large-load definition with two flexible services (FILI and FLIP); comments closed 2 September, board vote 28 October, FERC filing 16 November.
Still no US$/MW CIAC anywhere, and Virginia is still the only mandate
This tracker has said since August that no utility publishes a US$/MW contribution-in-aid-of-construction rate. That remains true after a full sweep of the window. Virginia’s SCC order of 31 July requiring mandatory CIAC for Dominion direct-connect loads is still the only mandate in the country, its compliance filing is due about 29 October, and no rate has been set. The EEI tracker, which has the broadest docket coverage at around 55 entries, carries no CIAC, collateral or exit-fee columns at all — which is why the consolidated table above had to be assembled from three separate sources.
Virginia SCC, 31 Jul 2026; EEI Large Load Projects and Tariffs, Aug 2026
Filed, approved and effective are three different things
No new large-load tariff was filed at any state PUC in this window, and none was approved. Every item above is either an enactment (Delaware), a board action outside PUC jurisdiction (TVA), a proposal at comment stage (CAISO), or a federal proceeding whose deadline moved (FERC). Trade press routinely conflates the three, and the distinction determines whether a term is something a borrower can be held to today.
EE sweep of state PUC dockets, 26 Aug to 12 Sep 2026
v2.2Sep 07 2026NewUpdateFix
The US$1.5m per MW everyone quotes as CIAC is collateral. They are different obligations and only one of them exists as a published rate.
Dominion’s US$1.5m/MW is a refundable security deposit, reducible up to 70 per cent on credit. The Virginia SCC’s 31 July 2026 order separately requires a mandatory contribution in aid of construction for direct-connect large loads covering substations and connecting lines — not refundable, and carrying no published rate. The compliance filing is due around 29 October 2026.
No utility in this tracker publishes a US$/MW CIAC figure. None. Searched every jurisdiction, every window. Most set CIAC cost-of-service and per-study rather than by tariffed rate, so a clean table would have to be derived from interconnection agreements and study results.
Georgia Power’s 3,200 MW OpenAI contract was approved 26–27 August 2026 in docket 71526 — the largest single large-load contract approved anywhere to date. OpenAI pays the full cost of infrastructure needed to serve it and has committed up to 1,000 MW of flexible demand. The executed contract is fully redacted: no term, minimum bill, take-or-pay, collateral or exit fee has been published, and a commissioner has confirmed the terms remain trade secrets post-execution.
ERCOT Batch Zero moved four times in the window. PUCT approved three good-cause exceptions on 20 August; ERCOT delayed issuance on 31 August; provisional conditional classifications were issued on 3 September. ERCOT published no project counts, MW or GW in any of it — the 200 GW and 474 GW figures in circulation are press reporting and are not ERCOT’s.
Microsoft and the Wisconsin Industrial Energy Group both protested ATC and We Energies large-load commitment agreements at FERC on 21 August, the first significant hyperscaler-versus-utility fight over large-load cost commitment structures.
Duke Energy Florida’s large-load tariff faced its first contested hearing under SB 484 on 25 August. A sitting commissioner called the petition facially noncompliant. No vote was taken.
No new state adopted a large-load tariff in the window.
Collateral and CIAC are being conflated in the market
A refundable deposit sized per MW and a non-refundable cost contribution set per study are different obligations with different balance-sheet treatment. The tracker now separates them and states that no CIAC rate exists to put in the second column.
Virginia SCC final order 25 Nov 2025 (collateral) · Virginia SCC order 31 Jul 2026 (CIAC mandate)
The Georgia contract is the largest approved and the least disclosed
Three point two gigawatts, full cost responsibility, and a fully redacted contract. Everything the tracker would want to record about its terms is a trade secret, so nothing about them is inferred here.
Georgia Power press release, 26 Aug 2026 · Georgia PSC docket 71526
v2.1Aug 23 2026Fix
Texas stopped data centers advancing through the interconnection process on August 3, and low tariff friction no longer means low friction.
Governor Abbott directed the PUCT and ERCOT to verify and audit every data center advancing through the interconnection process before any further approvals. ERCOT paused the Batch Zero study process in response. Applications are still accepted; what halted is advancement through the study.
The total request queue is roughly 474 GW, about 90 percent of it attributed to data centers, so the data center share is nearer 425 GW. ERCOT expects the audit to take under nine months. The 9 Apr 2027 Batch Zero deadline has not been formally extended, but ERCOT has said it will miss it and no replacement date exists (PUCT, 20 Aug 2026).
