Autoresearch: does negative grid headroom accrue to Bloom and GE Vernova BTM?

Generated by /autoresearch on 2026-08-17. Synthesized across 3 rounds from 10 successful web fetches (3 timeouts), anchored by Grokipedia Bloom_Energy. See Provenance. Treat as raw material — review before promoting into a project or thread. Context: vault/projects/stock-market

Priors capture skipped — unattended weekday shift; user instructed skip.

X pass enabled. 0 high-signal posts surfaced (0 via direct fetch, 0 via search snippets). Four site:x.com / site:twitter.com discovery queries returned no permalinks. X MCP needsAuth in this environment. See Provenance.

Summary

The June 2026 question asked whether collapsing US grid headroom routes a primary on-site generation TAM to Bloom Energy (BE, fuel cells) and GE Vernova (GEV, LM2500 turbines), distinct from the already-graduated genset bridge (CAT/CMI), with Vistra (VST) as a nuclear-co-location adjacent. This pass does not re-litigate the genset chain or the parent cascade's power-is-binding step. It tests the unverified order-flow-to-ticker gap.

Bloom's gap closes on first-party paper. Q2 2026 printed record revenue of $1.065B (+165.5% YoY) on product revenue of $935.4M (+215%), and management raised FY26 revenue to $3.9–$4.2B. CEO KR Sridhar called Bloom "a standard for AI onsite power": all major US hyperscalers plus "over a dozen" neoclouds, AI labs, and colos have validated the platform. Oracle's April 13 master agreement is up to 2.8 GW, with 1.2 GW contracted and deploying. Brookfield's financing shelf went $5B → $25B on June 30. That is named, dated, datacenter-attributed order flow — the citation the question's "What to watch" asked for.

GE Vernova's gap narrows but does not close as an LM2500-BTM monopoly. Gas equipment under contract rose 100 → 116 GW in Q2, with a year-end target of ≥125 GW. Management says ~20% of those gigawatts are data-center, 80% traditional utility/IPP. The named aero BTM order is still the July 2025 Crusoe 29× LM2500XPRESS (~1 GW) "power plants alongside AI datacenters." Q2's much-cited >$5B H1 data-center orders sit in Electrification (grid kit), not Gas Power. Aero is described as a bridge until HA heavy-duty commissions 18 months later. Siemens Energy's DC share is ~28% of 87 GW — higher than GEV's. The Aug-11 competitive leak (30+ OEMs, secondary market) is not refuted.

Vistra is not this equipment TAM. Comanche Peak is a 20-year 1.2 GW PPA with Amazon, first energy 2027, full ramp 2032. Management supports the Texas queue audit and wants it culled. That is nuclear-adjacent firm power, not Bloom/GEV kit.

The Aug-20 ERCOT watch item has not fired. This file is dated 2026-08-17. Batch Zero classifications remain paused; the PUCT good-cause meeting is still three days out. The audit does ask whether projects provide their own power. WLPUN/"bring your own generation" is the documented BTM path — and it is not a full ERCOT bypass.

Findings

Theme 1 — Bloom: the order-flow-to-ticker link is now first-party

Bloom's July 28 release is the primary (Bloom IR Q2):

YoY: revenue +165.5%, product +215.4%. FY26 guide raised to revenue $3.9–$4.2B, non-GAAP operating income $800–$900M, non-GAAP EPS $2.55–$2.85 (same release).

Sridhar, in the release: "customers who traditionally defaulted to combustion technologies are now proactively choosing Bloom as a superior power solution. Today, 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 for their AI factories. Bloom is now a standard for AI onsite power." That is the combustion-to-fuel-cell substitution the question hypothesized, in the issuer's own words, and it is the intended distinction from the CAT/CMI genset bridge.

On the July 28 call (Roic AI transcript; WebFetch timed out, text from search dump — treat as reprint, not a second primary): Sridhar said Bloom became that standard "in less than a year" after announcing Oracle as the first direct hyperscaler and delivering a data-center system in 55 days. "We have booked major customers as of today who are not in our reported backlog at the end of last year for whom we will ship some of our systems this year." "These customers are now placing longer-term orders, leading to our backlog growing at a faster pace than revenue." CFO Simon Edwards: the backlog "spans multiple hyperscalers, neoclouds, colocation providers and commercial and industrial operators"; one or two customers can dominate a single quarter's revenue because of delivery timing, not because the backlog is a single name (Roic AI Q2 transcript).

