brain/
sourcestock-market

Autoresearch: Texas PUCT/ERCOT 2026-08-20 Batch Zero good-cause exceptions and BTM datacenter generation

First-party readout of the 2026-08-20 PUCT good-cause order (process delay, not a BTM-exemption rewrite); no 2026 GEV LM2500 datacenter unit count; no Bloom-named MW increment past Oracle 1.2/2.8 GW.

Source

Autoresearch: Texas PUCT/ERCOT 2026-08-20 Batch Zero good-cause exceptions and BTM datacenter generation

Generated by /autoresearch on 2026-09-19. Synthesized across 3 rounds from 13 web pages (2 fetch failures), anchored by Grokipedia Bloom_Energy / Electric_Reliability_Council_of_Texas / public_utility_commission_of_texas after behind the meter power resolved to an unrelated cycling page. See Provenance. Treat as raw material — review before promoting into a project or thread. Context: vault/projects/stock-market Priors: skipped at request.

Summary

The 2026-08-20 PUCT action in Project No. 59142 granted ERCOT three good-cause exceptions to Batch Zero timelines, not a rewrite of the behind-the-meter / WLPUN eligibility rule. Classification of Large Loads (≥75 MW) was delayed so ERCOT can finish the Abbott-directed verification/audit; conditionally eligible base-load projects may still sit in the August and November 2026 quarterly stability assessments; late dynamic-data review got a 24-day cure. That is a process tightening (pause, audit, community-impact RFI) around grid interconnection, not a widening of a BTM exemption and not a first-party narrowing of WLPUN tech scope. First-party ERCOT HTML still treats WLPUN/BYOG as a Batch Zero study path that requires a Full Interconnection Study for new co-located generators; this pass did not retrieve a first-party HTML page that names fuel cells as a WLPUN resource (secondary counsel notes ERCOT examples of gas turbines and solar). GE Vernova's 2026-07-22 SEC exhibit names 61 aeroderivative turbines ordered in Q2 2026 without an LM2500 or datacenter split; the last first-party LM2500XPRESS datacenter unit count retrieved here remains Crusoe's 29 units / ~1 GW (July 2025). Bloom's April 2026 Oracle MSA is still 1.2 GW contracted / up to 2.8 GW; Bloom's July 28, 2026 Q2 release names no incremental MW. AEP's January 2026 8-K (pre-Oracle) records a "substantial portion" of a 900 MW SOFC option for a Cheyenne facility; this pass found no Bloom-named MW increment past Oracle.

Findings

What August 20 actually decided

On August 10, 2026, ERCOT asked the PUCT, in Project No. 59142, for good-cause exceptions under 16 TAC § 25.3(b) so it could deviate from Planning Guide §§ 9.3.1(2)(a), 5.3.5(5)(a)(ii), and 9.2.2(3)(a) while it paused Batch Zero to run the Abbott audit (ERCOT market notice M-A080326-02; PUCT Order Granting Good Cause Exceptions, Item 53, filed 2026-08-20).

The order, signed by Chairman Thomas J. Gleeson and Commissioners Kathleen Jackson and Courtney K. Hjaltman, grants exactly those three exceptions (PUCT Order, Project 59142 Item 53):

  1. Excuse the August 7, 2026 Batch Zero classification deadline. Planning Guide § 9.3.1(2)(a) had required ERCOT to classify Large Loads as base load, studied load, or not included by that date. The Commission lets ERCOT delay classification so it can verify developer submissions first.
  2. Allow conditionally eligible (not yet finally classified) base-load Large Loads into the August 1 and November 1, 2026 quarterly stability assessments. Without this, the classification pause would have emptied the QSA of new Large Loads.
  3. Let ERCOT finish dynamic-data review after August 7 and give ILLEs the same 24-day cure that on-time notices received under § 9.2.2(3)(a).

ERCOT's August 21, 2026 market notice restates the grant and says ERCOT intends to notify interconnecting DSPs/TSPs of each Large Load's conditional classification by August 31, 2026, which starts the Planning Guide § 9.3.1(2)(a)(i)–(iii) dispute/reconciliation process (M-A080326-02).