It is not a moratorium, but projects that fail the audit must be denied.
CenterPoint and Oncor both carry the pause on their alert line and a new interconnection status field reading paused, dated August 3 with a review date of April 9 2027. No score moved. A temporary processing state is not a jurisdictional characteristic, and encoding one as the other is how a model swings from low friction to maximum friction and back inside a year while the fundamentals sit still.
ERCOT still has the most flexible tariff structure of any tracked market. That flexibility is only available to load that can connect.
Fix
ERCOT repriced on access, not on terms.centerpoint.lltNote and oncor.lltNote now lead with the Aug 3 2026 Abbott directive and the ERCOT Batch Zero pause (~474 GW of requests, ~90% data centres, audit under nine months, Apr 9 2027 deadline unextended). Both alert lines carry it. In the risk ledger, centerpoint.exp.interconnect moves 3 → 1 and oncor.exp.interconnect2 → 1; the protection axes are untouched because no tariff term changed. Both tensionNote strings rewritten: the model previously concluded ERCOT had the most flexible structure of any tracked market, which is still true on terms and misleading without the access caveat.
Gov. Abbott letter to PUCT/ERCOT Aug 3 2026 · Gibson Dunn · Utility Dive
v2.0Aug 18 2026 · EE cross-model syncFixStructureNew
Forecast numbers now come straight from the standalone 2040 model, and two large collateral and downside errors are corrected.
Xcel Colorado's upfront collateral is US$600,000 flat. It had been shown as US$150 million for a 250 MW site, because a one-time charge was being multiplied by megawatts. That single error set the bar scale for every other utility in the collateral chart.
The constrained 2040 case drops from 337.6 GW to 303.3 GW, removing an 11 percent overstatement of the downside. The base case moves from 411.8 GW to 408.5 GW.
The tracker no longer runs its own copy of the forecast. It reads the published Forecast 2040 figures directly, so the two can no longer disagree.
We Energies triggers its financial security requirement at 100 MW, down from 500 MW, and Oracle dropped its challenge on August 17 2026.
43 of the 60 utilities tracked carry inferred rather than filed figures. The earlier count of 16 understated how much of this dataset is benchmark-grade.
Fix
Forecast 2040 demoted to a read-only mirror of the canonical series. This file was carrying its own copy of the fundamentals model at v3.1 while the standalone EE Forecast 2040 had moved to v1.9 and fixed a reversed LLT-friction exponent. The stale copy has been deleted, not re-ported: FM40_WEIGHTS and nine helper functions (sCurve, pueAt, computeGrowthAt, btmBypassAt, renewalRiskAt, computeNetDemandMW, lltFrictionAt, gridCeilingMW, computeDecomposition) are gone, and computeForecast2040() is now a thin shim over EE_SYNC.forecast. 2040 outcomes, stale → canonical: base 411.8 → 408.5 GW, accelerated 497.3 → 494.1 GW, constrained 337.6 → 303.3 GW, high-efficiency 125.7 → 134.7 GW. CAGR 2025–40: 15.1 → 15.0%, 16.5 → 16.5%, 13.6 → 12.8%, 6.3 → 6.8%. The constrained case was the material error — an 11% overstatement of the downside.
EE Forecast 2040 v1.9 · EE_SYNC v1.0
Fix
Collateral calculator: 250×–1000× overstatement on Xcel Colorado corrected. COLL_DATA.xcel carried perMW: 600000 and the calculator multiplied by facility MW, printing $150M at the 250 MW default. Every source string in this file describes it as a flat amount — RISK.xcel says "$120K study deposit + $600K total upfront before service begins", U.xcel says "$600K upfront". It is $600K flat. Entries now carry flat:true, renderCollateralCalc() skips the × MW multiplication for them, and the row is labelled "flat, not MW-scaled". This also repairs maxCap, which set the bar scale for every other utility in the chart.
CPUC Advice 2018-E · Apr 2 2026
Fix
Nav date restored. An init IIFE rewrote .nav-meta innerHTML and destroyed the #nav-date span, then two contradictory hardcoded writes followed ("Aug 17 2026", then "Jun 2026" four lines later). All three removed. The date is set by eeInitChrome() from EE_SYNC.vintage and now reads Data vintage Aug 2026.