The YE2025 backlog figure was ~$20B total / ~$6B product (product +140% YoY). Bloom only restates the dollar backlog once a year; Q2 did not print a new total (Motley Fool, 2026-08-05 — secondary restatement of the YE number; do not treat the $20B as a Q2 disclosure).

Named contracts that close the gap:

• Oracle, 2026-04-13. Master services agreement: Oracle "intends to procure up to 2.8 GW." "Initial 1.2 GW of capacity has been contracted, with deployment underway and continuing into next year." Last year Bloom delivered a fully operational system to Oracle in 55 days vs a 90-day schedule. OCI EVP Mahesh Thiagarajan: "By rapidly deploying Bloom's reliable, efficient fuel cell energy, we are quickly meeting the demands of our customers across the United States." (Bloom IR Oracle)
• Brookfield, 2026-06-30. Financing framework $5B → $25B (fivefold since October 2025) "to finance power projects for AI infrastructure." The release's own bullets call this "Bloom's islanded power solutions" and "rapidly deployable onsite power." (Bloom IR Brookfield)
• AEP, 2024-11-14. Supply agreement for up to 1 GW; initial 100 MW order; "co-located at customer sites"; "initial installation… will help meet the immediate power of AI data centers." (Bloom AEP) Older than this pass's window, but it is the utility-channel first-party that Grokipedia and later coverage keep citing.

Sridhar's on-call framing of why this is BTM rather than a wait-for-the-grid trade: "surplus grid capacity is now gone. New data center load now means new infrastructure… The faster, cheaper, more predictable path for that customer is islanded on-site power." He contrasts "legacy suppliers [who] celebrate backlog stretching to 2029" — "a 4-year backlog is not a trophy. It's a concession of constrained supply" (Roic AI Q2). That is the question's step 1–3 logic in the beneficiary's mouth.

Held honest. "Validated and approved" ≠ every hyperscaler has a disclosed GW order. Only Oracle's 1.2 GW contracted / 2.8 GW ceiling is a named hyperscaler MW figure in the fetched primaries. AEP's 1 GW is a utility procurement ceiling with 100 MW firm. Brookfield is a financing shelf, not a 25-gigawatt equipment order. Sridhar's "25 gigawatts of like deployments" visibility line on the call is pipeline color, not backlog.

Theme 2 — GE Vernova: DC share is real; the LM2500-BTM monopoly is not

GE Vernova's July 22 release (GEV IR):

• Orders $24.2B (+88% organic); book-to-bill "slightly more than two times" (call).
• Total backlog $176B (+$13B sequential); on track for $200B in 2027.
• Gas Power equipment backlog + slot reservations 100 → 116 GW; year-end target ≥125 GW.
• Shipped 3 GW of gas equipment in the quarter; signed 20 GW of orders + SRAs.
• Output path: 20 GW/yr by Q3 2026, 24 GW in 2028, actions toward 30 GW in 2030.
• Revenue $11.1B (+22% / +12% organic). FY26 revenue guide $45.5–$46.5B; FCF $11.5–$12.5B.

The data-center split, from the call (Roic AI reprint of GEV Q2 transcript; WebFetch timed out):

• Firm gas backlog 44 → 53 GW; SRAs 56 → 63 GW.
• "~100 customers in 26 different countries, of which approximately 80% are traditional customers and 20% are for data centers."
• Electrification: Q2 booked $2.7B of data-center orders, H1 >$5B, "more than double full year 2025." Equipment backlog $41B.
• Power orders more than doubled; Gas Power equipment orders ~4× YoY on HA and aeroderivative.
• Q2 shipped 29 gas turbines (+38% YoY), including 16 aeroderivatives.

Scott Strazik on aero vs heavy-duty (same call): "in a number of cases, the aero are complementary to the Heavy-Duty… customers in earlier periods leveraging the aero derivatives until the Heavy-Duty can ultimately be installed… commissioning of that aero derivative is reasonably quick. It could be six months, whereas the ultimate commissioning of the Heavy-Duty could take another 18 months at site… the aero derivatives are providing the bridge." That is GEV's own language. It puts LM2500 in a bridge-to-HA role — closer to the CAT/CMI "bridge power" category than to "always-on primary that replaces the grid," which is the question's claimed distinction.

