Power-Cycle Thesis - Hype-vs-Structural Pressure Test + Winners Survey

Why this is in the vault

Founder asked yesterday whether the power-cycle v1 thesis ([[2026-05-17-power-cycle-v1]]) is just social-media beta or whether specific names have structural advantages worth holding for outsized returns. Also asked to verify a specific case study (Boom Supersonic's Symphony engine pivoting to mobile power generation) and survey winners by tier beyond the locked TLN/CEG/CCJ basket. Output is reference for the v1.1 thesis update and for the 0.5R basket-addition proposals at the bottom.

Cap budget: ~45 min, ~$5 LLM spend, ~8 web queries. Spent: 6 queries, ~$2.50.

Bottom line

1. Hype-vs-structural: 70% structural, 30% premium-already-paid. The bottleneck is real (interconnection queue +300% YoY in ERCOT, GE Vernova electrification booked $2.4B of data-center orders in ONE quarter vs. all of last year). But CEG at 24.7x forward P/E vs. VST at 16.4x means the obvious nuclear name has had its re-rate, the second-derivative names (transmission, fuel, mobile gen) have not.
2. Boom Supersonic claim VERIFIED and CONCRETE. Boom pivoted to Superpower (42MW natural-gas turbines from the Symphony core); Crusoe ordered 29 units; $1.25B order book; $300M Series E closed Dec 2025. Boom is private — playable via Kratos (KTOS) whose Florida Turbine Technologies subsidiary is Symphony's engine designer.
3. Top 3 winners NOT in current basket: GE Vernova (GEV), Quanta Services (PWR), Kratos (KTOS). GEV has 13F validation from Tiger; PWR sits one layer back (transmission) where the bottleneck physically lives; KTOS is the asymmetric mobile-power read on a verified non-obvious story.

Hype vs structural - honest read

What holds up to current data

The interconnection queue is real and getting worse. ERCOT's large-load queue jumped ~300% YoY through end-2025; total at 233-346 GW depending on classification (Latitude Media, Utility Dive). Data centers = ~77% of those requests. ERCOT publicly said summer 2026 peak demand lands between 90.5-98 GW, vs. ~85 GW prior cycle.

The PPA cadence is accelerating, not slowing. Verified live as of May 2026:

• Amazon-Talen: 1,920 MW through 2042 (expanded June 2025) - enkiai
• Microsoft-Constellation Crane: 835 MW, 20-year, $1B DOE loan
• NEW: Meta-Vistra: 2.2 GW, multi-Ohio plants - enkiai
• NEW: Meta-Terrapower + Meta-Oklo: combined 4.0 GW of advanced reactor offtake - same source
• Meta total nuclear commitment: 3.2 GW operating + 4.0 GW advanced = 7.2 GW

This is a structural acceleration, not a peak. The Meta-Vistra/Terrapower/Oklo combo is the strongest signal: the customer is now funding development (Meta as anchor) not just buying off the shelf.

Picks-and-shovels backlog data is the cleanest "real" signal. From Q1 2026 prints:

• Quanta Services (PWR): $48.5B record backlog (+22% YoY revenue); 15-20% adjusted EPS CAGR through 2030 PRNewswire
• GE Vernova (GEV): $163B total backlog; Electrification segment booked $2.4B in data-center orders in Q1 2026 - more than all of last year combined; raised 2026 guidance GEV release
• Eaton (ETN): Record Electrical Americas Q4 sales $3.51B (+21%)

This is where the data is most concrete and the valuations most reasonable. The IPPs trade on a thesis; the equipment makers trade on a backlog.

SMR reality vs. narrative:

• NuScale (SMR): NRC approved the 77 MWe design. TVA + ENTRA1 announced a 6-GW deployment program. Deployment by 2030. - DOE
• GE Vernova/Hitachi BWRX-300: Construction in progress at Darlington (Ontario), completion target late 2029. TVA filed first US BWRX-300 construction permit; NRC review completes ~Dec 2026. Furthest along of all Western SMRs.
• X-energy: NRC safety evaluation expected Nov 2026; first commercial deployment maybe end of 2026.
• Oklo: No NRC design approval; liquid-sodium cooling is novel; 75 MWe design announced; speculative.
• Centrus (LEU): Already producing 900 kg/yr HALEU at Piketon Ohio; DOE contract extended through June 2026 (World Nuclear News). Only Western producer of the fuel that ALL the advanced-reactor designs require.

Honest read on SMR: only GEV/Hitachi BWRX-300 has dirt moving. NuScale is paper-approved. Oklo/X-energy/SMR (NuScale) are all still in license-review or design-approval phases. The "SMR = 2027" narrative is wrong; the realistic earliest commercial date is 2029-2030.

EIA macro numbers are revising up but slower than the loudest claims. EIA April 2026 outlook: 2026 US electricity consumption 4,250 BkWh, +1.3% YoY. Commercial sector +2.2% in 2026, +5.3% in 2027. ERCOT specifically expected to lead. This is real growth but the headline-grabbing "4x by 2030" claims aren't EIA-supported - those are AI-industry capex projections back-of-enveloped to load. The structural read: load growth is back after 15 years flat, but it's 1-3% not 10-20%.

Where valuation has run ahead of fundamentals

CEG at 24.7x forward P/E (44.6x trailing, vs. industry median 14.9x) is paying for the Crane Microsoft deal at full price - per MEXC analysis, that premium "only sustains if 25%+ EPS growth holds." Any miss against that bar gets re-rated. VST at 16.4x is materially less risky on multiple compression.

TLN has fully priced the Amazon deal. From the thesis filing, TLN is up enormously over 12 months. The asymmetric trade was 2025; 2026 entry is paying for delivered execution, not optionality.