The order's recitals also record the operational pause: ERCOT "intends to temporarily pause its transitional Batch Zero interconnection study of loads that are 75 MW or more (Large Loads), including data centers and virtual currency mining facilities, until the verification and audit are complete," and will request community-impact information "from all data centers and virtual currency mining facilities of 25 MW or more that have not yet energized" (PUCT Order, Project 59142 Item 53). That is an energization/classification pause plus a broader 25 MW community-impact sweep, not a change to what on-site generation qualifies as.

Did it widen or narrow the BTM exemption?

Neither, as a rule change. The August 20 instrument is a good-cause exception to deadlines, not an amendment of WLPUN, PURA § 39.169 net-metering, or any "BTM exemption" text (PUCT Order, Project 59142 Item 53; M-A080326-02).

What did tighten is the grid-interconnection process around Large Loads that still need ERCOT:

  • Abbott's August 3, 2026 letter directs PUCT and ERCOT to finish a "comprehensive verification and audit of all data centers advancing through ERCOT's interconnection process before any additional data centers are approved to move forward," and to deny noncompliant projects (Governor Abbott letter PDF, 2026-08-03; Governor's Office press release, 2026-08-03).
  • The letter cites approximately 474 GW of interconnection requests, "more than five times Texas' record peak," with "approximately 90 percent" described as data centers. It asks, among other things, "the extent to which data centers are providing their own power or depending on the ERCOT grid," including "any effort and progress to construct or procure on-site electric generation, including generation type" (Abbott letter PDF). The fetched letter does not say on-site generation "may not be subject" to the audit, and does not state an "1800+ projects" count.
  • ERCOT's Large Load Integration page, as of this fetch, already has a live Batch Zero Verification section (RFI user guide dated September 11, 2026; ILLE/TSP attestations dated early September) plus the unchanged WLPUN Form X path (ERCOT Large Load Integration).

WLPUN/BYOG itself predates August 20 and was not rewritten that day. ERCOT market notice M-B062326-02 (June 26, 2026) tells studied-load WLPUN applicants that each new co-located generator must have a complete Full Interconnection Study application accepted in RIOO by July 10, 2026 — "behind-the-meter" is not an ERCOT bypass (M-B062326-02). The Large Load page still hosts "Section 23 Form X: Withdrawal-Limited Private Use Network Designation" with the same July 10 / July 24 submission chain (ERCOT Large Load Integration).

On eligible on-site tech: the first-party ERCOT HTML fetched in this pass (market notices + Large Load page) names WLPUN/BYOG and Form X but does not list fuel cells, gas turbines, or solar. Counsel commentary dated August 11, 2026 — written before the meeting — says "ERCOT specifically identifies natural-gas turbines and solar generation as potential on-site resources" and repeats that new WLPUN generators remain subject to generation-interconnection requirements (Beatty & Wozniak, 2026-08-11). That is a secondary restatement of ERCOT's June "Trending Topics" PDFs, which this skill did not fetch (ercot.com PDFs are off the *.gov whitelist). No first-party page fetched here adds fuel cells to, or removes them from, the WLPUN example list.

Net: August 20 did not widen a BTM exemption. Practically it narrowed the near-term path for grid-dependent Large Loads (pause + audit + 25 MW community-impact collection) while leaving WLPUN's legal shape intact and still requiring FIS for new on-site gens.

GE Vernova 2026 aero / LM2500 datacenter unit count

GE Vernova's July 22, 2026 Q2 release, as filed on EDGAR, states Power-segment Q2 orders of $16.7 billion, "with 52 heavy-duty units, including 15 HA turbines, and 61 aeroderivative turbines" (SEC exhibit gevpressrelease2q26.htm). The same exhibit says Gas Power signed 20 GW of new gas equipment contracts in the quarter, moved backlog from 100 to 116 GW under contract / slot reservations, and that Electrification data-center orders reached "over $5 billion year-to-date, more than double our 2025 total" (same SEC exhibit; company HTML reprint GE Vernova Q2 2026 press release).