Named BTM-style aero order. July 22, 2025: Crusoe, 29 LM2500XPRESS packages, "nearly 1 GW," "rapidly building and operating power plants alongside AI datacenters." 10 units booked December 2024, 19 booked June 2025. Five-minute black-start, 95% factory-assembled, ~35 MW per unit (GEV Crusoe). This is the cleanest first-party LM2500-to-datacenter citation in the fetched set. It is thirteen months old. No fetched 2026 primary restates a new LM2500 datacenter unit count.

Do not conflate the $5B. Strazik's release sentence is "data center orders reaching over $5 billion year-to-date" in Electrification. The call repeats it as Electrification segment DC orders. An analyst who added "$5B of orders… versus $13.5 billion in total" was talking Electrification H1, and Strazik confirmed those numbers (GEV Q2 transcript). Using $5B as "GEV turbine BTM TAM" would be a category error.

GEV's own scarcity hedge. On an earlier call, Strazik said "in many of the cases with these projects, the gas turbines are really not the gating item" versus a three-year EPC / permitting / fuel cycle (Natural Gas Intel — recap of the Q1-era 100 GW book). The wiki already flagged this as "a seller talking down its own scarcity." It remains the company's public line.

Theme 3 — The competitive leak is now two-sourced, not just SemiAnalysis

The Aug-11 question update treated SemiAnalysis's "GEV's are [5yr+ backlogged], but there are plenty of other options" as a single-source leak. This pass adds a second, independent OEM split:

• Siemens Energy: data-center commitments 24 GW of 87 GW (~28%) as of March; FY-end target 90–100 GW. CEO Christian Bruch: customers shuffle sites to wherever air permits clear faster — the fuel-cell "avoid combustion permitting" claim in the question is the other side of this sentence (Natural Gas Intel).
• Mitsubishi Power: 35 large-frame turbines booked in the FY ended March, +40% YoY; CEO Eisaku Ito called it record intake (same).
• GEV remains the largest book (116 GW vs Siemens ~60–87 GW depending on firm vs commitment definition) but not the highest data-center share.

East Daley (secondary): GEV 116 GW vs ~20 GW/yr capacity; more than half of the 116 is slot reservations, not firm orders; ~20% DC-associated; reservations now for 2031 (East Daley — not fetched; snippet only, do not load-bear).

The July 2025 Crusoe deal is evidence GEV does book named BTM aero. It is not evidence that GEV "dominates the on-site filings" to the exclusion of Siemens/Mitsubishi/second-tier/secondary-market — the claim the question still marks unverified. The SpaceX Southaven "no supplier named / GEV routed around" point from the Aug-11 SemiAnalysis update is not revisited here and is not contradicted by anything fetched.

Theme 4 — Texas: the policy driver is live; the Aug-20 read is still ahead

This file is dated 2026-08-17. The question's dated watch item is the 2026-08-20 PUCT open meeting.

What is already on paper:

• Abbott's Aug-3 directive: audit data centers in the ERCOT queue before further approvals; the letter asks whether projects depend on ERCOT or provide their own power, including progress toward on-site generation and the type proposed (Beatty & Wozniak, 2026-08-11).
• ERCOT Market Notice M-A080326-01 (Aug 3): will not issue Batch Zero Large Load classifications by Aug 7; will seek a good-cause exception at the Aug 20 PUCT meeting (ERCOT notice).
• Queue size cited: ~474 GW large-load requests, ~90% data centers (ERCOT July 29 Senate presentation, via Beatty).
• The documented BTM path is ERCOT's Withdrawal-Limited Private Use Network (WLPUN), "sometimes described as Bring Your Own Generation." It nets on-site generation against load behind an enforceable withdrawal limit. ERCOT names natural-gas turbines and solar as potential on-site resources. Fuel cells are not named in the fetched WLPUN write-up.
• Critical qualifier, same memo: "Behind-the-meter generation is not an ERCOT bypass: new generators proposed as part of a WLPUN remain subject to applicable generation-interconnection requirements."

Gibson Dunn (not fetched; search snippet only): ERCOT's Aug 10 good-cause filing is in; at the Aug 14 open meeting ERCOT said verification would take several months, less than nine. The April 9, 2027 study-results date has not yet been extended. Do not treat that as fetched.

Net for the chain. The Aug-10 wiki update already recorded the policy driver (grid connection unavailable, with an explicit off-grid carve-out). This pass adds: (1) the carve-out is operationalized as WLPUN/BYOG, which names gas turbines; (2) it is not a clean islanded exemption; (3) the Aug-20 meeting that would tell us whether the exemption widens or narrows has not happened.

