Interpretation guideGUIDE-02

The same number can describe a request, a scenario, a contract or a physical load.

The Evidence Ledger keeps each figure attached to its register, state, unit basis, geography, vintage and caveat. A cross-record total is prohibited when any of those dimensions differ.

Executive summary

Read the conclusion before the archive.

  1. A shared MW label does not create a shared denominator.
  2. Forecast, request, contract, energized load, MW-IT, average MW and annual energy answer different questions.
  3. A conversion is allowed only when explicit bridge data and assumptions are published.

01 · Core argument

A number is not portable across evidence regimes.

Every quantity remains attached to what it measures, when it was observed, which geography it covers and which state it describes.

The absence of those dimensions makes an impressive total less informative, not more.

02 · Four-register discipline

Load, Supply, Money and AI-share are separate ledgers.

They can be read together as context. They cannot be collapsed into a single AI-compute capacity figure.

Four-register discipline

Every figure stays inside its evidentiary register.

Movement between registers requires an explicit, source-stated bridge. Missing bridge data remains unknown rather than being imputed.

  1. R01

    Load

    Requests, forecasts, design power, energized load, average MW and energy remain distinct.

    CLM-0005CLM-0033

  2. R02

    Supply

    Service model, deliverability, generation and physical inputs stay attached to their own pathway.

    CLM-0053

  3. R03

    Money

    Capex, backlog, RPO, contracted power, revenue and utilization are not bridge-equivalent.

    CLM-0180

  4. R04

    AI-share

    Total data-center measures are not AI-only unless the source supplies the cut.

    CLM-0029

AI-share boundary

Broader infrastructure does not become AI-only by implication

The source must define which part of the broader quantity is attributable to AI.

01

Broader measured scope

Campus, data-center, portfolio, utility, accelerator or financial quantity.

Explicit source bridge

Named AI workload or customer scope · denominator · period · allocation method.

02

AI-attributable subset

Reportable only to the extent directly defined by the source.

Required bridge test

Without the bridge, retain the broader label and stop the attribution.

Control rule: GPU count, company narrative, total campus MW or portfolio capex does not supply an AI-share denominator on its own.

03 · Quantity types

Six familiar labels answer six different questions.

The comparison below is semantic, not additive.

Requested MW
What a customer asked a utility or grid process to study. It is not approval or delivery.
Forecast MW or GW
What a planning model includes under defined assumptions and screening rules. It is not a queue total.
Contracted MW
What an agreement or company disclosure describes. It is not energized or average load.
Energized MW
What electrical infrastructure has reached a live state. It does not by itself establish installed or utilized compute.
MW-IT, average MW and annual MWh/TWh
Equipment load, time-averaged power and annual energy are different physical measures. Conversion requires explicit assumptions.
Economic or utilization measures
Revenue backlog, active power, cabinet utilization and customer use have different definitions and cannot be translated without bridge data.

04 · Bridge test

A conversion requires a denominator, period and mechanism.

Without all three, the original quantity may be reported but not translated.

01

Same object and scope?

Portfolio, region, campus, phase and building totals cannot be substituted for one another.

02

Same state and time basis?

Requested, contracted, peak, average and annual values describe different observations.

03

Published conversion rule?

Efficiency, load factor, PUE, duration and AI-share assumptions must be explicit.

Two-sided gap

The conversion gap can widen from either side.

Upstream contraction: screening, service-model limits, policy uncertainty and local process can reduce or delay the quantity that reaches operation.

CLM-0205CTR-0001

Demand-side contraction: efficiency, forecast revision and demand quality can reduce the load or economic use associated with an upstream figure.

CTR-0002CTR-0007

Decision boundary

No synthetic total

Permitted conclusion

Display each figure with its original label, state, unit basis, geography, vintage and caveat.

Prohibited stronger claim

Do not add, rank or convert heterogeneous figures into a market total or realization percentage.

Quantitative ceiling
Not summable in this edition.
Causal ceiling
No quantity alone establishes the mechanism that produced it.

05 · Claim register

Evidence boundaries behind the unit rules.

Each Claim ceiling and set of Linked sources remains visible for audit.

CLM-0005 · interpretation

Announced, requested, approved, contracted, deliverable, energized, installed, utilized, and economically productive capacity are non-equivalent states and units.