EE_SYNC.vintage
Fix
Oracle / We Energies contradiction resolved. COLL_DATA.we said Oracle dropped its lawsuit Aug 17 2026 and the financial security requirement stands unchallenged, while the chart-note directly beneath the same table still read "Oracle legal challenge active". The note now carries the Aug 17 2026 state. The VLC threshold was also reconciled across five places: U.we, RISK.we (four descriptions) and COLL_DATA.we now all read ≥100 MW (lowered from 500 MW). The "highest threshold in US" clause was dropped — at 100 MW it ties Georgia Power’s ≥100 MW bespoke rather than leading.
Collateral tab no longer renders blank on first open. renderCollateral() was defined but never called — not from switchView() and not from init — so the tab was empty until a facility-size button was clicked, even though collMWSize defaults to 250 and the 250 MW button ships .active. Now wired to both.
Fix
f40-pue-2040 TypeError eliminated. updateForecast40KPIs() wrote to an element id that did not exist in the document, throwing on every Forecast 2040 tab open and every scenario toggle. A PUE 2040 KPI tile now exists and reads EE_SYNC.forecast.pue[2040] = 1.18; a BTM 2040 tile was added alongside it. The former "AI share 2040" tile was retired — AI share of IT load is a model internal owned by the canonical model and is not mirrored in EE_SYNC, so this file has no honest source for it.
Fix
Vintage bar count computed at runtime. buildVintageBar() hardcoded "Remaining 16 utilities" — written for a 25-utility dataset and stale ever since. It now counts conf:’inferred’ entries in U{} directly: 43 of 60. It cannot drift again.
U{} · 60 US utilities
Fix
Ramp green removed from text. #6b9a1a is a chart-ramp colour and fails contrast as small text. The "Tier 2" market-ranking label and the Dominion (VA) rate-trajectory series label both used it. Tier 2 moves to var(--green) and Tier 1 to the darker ink companion so the two stay distinguishable; the trajectory series moves to the var(--green) value. Fills are untouched.
Structure
Dead 2030 forecast system deleted — 553 lines. FC_MARKETS (46 market clusters), FC_SCENARIOS, computeForecast(), setScenario(), renderForecast() and its six chart renderers were unreachable: renderForecast() appeared exactly once in the file, its own definition. It targeted 11 DOM ids that do not exist here (chart-fc-area, chart-fc-cagr, chart-fc-power, chart-fc-scenarios, chart-fc-share, chart-fc-table, fc-baseline, fc-2030, fc-cagr, fc-gainer, fc-loser, scen-desc-text) and referenced seven utility keys absent from U{} (pepco, pso, entergy_ar, entergy_la, heco, gvea, delmarva). Its "14% / 18% / 10% sector CAGR" scenario blurbs had been copy-pasted into setScenario40() and were still on screen; those now read the canonical emergent CAGRs from EE_SYNC.
Structure
Forecast 2040 charts rebuilt on canonical data. The driver-decomposition waterfall, driver-trajectory series and sensitivity tornado all required mutating the local weight vector and re-running the model, so they could not survive its removal. In their place: a supply decomposition chart plotting EE_SYNC.forecast.supply (ftm + nuc + btm, which sums exactly to the base total — 325.9 + 24.2 + 58.4 = 408.5 GW at 2040), and a canonical scenario table reading capacity, CAGR and scenario probabilities straight from EE_SYNC. The supply split is published base-case only; non-base scenarios are scaled pro-rata and the chart says so. Parameter sensitivity now lives only in the canonical model.
EE_SYNC.forecast.supply
Structure
Palette migrated to the warm-paper house style. Shared token block, product bar, footer, sync badge and nav chrome brought onto the same system as the other five EE models. Cross-links added from Forecast 2040 to the standalone model and from Collateral to the Risk Ledger; the RV Model is rendered as a disabled chip because it has no deployed URL in EE_SYNC.products yet.
EE house style v1.0
New
This Changelog view. Matching the pattern used by the other EE models, so revisions to this tracker are auditable rather than silent.
Ownership Contract
Each number has exactly one authoritative model. A consuming model reads from EE_SYNC and never restates the figure inline — restating is how this file drifted.
Reads — $32.75/W is per critical IT watt, not facility watt
gas
Modular Gas Tracker
Reads — BTM pipeline, OEM backlogs, lead times
rv
RV Model
Reads — not yet deployed
The audit fails the build if these blocks drift between files. If a forecast number needs to change, it changes in EE Forecast 2040 first and propagates here — editing it in this file is the failure mode this version exists to close.
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.