brain/
sourcestock-market

Autoresearch: Evidence to graduate or keep-as-hypothesis the siting-gate rotation chain

Post-2026-08-24 pass on the stock-market hypothesis that state DC siting gates plus stranded capacity rotate demand from EPC/construction to existing-generation/BTM. Items 1 (PPA in lieu of cancelled new-build/BTM), 2 (PWR/CAT citing deferral as headwind), and 4 (flexibility bookings up vs EPC down) remain unmet. PA/NY/TX gates still in force; CAT BTM commentary is a tailwind.

Source

Autoresearch: Evidence to graduate or keep-as-hypothesis the siting-gate rotation chain

Generated by /autoresearch on 2026-09-19. Synthesized across 3 rounds from 13 web pages, anchored by Grokipedia Data center. See Provenance. Treat as raw material — review before promoting into a project or thread. Context: vault/projects/stock-market

Graduation bar under test (from siting-gate-plus-stranded-capacity-to-demand-side-beneficiary-rotation, last wiki update 2026-08-24): item 3 (second state) already met. Items 1, 2, 4 unmet. CAT was incorrectly lumped with PWR; BTM on-site gen is a beneficiary of the gate. This pass does not mint a 10th ai-infrastructure chain and names no buy/sell/size.

Summary

Three rounds of primary and IR search after 2026-08-24 do not graduate the chain. No hyperscaler or large load was found announcing a power-purchase agreement off existing generation in lieu of a previously announced new-build or behind-the-meter project (item 1). Neither Quanta (PWR) nor Caterpillar (CAT) has printed guidance that treats datacenter-project deferral or permitting delay as a demand headwind (item 2): PWR’s last print is Q2 (2026-07-30), with CEO Duke Austin saying a New York pause “doesn’t bother me at all”; CAT has not yet printed Q3, and at a 2026-09-10 Wells Fargo conference CEO Joe Creed said customers are not slowing and that “bring your own power” favors Cat. Storage bookings are up (Fluence record ~$6.4B backlog; Tesla 13.5 GWh Q2 deployments), but EPC is not decelerating — Quanta’s total backlog hit a record $53.4B and 2026 revenue guidance was raised to $39.3–$39.7B (item 4 fails the “vs EPC decelerating” half).

The gate leg is still live and, after 08-24, more instrumented: Pennsylvania PUC (2026-09-10) opened curtailment-rulemaking and a large-load cost-allocation conference to implement EO 2026-05; New York’s Community Investment Framework (~$1M/MW recommendation, 2026-09-15) and a PSC GEIS positive declaration / draft scope (comment through 2026-10-20) keep EO 62 in force; Texas’s ERCOT pause is still in force, with Batch Zero verification RFIs (2026-09-09), a community-impact RFI due 2026-10-12, reports due 2026-12-10, and a new 2026-09-14 Abbott directive that TWDB penalize water-survey noncompliance (criminal referral / TCEQ permit ineligibility). FERC ER26-3380 still has no order as of this pass; PJM’s 2026-09-10 answer still requests effectiveness by 2026-09-29.

Honest disposition: keep hypothesis. The gate is corroborated; the rotation (re-contracting over construction; PWR headwind; flexibility-vs-EPC split) is not.

Findings

Theme 1 — Item 1 unmet: no PPA off existing generation in lieu of a named cancelled new-build or BTM project

The bar is specific: a hyperscaler or large load announcing offtake from existing generation instead of a previously announced new-build or behind-the-meter campus — not merely signing another nuclear PPA, and not cancelling a campus without a substitute contract.