Claim ceiling
No conversion without an explicit method and evidence.
Evidence
CONTROLLING/BOUNDING INPUT · HIGH
Locator
Operating Prompt §7; Seed Charter §6
Linked sources
SRC-0001internal recordSRC-0002internal record

CLM-0029 · forecast

LBNL estimates U.S. data centers used about 176 TWh in 2023, approximately 4.4% of U.S. electricity; this is a modeled total-data-center baseline, not AI-only demand or operating site capacity.

Claim ceiling
National total-data-center annual energy estimate only; no AI-only or project-state inference.
Evidence
Tier A · HIGH for report values; MEDIUM for national estimate precision
Locator
Executive Summary pp. 5–7; historical series and Figure 5.5
Linked sources
SRC-0023

CLM-0033 · interpretation

TWh/year, equivalent average GW, source-defined facility capacity, system peak GW, requested load, design capacity, MW-IT, GPU count, and announced capex are non-interchangeable metrics and are held in separate comparable groups.

Claim ceiling
Normalization rule; no conversion beyond explicit formula and assumptions.
Evidence
Tier A + CONTROLLING INPUT · HIGH
Locator
Operating Prompt §§7–8; Seed Charter §§3,6; quantitative source locators in Pass 02 ledgers
Linked sources
SRC-0001internal recordSRC-0002internal recordSRC-0023SRC-0012SRC-0018

CLM-0016 · forecast

PJM’s 2026 load forecast process screens large-load adjustment requests rather than automatically including the full requested amount.

Claim ceiling
Forecast-inclusion method, not validation of each included project.
Evidence
Tier A · HIGH
Locator
SRC-0029 methodology; SRC-0030 criteria; SRC-0031 slides 30–31

CLM-0019 · interpretation

ERCOT’s April 2026 aggregate of roughly 410 GW seeking interconnection is a request-pipeline measure, not approved, deliverable, energized, or forecast peak load.

Claim ceiling
Request-pipeline snapshot only; no project count or capacity realization rate.
Evidence
Tier A · HIGH
Locator
SRC-0038 aggregate request slides; process page

CLM-0098 · technical state

A reported contract or committed MW quantity is a contractual/company-disclosure state and cannot be counted as deliverable power, energized load, average consumption or operating AI capacity.

Claim ceiling
No capacity-state upgrade without project-specific evidence.
Evidence
Tier A-minus · HIGH
Locator
Company aggregate contract disclosures
Linked sources
SRC-0101SRC-0116

CLM-0180 · interpretation

RPO, revenue backlog, equipment backlog, active power, contracted power, cabinet utilization and development capacity are non-equivalent measures.

Claim ceiling
No cross-metric aggregation or conversion to economic compute without disclosed bridge data.
Evidence
Tier A · HIGH
Locator
Referenced filing definitions

CLM-0205 · interpretation

At least five rival theses win materially across the selected portfolio: forecast inflation/screening, policy and data uncertainty, partial private-risk allocation, local veto/process risk and demand-quality limits.

Claim ceiling
Rival wins are mechanism-specific and do not generalize to all U.S. projects.
Evidence
Tier A · HIGH for bounded portfolio findings
Locator
Selected case evidence and PASS_08_RIVAL_WIN_REGISTER.csv

Rival evidence that narrows pipeline, demand and contract figures

CTR-0001 · Forecast inflation

Requested and announced pipelines can materially exceed screened forecasts or realized load.

Evidence result:PJM differentiates utility demand asks from preliminary/final included adjustments; FERC reported a 15% lower 2032 AEP-zone forecast versus the prior vintage after large-load adjustments.

Ceiling effect:No pipeline-to-realization conversion; preserve vintage and screening.

CTR-0002 · Efficiency response

Hardware, software, workload, and facility efficiency can reduce energy per useful compute and alter forecast demand.

Evidence result:The LBNL national report explicitly models efficiency, utilization, shipments, and cooling assumptions and records historical efficiency effects.

Ceiling effect:Forecasts require scenario ranges; no fixed power-per-compute assumption.

CTR-0007 · Demand-quality problem

Contracted or installed capacity may not yield high utilization, customer use, revenue, or durable economic output.

Evidence result:NOVEC testimony shows a large difference between served and contracted aggregate load; the program’s C18–C20 states require later operating/economic evidence.

Ceiling effect:No contracted-to-utilized/economic conversion.

CTR-0008 · Policy and data uncertainty

Evolving rules, changing forecast vintages, and inconsistent definitions may be a principal planning and evidentiary cost.

Evidence result:FERC, NERC, PJM, ERCOT, and state records are contemporaneously changing and use different units/states.