Theme 5 — Vistra is a nuclear PPA, not a BTM equipment beneficiary

The question's step 5 ("VST adjacent via nuclear co-location… ⚠ unverified") used SemiAnalysis's "1,200 MW net-metering at Comanche Peak." The 2026 primary is a PPA, not a net-metering equipment order.

• Sept 29, 2025 8-K: 20-year PPA (extendable +20) for 1,200 MW of carbon-free power from Comanche Peak; delivery starts Q4 2027, ramps to full by 2032; incremental Adj. FCF before growth ~8–10% if the customer takes full capacity (Vistra 8-K — customer unnamed in the filing).
• Q2 2026 call / Utility Dive (2026-08-15 recap of the Aug 14 call): the customer is Amazon; $5B data-center project near the plant; CEO Jim Burke does not expect Comanche Peak to be hit by the Abbott audit; "this load… 2027, 2028 time frame"; "I'd like to see the queues culled"; ERCOT load-growth guide cut from 5–6% to 4–6%, with Burke saying a "significant component" of growth is not data centers (Utility Dive).
• Stacey Doré (same): "very pleased with FERC's co-location order that came out in June" — PJM/TOs must accommodate co-location via new transmission services. That is the front-of-meter / tariff path the wiki already documented for CEG, not a Bloom-style islanded kit sale.

Read: VST confirms that hyperscalers will pay for firm nuclear next to a campus. It does not confirm that the BTM scramble books incremental Vistra equipment or that Comanche Peak is behind-the-meter in the SemiAnalysis sense. Treating VST as a third name in the BE/GEV equipment TAM overstates the source.

Theme 6 — Grokipedia primer (anchor only)

Grokipedia: Bloom Energy (25,017 chars fetched) is the vocabulary primer: SOFC, no combustion, modular Energy Servers, BTM or front-of-meter, fuel-flexible (gas / biogas / hydrogen), claimed 90-day data-center deployments, Equinix >100 MW / 19 sites (2025), Oracle 2025, Brookfield $5B (Oct 2025), AEP 1 GW, "over 300 MW" installed at data centers in one passage vs "1.4 GW deployed globally as of 2024" in another. Also: never recorded annual profits in the encyclopedia's telling, 2020 revenue restatement, short-seller history. Treat as a primer. Load-bearing 2026 figures in this file come from the IR releases, not from Grokipedia. The encyclopedia does not yet reflect Q2 2026, the Oracle 2.8 GW MSA, or the Brookfield $25B expansion.

Contradictions and open questions

• BE "standard" vs named MW. First-party validation across "all major US hyperscalers" is a qualitative claim. The only named hyperscaler GW figure fetched is Oracle 1.2 / 2.8. A 10-Q split of product backlog by data-center vs C&I was not extracted (SEC 10-Q HTML was search-dumped at 206 KB; this pass did not mine it for a DC %). That would be the next tightening citation.
• GEV $5B DC orders are Electrification, not turbines. Anyone still citing that print as LM2500 BTM TAM is misreading the segment. The turbine DC claim that is first-party is the ~20% of 116 GW mix, plus Crusoe 29× LM2500XPRESS (2025).
• Aero-as-bridge vs "primary on-site." The question's claimed distinction from CAT/CMI is that BE/GEV are always-on primary, gensets are bridge/backup. GEV's CEO now describes aero as the bridge to HA. That leaks the distinction on the GEV leg. Bloom's islanded-primary claim is cleaner.
• Siemens DC share > GEV DC share. 28% of 87 GW vs 20% of 116 GW. GEV can still be the largest absolute DC turbine book (~23 GW if 20% × 116) and not "dominate filings." SemiAnalysis's 30+ OEM / secondary-market leak is unrefuted.
• WLPUN names turbines and solar, not fuel cells. Texas policy helps the category (own-power). It does not, in the fetched legal memos, name Bloom.
• Aug-20 has not happened. Do not update the exemption as widened or narrowed until the PUCT meeting files.
• VST "net-metering" vs PPA. Wiki step 5 should be corrected toward the 8-K/PPA language. Co-location appetite is real; it is a different chain (nuclear PPA / FERC tariff), already closer to nuclear-baseload-for-ai-data-centers than to this equipment TAM.
• X discovery returned zero permalinks. No primary-participant posts (Sridhar, Strazik, Crusoe, Oracle) were citable from X this pass.