  • Constellation’s last IR print remains Q2 (2026-08-06): an additional 920 MW of long-term nuclear PPAs (15–20 years, start 2029–2032) with investment-grade customers, of which the 176 MW Walmart / Dresden deal is named and supports a 30 MW uprate at an existing Illinois plant (Constellation Q2 2026 results). That is offtake from the existing fleet (plus a small uprate). It is not disclosed as replacing a cancelled datacenter new-build or BTM project, Walmart is not a hyperscaler, and the counterparties for the remaining ~744 MW are unnamed. Crane (Microsoft) remains a restart of a retired unit, not a substitution for a cancelled campus; FERC CIR-transfer waiver and NRC fuel-license amendment are restart-path milestones, not item 1 (same CEG release).
  • The closest cancellation in the window is AWS Calvert Cliffs. Calvert County’s 2026-08-04 primary states Amazon Data Services “has formally withdrawn its Calvert Technology Center Conceptual Site Development Plan application and has decided not to pursue the proposed data center project in Calvert County” (Calvert County press). Trade press describes a ~500 MW / ~2.46 million sq ft colocated campus next to Constellation’s Calvert Cliffs plant, withdrawn amid a pending local six-month moratorium vote; AWS’s statement cites “development timelines, operational requirements, and our ability to deliver quickly for customers,” and says it “remains a valued partner” to Constellation — without announcing a replacement PPA off existing generation (DCD, 2026-08-05; Maryland Matters, 2026-08-04). Withdrawal ≠ “PPA in lieu.” The date is also before 2026-08-24; this pass found no post-08-24 AWS substitute contract that would close the test.
  • Other named campus changes in search (Oracle/OpenAI dropping an Abilene expansion; Meta/Clinton 2025 VPPA; CyrusOne new 380 MW facility adjacent to existing Freestone gas) are either not existing-fleet substitution, not in lieu of a cancelled BTM/new-build, or predate this window. CyrusOne/Freestone is new construction next to existing generation, the opposite geometry of item 1 (Constellation/CyrusOne, 2026-02-09).
  • No post-2026-08-24 Constellation IR or hyperscaler release was found that names a cancelled campus and a replacement existing-fleet PPA.

Item 1: unmet.

Theme 2 — Item 2 unmet: neither PWR nor CAT has cited DC deferral as a demand headwind; CAT’s latest commentary is a BTM tailwind

Quanta (PWR) — last print still Q2 2026-07-30; Q3 not yet issued. Consensus calendars put the next print around 2026-10-29 (estimate, not a company-confirmed IR calendar item as of this pass) (public.com earnings calendar). The Q2 8-K/EX-99.1 reports record remaining performance obligations of $33.6B and total backlog of $53.4B, and raises full-year 2026 revenue to $39.3–$39.7B, adjusted EPS $16.45–$16.95, adjusted EBITDA $4.09–$4.21B (Quanta EX-99.1, 2026-07-30). Permitting appears only as generic forward-looking risk language (“weather, regulatory, permitting, supply chain challenges… may impact… project timing”), not as a named datacenter-deferral headwind (same).

On the Q2 call, asked about the New York pause and whether state bans could knock projects out of backlog, Austin said technology is ~15–20% of the business, “I haven’t seen much in New York… having a moratorium in New York doesn’t bother me at all,” and that datacenter work enters backlog only with limited notices to proceed (“we tend to put in things that have only LNTPs, we don’t put the rest of it until the project is a go”) (stockanalysis Q2 2026 transcript). That is explicitly not item 2.

Caterpillar (CAT) — Q3 2026 not yet printed (last official print is Q2, 2026-08-04: sales $20.5B, +24%; Power Generation sales-to-users +72% on data-center large gensets/turbines; backlog $72B, +$9B sequential, +92% YoY; Creed: “no one is slowing down at the moment”) (Motley Fool CAT Q2 2026 transcript). Official Q2 HTML on caterpillar.com returned HTTP 500 this pass; figures above are from the Motley Fool transcript reprint of the 2026-08-04 call.

The post-08-24 CAT primary is not an earnings print. On 2026-09-10 Creed appeared at a Wells Fargo event (mislabeled “earnings call” on some aggregators). Analyst Jerry Revich asked about “slower pace of permitting” and ~100 GW of competing capacity. Creed: “We’re not seeing customers slow down right now in their need for power generation… there’s a lot more of bring your own power, which I think favors us actually… from a demand standpoint, we’re not really seeing customers change their behaviors right now.” He separately said “we’re seeing a lot more interest in natural gas sort of prime power behind the meter,” that standby diesel is not falling, that Texas-style curtailment-plus-local-gen is in customer design talks, and that “the great majority of the sites that we’re talking to, the plan is they’re the prime power” (earningscalls.dev / Wells Fargo 2026-09-10 transcript; stockanalysis investor-update transcript).