Ceiling effect:Every current claim requires cutoff and reopen trigger.

CTR-0030 · Forecast inflation / demand-quality

Contract and commitment aggregates do not establish project realization.

Evidence result:Georgia company disclosures lack a public project-level deliverability/energization crosswalk.

Ceiling effect:No pipeline or operating-capacity addition.

Public-record gaps

NEG-0013

Question checked:Can FERC’s reported >50 GW in-service data-center capacity be reproduced project by project from the public report?

SOURCE GAP CONFIRMED IN PASS 09. The FERC report identifies a Yes Energy Load Center Project Database source, while public Yes Energy pages describe a proprietary/commercial dataset but do not publish project-level inclusion, status-transition, duplicate-control or C14–C20 reconciliation rules sufficient for reproduction.

Implication:Use only as a source-defined external benchmark; do not add to the project ledger or treat “in service” as a program C-state.

NEG-0014

Question checked:Is public measured operating MW available for the four bounded C17 project records?

UNKNOWN — PUBLIC EVIDENCE NOT LOCATED. Operating states are disclosed, but common measured average/peak operating load is not.

Implication:No aggregate operating MW; unknown must not be recorded as zero.

NEG-0022

Question checked:Can ERCOT aggregate large-load interest be reconciled to project-level approved, deliverable and energized MW?

SOURCE GAP. Aggregate interest/process evidence does not provide a complete public project-level C7–C17 crosswalk.

Implication:No project realization ratio or operating total from aggregate request figures.

06 · Maintenance

Evidence that can change a quantity or its basis.

Revisions are recorded when the source publishes a changed value, denominator, period, state or conversion method.

Evidence that could change a quantity or its basis

  • OIR-0005 · ERCOT releases a newer large-load request, screening, forecast, withdrawal, or interconnection-status dataset.

    Required update: Preserve prior snapshot; add new vintage, state definitions, attrition/duplication changes, and affected aggregates.

  • OIR-0008 · New official demand forecast, project pipeline, operational-data accuracy report, or material project cancellation/energization is published.

    Required update: Add vintage rather than overwrite; reconcile project IDs; update state and double-counting register.

  • OIR-0017 · A new official national or regional demand forecast, AEO/LBNL update, or method revision changes scope, horizon, scenario range, or unit bridge.

    Required update: Add new vintage without deleting prior; reconcile scope, method, denominator and supersession; rerun normalization.

  • OIR-0020 · Public evidence provides measured operating load, commissioned IT capacity, customer-use, utilization, revenue/economic incidence, or independent verification for a C17 project.

    Required update: Verify unit, period, site boundary and source function; upgrade only the supported state; preserve contrary evidence and C18–C20 distinctions.

  • OIR-0030 · Project-level official records resolve Georgia aggregate contracts/commitments into request, contract, construction, energization or operating states.

    Required update: Create project-specific state observations; prevent double counting and preserve units.

Record navigation

These records share claims or sources with this report. The linkage supports navigation, not a causal inference.

Bibliography and exact locators

  1. SRC-0001 — AI Compute Infrastructure — autonomous research operating prompt

    Operator-supplied controlling input · 2026-06-18 · CONTROLLING INPUT

    Locator: Full file; sections 0–14

    OPERATOR-SUPPLIED PROGRAM INPUT

  2. SRC-0002 — AI Compute Infrastructure — research seed charter

    Operator-supplied bounded brief · 2026-06-18 · BOUNDING INPUT

    Locator: Full file; sections 1–14

    OPERATOR-SUPPLIED PROGRAM INPUT

  3. SRC-0023 — 2024 United States Data Center Energy Usage Report

    Lawrence Berkeley National Laboratory / U.S. Department of Energy · 2024-12-19 · Tier A

    Locator: Executive Summary pp. 5–7; Figure 5.5 pp. 52–53; methods and limitations

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  4. SRC-0012 — 2025 State of the Markets Report

    Federal Energy Regulatory Commission, Office of Enforcement · 2026-03-26 · Tier A

    Locator: PDF pp. 9–10 (report pp. 7–8), Figure 5 and note 14; other market-review figures as separately cited

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  5. SRC-0018 — 2025 Long-Term Reliability Assessment

    North American Electric Reliability Corporation · 2026-01-23 · Tier A

    Locator: PDF p. 16 for PJM 210/56 GW; PDF p. 131 for ERCOT 6,700/23,000/25,000 MW and realization calibration