Provenance

Rounds run: 3 of 3

Sub-questions by round:

Round 1 (broad survey):

1. Do Bloom's latest earnings / backlog disclosures attribute orders to datacenter BTM (not generic C&I)?
2. Do GEV's latest earnings disclose LM2500 / gas-turbine book-to-bill on datacenter / BTM accounts vs utility/IPP?
3. Has the ERCOT Batch Zero / Abbott BTM carve-out been decided (Aug-20 meeting)?
4. What is the current competitive set for BTM turbines (Siemens, Mitsubishi, secondary market)?
5. Is Vistra Comanche Peak / nuclear co-location confirmed as BTM-scramble revenue?

Round 2 (drill-down):

1. Official Oracle / Brookfield / AEP contracts — targeting the named-MW gap on BE.
2. Crusoe LM2500 primary — targeting the missing 2026 LM2500 unit citation.
3. Vistra Q2 / Utility Dive — targeting PPA vs net-metering.
4. Siemens/Mitsubishi DC share — targeting the competitive leak.

Round 3 (resolve remaining uncertainty):

1. Bloom official Brookfield $25B release — targeting a first-party financing-shelf citation rather than ESG Today.
2. Natural Gas Intel on GEV "turbines aren't gating" + Siemens 24/87 GW — targeting a second source for the competitive leak.

Anchor source (Grokipedia, fetched before round 1):

• Bloom Energy — 25,017 chars extracted — SOFC / BTM vocabulary, Equinix / Oracle 2025 / Brookfield $5B / AEP 1 GW primer. Does not include Q2 2026, Oracle 2.8 GW, or Brookfield $25B.

X sources (via x-fetch skill + _lib/x.py):

• 0 items surfaced (0 direct fetch, 0 via search snippets)
• Queries: Bloom Energy datacenter site:x.com OR site:twitter.com 2026; GE Vernova datacenter site:x.com OR site:twitter.com 2026 turbine; "behind the meter" OR BTM Bloom OR "GE Vernova" OR LM2500 site:x.com 2026; plus the first Bloom Grokipedia-adjacent pass.
• X MCP serverStatus needsAuth in this environment. Same outcome as the 2026-08-17 CSP / private-credit / Amkor / BESI layers.

URLs fetched (10 successful, 3 failed):

Round 1:

• Bloom Energy Q2 2026 results — official IR — record $1.065B, guide raise, "standard for AI onsite power"
• Beatty & Wozniak on Batch Zero / WLPUN — energy-law memo — audit asks own-power; WLPUN names turbines/solar; BTM ≠ ERCOT bypass
• GE Vernova Q2 2026 results — official IR — 116 GW, $176B backlog, Electrification DC >$5B H1
• [Failed: https://www.roic.ai/quote/BE/transcripts/2026-year/2-quarter] — timeout; search dump used (same method as the BESI layer)
• [Failed: https://www.roic.ai/quote/GEV:UN/transcripts/2026-year/2-quarter] — timeout; search dump used
• [Failed: https://www.utilitydive.com/news/ge-vernova-gas-turbine-backlog-climbs-to-116-gw/826039/] — timeout; figures corroborated by GEV IR + call dump

Round 2:

• Bloom / Oracle 2.8 GW MSA — official IR — 1.2 GW contracted and deploying
• Bloom / AEP 1 GW — official — Nov 2024 ceiling + 100 MW initial
• GEV / Crusoe 29× LM2500XPRESS — official — ~1 GW, July 2025, "power plants alongside" DCs
• Vistra supports Texas pause — trade press — Amazon 1.2 GW PPA, 2027 energize, queues-culled
• ESG Today Brookfield $25B — secondary; superseded by official IR in round 3

Round 3:

• Bloom / Brookfield $25B — official IR — financing shelf, islanded/onsite language
• Natural Gas Intel: turbines aren't gating — trade press — GEV 20% DC; Siemens 24/87 GW; Bruch air-permit shuffle

Not fetched (noted, not cited as primary): East Daley turbine-delay note (snippet only); Gibson Dunn Batch Zero memo (snippet only); CNBC Greenville plant tour (www.cnbc.com is Persistent-failure / 403); Bloom 10-Q HTML (search-dumped, not mined for a DC backlog %); ERCOT M-A080326-01 body was in search results and is cited via the Beatty restatement plus the notice URL.

Tools used: WebSearch, WebFetch, grokipedia-fetch (skill / _lib/grokipedia.py), x-fetch discovery via site:x.com WebSearch (X MCP needsAuth). Generated: 2026-08-17 14:20 UTC