That is the opposite of item 2 as a PWR-style headwind, and it is consistent with the 2026-08-10 correction that CAT/BTM is a beneficiary of interconnection gates, not a co-victim with PWR. Treating CAT’s lag as the same evidence as PWR’s remains invalid.

Item 2: unmet. Next dated PWR test: ~2026-10-29 print. CAT Q3 historically prints late October.

Theme 3 — Item 4 unmet: storage bookings are up, but EPC is not decelerating

Item 4 requires both halves: grid-flexibility/storage bookings accelerating and EPC bookings decelerating.

Storage / flexibility (accelerating — half met as description, not as the test):

  • Fluence, fiscal Q3 2026 (quarter ended 2026-06-30, reported 2026-08-05): backlog ~$6.4B (company-record); ~$850M of data-center business through July, including a first large behind-the-meter order in the quarter and ~$550M hyperscaler awards in July 2026. Revenue guidance was cut (to ~$2.9–$3.1B) because of manufacturing/BOP delays, with ~$400M of deliveries slipped into FY2027 — a supply/execution hit, not an order-book collapse. CEO: “record order intake and backlog, and increasing momentum with all of our customer segments including data centers” (Fluence Q3 FY2026 release reprint; Fluence 10-Q via last10k).
  • Tesla energy, calendar Q2 2026: 13.5 GWh storage deployments (+53% sequential, +~40% YoY); CFO described the energy order backlog as “robust” and tied build plans to “future demand we expect from data centre growth”; Musk discussed Megapacks smoothing AI-training swings and unlocking off-peak grid capacity (Motley Fool TSLA Q2 2026 transcript; Tesla 10-Q reprint). No dollar backlog comparable to Quanta’s was disclosed.

EPC (not decelerating — the failing half):

  • Quanta total backlog $53.440B at 2026-06-30 vs $43.976B at 2025-12-31 vs $35.845B at 2025-06-30; Electric-segment backlog $43.790B. Full-year 2026 revenue raised to $39.3–$39.7B (Quanta EX-99.1). Austin’s Q2 language is “customers accelerate investment in the electric grid, power generation and mission-critical infrastructure,” not a siting-gate-driven slowdown.

Both books can rise at once. That is not a rotation from EPC to flexibility. Fluence’s guidance cut is a factory problem, not evidence that EPC is losing bookings to storage.

Item 4: unmet.

Theme 4 — Post-2026-08-24 gate updates: PA, NY, TX still in force; new implementing instruments; FERC still no order

These updates strengthen the already-met gate leg (item 3). They do not substitute for items 1, 2, or 4.

Pennsylvania EO 2026-05 (signed 2026-08-18) — still in force; PUC now implementing.

  • 2026-09-10: PA PUC voted 5–0 on Vice Chair Barrow’s motion to draft updated emergency load-control / curtailment rules (order, criteria, applicability) and to notice a fall technical conference on cost allocation for large computational loads, including data centers. Staff tentative order targeted for the 2026-10-01 public meeting; final order targeted 2026-01-28. The release states the actions “follow Gov. Josh Shapiro’s Aug. 18 Executive Order encouraging the PUC to examine data center curtailment procedures and measures addressing the allocation of data center-related costs.” PJM July 2026 BRA shortfall cited as 6,831 MW for 2028–29 (PA PUC press, 2026-09-10; docket M-2026-3064961). This is implementation of EO 2026-05, not a new moratorium and not a PPA-in-lieu.
  • 2026-09-14: PUC released a Synapse Energy Economics resource-adequacy study. Chairman DeFrank: under the reference case, 2030 reliability risk is “nearly six times worse than PJM’s planning standard”; a severe high-load/low-supply case models >13 loss-of-load events per year from 2030 vs the 1-in-10 standard. The commission says it is “supporting Gov. Josh Shapiro’s recent Executive Order” and last week adopted a motion that “existing customers must be protected ahead of newly-connecting large loads such as data centers” (PA PUC press, 2026-09-14). Report URL: Assessing Electricity Resource Adequacy in Pennsylvania and PJM (state .gov PDF; text extracted in search). The study argues more generation is needed to match load — which is not by itself evidence that the marginal MW is met by re-contracting rather than construction.