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  6. SRC-0029 — 2026 PJM Load Forecast Report

    PJM Interconnection · 2026-01-14 · Tier A

    Locator: Executive summary; large-load adjustment methodology; zone tables

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  7. SRC-0030 — Load Adjustment Request Implementation

    PJM Interconnection · 2025-07-01 · Tier A-minus

    Locator: Implementation criteria and process

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  8. SRC-0031 — Load Adjustment Requests Summary for 2026 Load Forecast

    PJM Interconnection · 2025-11-24 · Tier A-minus

    Locator: Slides 30–31; ask versus preliminary included

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  9. SRC-0037 — Large Load Integration

    Electric Reliability Council of Texas · 2026-01-01 · Tier A-minus

    Locator: Current forms and process links

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  10. SRC-0038 — ERCOT Large Load Update — House State Affairs Hearing

    Electric Reliability Council of Texas · 2026-04-09 · Tier A-minus

    Locator: Aggregate request total and composition slides

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  11. SRC-0039 — 2025 ERCOT Annual Report

    Electric Reliability Council of Texas · 2026-03-19 · Tier A

    Locator: Large-load and planning sections

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  12. SRC-0101 — Georgia Power highlights first contracts under new rules and regulations

    Georgia Power · 2025-09-30 · Tier A-minus

    Locator: Lines 251-264; nearly 2 GW of new contracts filed; additional negotiations and forecast assertions

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  13. SRC-0116 — Georgia Power plans additional generation and reports large-load commitments

    Georgia Power · 2026-04-23 · Tier A-minus

    Locator: 32 large-load customers committed to approximately 15,600 MW; 21 projects under construction; resource requests remain subject to PSC process

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  14. SRC-0152 — Alphabet Inc. — Quarterly Report on Form 10-Q for quarter ended March 31, 2026

    Alphabet Inc. · 2026-04-23 · Tier A

    Locator: Lines 590; 1968–1972; 2206–2217; capital expenditures, Cloud RPO, asset placement in service

    PUBLIC COMPANY/SEC RECORD — CITE/SUMMARIZE

  15. SRC-0154 — Quarterly Report on Form 10-Q for quarter ended February 28, 2026

    Oracle Corporation · 2026-03-10 · Tier A

    Locator: Lines 1666–1670, 2948, 6529–6756; RPO, power obligations, capex and financing

    PUBLIC COMPANY/SEC RECORD — CITE/SUMMARIZE

  16. SRC-0155 — CoreWeave, Inc. — Annual Report on Form 10-K for year ended December 31, 2025

    CoreWeave, Inc. · 2026-02-27 · Tier A

    Locator: Lines 330–351; 3001–3041; active and contracted power, RPO, take-or-pay contracts, concentration and financing

    PUBLIC COMPANY/SEC RECORD — CITE/SUMMARIZE

  17. SRC-0156 — CoreWeave Reports Strong First Quarter 2026 Results

    CoreWeave, Inc. / SEC Exhibit 99.1 · 2026-05-07 · Tier A-minus

    Locator: First-quarter 2026 financial highlights; revenue backlog $99.4 billion at 2026-03-31

    PUBLIC COMPANY/SEC RECORD — CITE/SUMMARIZE

  18. SRC-0157 — Equinix, Inc. — Annual Report on Form 10-K for year ended December 31, 2025

    Equinix, Inc. · 2026-02-13 · Tier A

    Locator: Lines 350–384; 1044–1045; 1553–1571; 1736–1756; 2191–2198; 5739; 6154–6167

    PUBLIC COMPANY/SEC RECORD — CITE/SUMMARIZE

  19. SRC-0162 — Eaton Corporation plc — Quarterly Report on Form 10-Q for quarter ended March 31, 2026

    Eaton Corporation plc · 2026-05-05 · Tier A

    Locator: Lines 492–493, 1128–1163; backlog, data-center demand, margins; acquisition notes lines 327–396

    PUBLIC COMPANY/SEC RECORD — CITE/SUMMARIZE

  20. SRC-0163 — Vertiv Holdings Co — Annual Report on Form 10-K for year ended December 31, 2025

    Vertiv Holdings Co · 2026-02-18 · Tier A

    Locator: Lines 378–389 and 637–639; combined backlog, delivery expectation, cancellation/rescheduling risk