Those 61 aeros are not labeled LM2500, not labeled datacenter, and not given a 2026 year-to-date datacenter unit total. The company HTML release likewise has no LM2500 or datacenter-turbine unit count (GE Vernova Q2 2026 press release).

The last first-party LM2500XPRESS × datacenter unit count retrieved in this pass is still the July 22, 2025 Crusoe announcement: 29 LM2500XPRESS packages (10 booked December 2024 + 19 booked Q2 2025), "expected to provide nearly 1GW," ~35 MW per unit (GE Vernova–Crusoe release, 2025-07-22). No 2026 first-party GEV LM2500 (or aero) datacenter unit count was found.

Bloom named MW past Oracle 1.2 / 2.8 GW

Bloom's April 13, 2026 IR release remains the first-party Oracle figure: initial 1.2 GW contracted, deploying in the U.S. "into next year"; MSA "supports up to 2.8 GW"; warrant issued April 9, 2026 on October 30, 2025 terms (Bloom IR, 2026-04-13).

Bloom's July 28, 2026 Q2 release reports $1.065 billion quarterly revenue, product revenue $935.4 million, and raised FY2026 revenue guidance of $3.9–$4.2 billion. KR Sridhar says "all the major US hyperscalers and over a dozen US neoclouds, AI labs, and colocation data center operators have validated and approved our power solutions" and calls Bloom "a standard for AI onsite power." The release names no new GW/MW customer increment — no Google, no AEP, no Oracle add-on (Bloom IR Q2 2026, 2026-07-28).

The only other first-party named SOFC MW retrieved here is pre-Oracle: AEP's January 4/8, 2026 8-K says a subsidiary had a November 2024 agreement for 100 MW plus an option for an additional 900 MW of solid-oxide fuel cells, and on January 4, 2026 executed "an unconditional purchase agreement to acquire a substantial portion of its option" for a Cheyenne, Wyoming fuel-cell facility at "approximately $2.65 billion," plus a 20-year offtake with an unnamed "high investment grade third party customer," conditions expected by Q2 2026 (AEP 8-K, 2026-01-04 event). That 8-K does not name Bloom, Google, or a closed MW figure for the exercised "substantial portion."

No Bloom-named MW increment past the April 2026 Oracle 1.2 / 2.8 GW pair was found in first-party Bloom IR or in the AEP 8-K. Third-party September 2026 Wyoming campus coverage (Google / Project Tembo / fuel cells) was not successfully fetched this pass (wyomingnews.com timed out); SemiAnalysis's "900 MW Bloom / Google" line was search-visible only and is not treated as a fetched, Bloom-named increment.

Encyclopedic anchors (not load-bearing for 2026 outcomes)

Grokipedia's Electric_Reliability_Council_of_Texas and public_utility_commission_of_texas pages define ERCOT as the ISO for ~90% of Texas load under PUCT oversight and recount the 2021 Uri reforms; they do not cover Batch Zero, PGRR 145, or the August 2026 audit (Grokipedia ERCOT; Grokipedia PUCT). Bloom_Energy is a company primer (SOFC, Sridhar, 2018 IPO) and is stale relative to the 2026 Oracle MSA (Grokipedia Bloom Energy).