New York EO 62 — still in force; CIF posted; GEIS scoping started; 2026-10-12 DPS 90-day report not yet due.

  • Reuters (2026-09-15/16): Hochul/ESD recommending host communities seek at least $1 million per megawatt of utility demand as a Community Investment Framework benchmark; ESD President Hope Knight: data centers “are not major local job generators.” Amazon, Meta, Google did not immediately comment; Microsoft declined (Reuters). This is the EO 62-directed ESD framework (60-day clock from 2026-07-14). ESD’s own page still showed “will be released by mid-September” when fetched as a stub (esd.ny.gov/communityinvestmentframework); governor.ny.gov EO 62 HTML timed out this pass — cite the standing order URL without re-extracted body: EO No. 62.
  • Enerknol (2026-09-18), citing a 2026-09-17 PSC action: positive declaration under SEQRA, draft GEIS scope out for comment through 2026-10-20, covering data-center development and large-load interconnection/cost-allocation reforms under Case 26-E-0045 (Enerknol). A matching DPS PDF of the data-center draft scope was not retrieved this pass (a 2024 RAPID-Act GEIS scope is a different proceeding — do not conflate). Treat the 09-17 scoping as trade-press pending primary PDF. The pause still ends on an event (final GEIS + findings), not a date. The first EO 62 dated milestone — DPS’s 90-day report to the Commission on TO large-load study practices — is 2026-10-12 and has not arrived.

Texas ERCOT large-load pause — still in force; verification underway; new TWDB enforcement overlay.

  • 2026-09-09 ERCOT Market Notice M-A090926-01: Batch Zero verification RFIs issued; complete responses + notarized attestations due 10 business days (5-day extension cap); successful verification is a condition of Batch Zero inclusion; failure → exclusion from Batch Zero (ERCOT M-A090926-01).
  • Houston Public Media (2026-09-11), reporting ERCOT’s Friday commission presentation: community-impact survey of 461 data centers (247 ≥75 MW, 214 of 25–75 MW), ~130 questions, responses into mid-October; 362 large loads conditionally in Batch Zero totaling 191.8 GW; 373 projects excluded, requesting >300 GW; 231 verification RFIs already out, first responses ~Sep 23; final audit results December; individual responses confidential (Houston Public Media). Energization pause for not-yet-connected large-load data centers/crypto ≥75 MW continues until verification + community-impact review complete.
  • New after 08-24: Governor Abbott, 2026-09-14, directed TWDB to enforce Water Code survey requirements against major water users including data centers (criminal offense for failure to return the survey; ineligibility for TCEQ Ch. 11 permits; referral to county/district attorney) and to partner with ERCOT on the audit. TWDB progress update due 2026-10-14 (gov.texas.gov press; letter PDF TWDB_Letter.pdf).
  • Same day, ERCOT Market Notice M-B091426-01 issued the State and Community Impact RFI to Batch Zero ILLEs not yet energized at ≥60 kV and 25–75 MW “Substantiated Load” data centers not yet energized; responses due 2026-10-12 17:00 CPT; Section C water questions issued on TWDB’s behalf; failure to fully answer Section C “may subject an RFI recipient to criminal penalties”; failure to respond reported to PUC and the Governor; report expected no later than 2026-12-10. This RFI is separate from the Sep 9 Batch Zero Eligibility RFI (ERCOT M-B091426-01; ERCOT Data Center Impact RFI page).

FERC ER26-3380 — still no order.

  • PJM’s 2026-09-10 answer in ER26-3380 still asks the Commission to accept the Reliability Backstop Procurement no later than 2026-09-29; the Sep 30–Oct 21 offer window is contingent on that order (PJM 2026-09-10 answer PDF). No FERC order accepting or rejecting the filing was found in this pass. Calendar hold continues, 10 days before the requested date.