    PUBLIC COMPANY/SEC RECORD — CITE/SUMMARIZE

  21. SRC-0164 — GE Vernova Reports First Quarter 2026 Financial Results

    GE Vernova Inc. / SEC Exhibit 99.1 · 2026-04-22 · Tier A-minus

    Locator: Lines 40–53; $2.4 billion Electrification equipment orders supporting data centers; total orders and revenue

    PUBLIC COMPANY/SEC RECORD — CITE/SUMMARIZE

  22. SRC-0006 — Sunshine Notice — June 2026 Commission Meeting

    Federal Energy Regulatory Commission · 2026-06-11 · Tier A

    Locator: Agenda Item E-1; RM26-4-000

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  23. SRC-0114 — SCC Data Center Initiatives — Facts and Figures

    Virginia State Corporation Commission · 2026-02 · Tier A-minus

    Locator: pp. 1-2: 14-year obligation, 85% T&D minimum, up to 60% collateral, alternative cost-allocation filings

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  24. SRC-0120 — Digital Gateway rezoning appeals opinion

    Court of Appeals of Virginia · 2026-03-31 · Tier A

    Locator: Opinion pp. 4–5 and 52–53; records identified in Pass 05

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  25. SRC-0168 — Public-interest intervenor filing concerning Georgia Power large-load forecast and pipeline

    Georgia PSC public docket participant filing · 2025-12-16 · Tier A-minus

    Locator: PDF pp. 6–10 and 17; discusses 14,345 MW removal, pipeline screening, contract uncertainty and flexibility assumptions

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  26. SRC-0165 — Order Accepting Tariff Revisions, Subject to Condition — Conditional High Impact Large Load Service

    Federal Energy Regulatory Commission · 2026-06-05 · Tier A

    Locator: 195 FERC ¶ 61,196; paras. 6–7, 104–105; ordering paras. A–B; effective July 1, 2026

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  27. SRC-0066 — Crusoe Announces Flagship Abilene Data Center is Live

    Crusoe · 2025-09-30 · Tier A-minus

    Locator: Construction began June 2024; first two buildings energized; first racks June 2025; early training and inference workloads running

    PUBLIC COMPANY/SEC RECORD — CITE/SUMMARIZE

  28. SRC-0032 — Load Forecast Accuracy Report — May 2026

    PJM Interconnection · 2026-05-29 · Tier A

    Locator: Overview p. 3; operational-data comparisons

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  29. SRC-0027 — Best Practice Guide for Data Center Design

    U.S. Department of Energy · 2024-07-01 · Tier A-minus

    Locator: Full guide

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  30. SRC-0052 — NOVEC Evidence for Virginia Load Growth Technical Conference

    Virginia SCC docket / Northern Virginia Electric Cooperative · 2024-12-16 · Tier A-minus

    Locator: Testimony: 58 buildings; 1,408 MW served; 3,513 MW contracted

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  31. SRC-0003 — Interconnection of Large Loads to the Interstate Transmission System (Docket No. RM26-4-000)

    Federal Energy Regulatory Commission · 2026-01-13 · Tier A

    Locator: Docket landing page; At a Glance; RM26-4-000; official page rechecked in Pass 09

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  32. SRC-0015 — Assessment of Gaps in Existing Practices, Requirements, and Standards for Emerging Large Loads

    North American Electric Reliability Corporation · 2026-03-01 · Tier A

    Locator: Executive Summary p. vi; gap categories

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  33. SRC-0040 — Long-Term Load Forecast Update 2025–2031 and Methodology Changes

    Electric Reliability Council of Texas · 2025-04-07 · Tier A-minus

    Locator: Forecast methodology and adjustments

    PUBLIC OFFICIAL RECORD — CITE/SUMMARIZE

  34. SRC-0071 — Meta’s Kansas City Data Center is now operational

    Meta · 2025-08-01 · Tier A-minus

    Locator: Company statement that facility is operational and serving traffic

    PUBLIC COMPANY/SEC RECORD — CITE/SUMMARIZE

  35. SRC-0072 — Colossus

    xAI · UNSTATED — CURRENT PAGE AT RETRIEVAL · Tier A-minus

    Locator: Company statement: 200,000 NVIDIA H100 GPUs in a single interconnected cluster

    PUBLIC COMPANY/SEC RECORD — CITE/SUMMARIZE

  36. SRC-0076 — Google announces new investment in Jackson County, Alabama data center

    Google · 2026-06-15 · Tier A-minus

    Locator: Existing campus operating since 2019; $1.5 billion expansion announced for 2026–2027

    PUBLIC COMPANY/SEC RECORD — CITE/SUMMARIZE