Contradictions and open questions

  • BTM "exemption" vs. process exception. Market and prior vault language framed August 20 as a "good-cause-exemption / Batch Zero meeting for behind-the-meter datacenter generation." First-party documents frame it as good-cause exceptions to Planning Guide deadlines during an audit (PUCT Order). Treating the order as a BTM-rule widening is not supported.
  • On-site generation "may not be subject." Not in the fetched Abbott letter or press release. The letter asks for on-site generation type and progress as audit fields; it does not carve BTM projects out of the verification (Abbott letter PDF).
  • "1800+ projects." Not stated in the fetched Abbott letter or press release (those use 474 GW / ~90% data centers). Left unconfirmed this pass.
  • WLPUN tech list (gas/solar vs. fuel cells). Secondary counsel attributes gas-turbine and solar examples to ERCOT (Beatty & Wozniak); first-party HTML fetched here is silent. ERCOT June "Trending Topics" PDFs were not fetched (off *.gov PDF whitelist).
  • Post-August-20 implementation detail. ERCOT Large Load page now shows September 2026 verification RFIs (Large Load Integration). Search indexed a September 11, 2026 ERCOT "Batch Zero Update" PDF (provisional classifications September 3; reports due December 10) that was not fetched (ercot.com PDF).
  • GEV 61 aeros ≠ datacenter LM2500s. First-party Q2 exhibit gives 61 aeros with no end-use or model split (SEC exhibit). A 2026 DC-specific unit count remains open.
  • Bloom / Google 900 MW. AEP 8-K is a pre-Oracle "substantial portion" of a 900 MW option for unnamed offtaker (AEP 8-K). Bloom Q2 does not name it. SemiAnalysis / local Wyoming pages were not successfully fetched as citable first-party Bloom confirmation.

Provenance

Rounds run: 3 (full)

Sub-questions by round:

Round 1 (broad survey):

  1. What did the PUCT actually grant on 2026-08-20 regarding Batch Zero / good-cause exceptions?
  2. What is the first-party WLPUN / BTM / Large Load rule text still in force after that meeting?
  3. What 2026 first-party GEV LM2500 or aero datacenter unit counts exist?
  4. Has Bloom named any MW increment past Oracle 1.2 GW contracted / 2.8 GW MSA?
  5. What did ERCOT/WLPUN materials name as eligible on-site generation?

Round 2 (drill-down):

  1. Did any post-August-20 PUCT/ERCOT filing change WLPUN or BTM eligibility? — targeting the widen-vs-narrow gap
  2. Can GEV's 61-aero / DC-order figures be pinned to an SEC HTML exhibit (not an off-whitelist IR PDF)? — targeting first-party unit-count integrity
  3. Is there a first-party Bloom or counterparty 8-K naming MW beyond Oracle? — targeting the Google/AEP 900 MW claim
  4. What does Abbott's August 3 directive actually say about on-site generation? — targeting the "may not be subject" / 474 GW framing

Round 3 (resolve remaining uncertainty):

  1. Fetch the Abbott letter PDF itself — targeting missing "1800+" / "may not be subject" language
  2. Fetch pre-meeting WLPUN counsel + ERCOT WLPUN HTML — targeting the gas/solar-not-fuel-cell example list
  3. Fetch local/Google Wyoming coverage for a named 900 MW Bloom increment — targeting Bloom-past-Oracle (fetch failed)
  4. Fetch the 2025 Crusoe LM2500XPRESS release as the last first-party DC unit baseline — targeting the 2026-count negative

Anchor source (Grokipedia, fetched before round 1):

URLs fetched (13 successful, 2 failed):

Round 1:

Round 2:

Round 3:

Not fetched (rule / reliability):

  • ERCOT Batch Zero PDFs on ercot.com (Aug 20 presentation; Sep 11 update; June Trending Topics) — PDF host is not *.gov.
  • GEV IR PDFs on gevernova.com — same; SEC HTML used instead.
  • www.whitecase.com — Persistent-failure in SOURCE_RELIABILITY.md; skipped.
  • www.reuters.com / bbc.com — Hard-blocked; skipped.
  • SemiAnalysis newsletter — search-visible only; not fetched; not cited as Bloom-named MW.

Tools used: WebSearch, WebFetch, _lib/grokipedia.py (skill-equivalent helper). X pass not enabled. Generated: 2026-09-19 20:45 UTC SOURCE_RELIABILITY applied: preferred puc.texas.gov / interchange.puc.texas.gov, ercot.com HTML, gov.texas.gov, sec.gov, investor.bloomenergy.com, www.gevernova.com HTML; skipped Hard-blocked; PDFs only from *.gov.

Referenced by
Questions
Grid headroom turns negative → datacenters self-generate behind-the-meter → fuel-cell + gas-turbine OEMs (BE, GEV) capture a ~50GW/yr equipment TAM distinct from the genset bridge-power chain