Theme 5 — CAT vs PWR split (not a new chain)

Creed’s 2026-09-10 BYOP / BTM-prime-gas comments are independent corroboration of the 08-10 correction, not graduation of the rotation. If the BTM exemption is the live trade, CAT/CMI/BE/GEV should decouple upward from PWR over the gate’s duration (graduation item 5 on the wiki page). This pass collected commentary, not a tape test, and does not re-rate anyone.

Grokipedia’s Data center primer (U.S. electricity share, EIA growth, sustainability controversies) is background only. It does not speak to state gates or the graduation tests.

Contradictions and open questions

  • Item 1 remains the binding identification test. Existing-fleet PPAs (CEG 920 MW, Meta/Clinton, Microsoft/Crane) show incumbents can re-contract. They do not show a hyperscaler abandoned a named new-build or BTM project in favor of that offtake. AWS Calvert is a cancellation without a substitute PPA. Until a named substitution exists, steps 4–5 of the chain stay open/partial at best.
  • Item 2 cannot be tested on a Q3 print that has not happened. PWR ~2026-10-29 is the next clean window. Using CAT as a co-headwind name would re-introduce the 08-10 error.
  • Item 4 is currently the wrong sign on the EPC leg. Fluence/Tesla up and Quanta up is compatible with “everything AI-power related is still booking,” which is the null of a rotation.
  • PA PUC Synapse study vs the rotation hypothesis. The study’s implication is that PJM needs new resources to restore LOLE, not that stranded capacity plus DR can absorb the load. That is in tension with CEG’s “bedrock is existing generation” Q2 framing (already in the wiki from 2026-08-06) and should be held as a live contradiction, not silently resolved.
  • NY GEIS primary PDF not fetched. Enerknol reports a 2026-09-17 positive declaration and an Oct 20 comment deadline; confirm against documents.dps.ny.gov Case 26-E-0045 before treating scope language as primary.
  • FERC ER26-3380 may print as soon as 2026-09-29. An order (accept / reject / deficiency) is a calendar event, not a graduation event for this chain.
  • ERCOT Batch Zero counts will move. Seely (via Houston Public Media) said dispute processes are open; 362 / 191.8 GW is conditional.
  • SemiAnalysis 2026-09-15 (already in the wiki on 2026-09-16: ~2.3 GW genuine delay, Texas pause as 3–4 months of admin delay offset by BTM) was not re-fetched this pass and is not used as a citation here.

Provenance

Rounds run: 3 of 3 (full)

Sub-questions by round:

Round 1 (broad survey):

  1. Hyperscaler PPA off existing generation in lieu of a previously announced new-build or BTM project
  2. PWR or CAT guidance citing DC deferral / permitting delay as a demand headwind
  3. Flexibility/storage bookings accelerating vs EPC decelerating
  4. Post-2026-08-24 updates to PA EO 2026-05 / NY EO 62 / Texas ERCOT large-load pause

Round 2 (drill-down):

  1. AWS Calvert Cliffs withdrawal — cancellation vs substitute PPA — targeting item 1 near-miss
  2. Quanta Q2 backlog/guidance vs Fluence/Tesla storage — targeting item 4
  3. NY CIF / GEIS scoping after 08-24 — targeting EO 62 duration/instrument
  4. FERC ER26-3380 order vs calendar hold — targeting PA/PJM overlay

Round 3 (resolve remaining uncertainty):

  1. Abbott TWDB water-penalty directive — new primary after 08-24? — targeting Texas gate update
  2. NY PSC GEIS positive declaration — primary vs retell
  3. Any post-08-24 named PPA substituting a cancelled campus

Anchor source (Grokipedia, fetched before round 1):

  • Data center — 12,017 chars extracted (capped), 375 citations, source: api — encyclopedic primer on facility types, power/cooling, U.S./global electricity use. Not used as load-bearing evidence for graduation items.

URLs fetched (13 successful, 2 failed):

Round 1:

Round 2:

Round 3:

Search hits used with extracted body (not a separate WebFetch round):

Tools used: WebSearch, WebFetch, grokipedia-fetch (_lib/grokipedia.py fetch). Generated: 2026-09-19 ~20:00 UTC

Referenced by