Bloom Energy (BE): Order backlog has more than doubled but the founder's thesis filing already flagged BE at +1,647% 12-month — extended. Skip per the parent thesis judgment.

SMR pure-plays (NuScale, Oklo) trade on 2030+ optionality. Both are real businesses with real customers (TVA, DoD, Air Force) but earnings are years out. Speculation, not investment.

Verdict on hype vs. structural

The bottleneck is structural and verified. The "AI-energy" trade as such is partially priced. The asymmetric edge has shifted FROM the obvious IPPs (TLN, CEG) TO:

1. Picks-and-shovels grid + electrical equipment (PWR, GEV, ETN) - backlog growth is real and not fully priced
2. The second IPP (VST) - cheaper than CEG, similar nuclear leverage, Meta anchor
3. Asymmetric mobile-power read (KTOS via Boom)
4. Fuel cycle pure-plays (LEU/Centrus) - only Western HALEU producer

Boom Supersonic mobile-power case - VERIFIED

Founder's claim verified in detail. The story is real, the engineering pivot is real, the order book is real, and there is a tradeable public-market expression.

What's actually happening (from multiple primary sources):

• Boom Supersonic announced Dec 2025 a pivot to develop "Superpower" — a 42 MW natural-gas burning electricity-generating turbine — using nearly the same core as the Symphony aircraft engine. Boom press release
• Closed $300M Series E specifically to fund Superpower
• Crusoe ordered 29 Superpower turbines for AI data centers
• Total Superpower order book: $1.25B as of Dec 2025
• Strategic logic per Boom CEO Blake Scholl: power generation revenue funds Overture aircraft development; revenue arrives faster than supersonic flight regulatory approval (which is the founder's exact framing — verified accurate). Flightglobal
• Symphony engine assembly begins May/June 2026 ("within weeks" per Aerospace Global News)

Public-market expression:

Boom is private (Series E). The cleanest public play is Kratos Defense & Security (KTOS). Kratos' Florida Turbine Technologies (FTT) subsidiary is Symphony's engine design partner — the FTT team comes from F-22/F-35 engine programs. Kratos stock specifically rose on the Boom Superpower announcement (Investing.com).

Caveat: KTOS revenue from FTT/Boom is a small share of the consolidated business (most of Kratos is defense drones — Valkyrie, hypersonic). The Boom exposure is real but partial — KTOS is not a pure play on Superpower. The wider Kratos defense franchise is also a beneficiary of mobile-power demand via Kratos Power Systems (separate from FTT).

Flag as watchlist candidate. Asymmetric: real story, partial exposure, currently underpriced because the market is treating KTOS as a defense drone play not an AI-power play.

Winners survey by tier

Tier 1 - Front-running PPA exposure (IPPs)

Verdict: VST is the missing third IPP. Cheaper valuation, comparable hyperscaler beta, fresh Meta anchor. Add to basket at 0.5R.

Tier 2 - Picks-and-shovels grid + electrical

Verdict: PWR is the underappreciated transmission play. GEV is partially priced but Tiger holds it for a reason — the data-center electrification flow is one-quarter of last year's annual rate. ETN is the boring add.

Tier 3 - Upstream fuel cycle

Verdict: LEU is the asymmetric fuel-cycle add for advanced reactor exposure - if SMR/advanced reactors ever ship, ALL of them need HALEU and Centrus is the only domestic source. Speculative but cheap at sub-$5B market cap.

Tier 4 - Mobile / distributed / backup generation

Verdict: GNRC is the cleanest pure-play (Stargate exposure verified). KTOS is the asymmetric optionality play via verified Boom story. Skip BE (extended).

Tier 5 - Speculative SMR

Verdict on SMR pure-plays: GEV-via-BWRX-300 is the only one with dirt moving. SMR (NuScale) is the second-best paper position. Skip OKLO. The HALEU layer (LEU) is the more interesting indirect bet.

Tier 6 - Adjacencies

Verdict: Not basket adds at v1.1. Watchlist for v2 expansion.

Smart-money positioning cross-ref

From [[2026-05-17-aggregate-2yr-backfill-summary]] - the only 13F-tracked names with explicit power-thesis exposure across 7 surveyed managers (8 quarters each):

TLN, CEG, CCJ, NuScale, Oklo, BE, GNRC, KTOS, LEU, PWR, ETN, BWXT, CMI, CAT — none currently held by tracked smart-money set. This is two-edged: (a) the obvious nuclear names lack smart-money validation, which weakens conviction; (b) the absence is also opportunity — these aren't crowded smart-money trades.

The strongest smart-money triangulation in our basket would be VST (2 of 7 managers) and GEV (1 of 7). Both not currently in v1 basket. This is the strongest argument for the basket additions below.

Rising-tide vs differentiation verdict

Precedents:

• Dot-com 1999: rising tide for ~18 months, then differentiation brutally separated AMZN/EBAY/MSFT (survivors) from pets.com/webvan (terminal). Sector beta dominated until it didn't, then specific names mattered for 2+ years.
• Telecom 2000: sector beta dominated UP, then dominated DOWN — almost nothing differentiated through the bust. Sector capex front-runners (Lucent, Nortel, Worldcom) all collapsed. Picks-and-shovels (Corning fiber) round-tripped 95%. Bad analog for current bull conviction.
• Solar 2007: total sector wipeout from oversupply (China subsidies); the "winners" (First Solar, SunPower) round-tripped 80-90%. Differentiation only mattered AFTER the bust — survivors capitalized the recovery.

Current power-cycle read:

Stage 1 (now through ~2027): Rising tide dominates. Anyone with hyperscaler-PPA exposure or backlog gets re-rated together. Hard to lose if you own the basket.

Stage 2 (2028-2030): Differentiation kicks in when supply finally arrives. The losers will be the IPPs whose PPAs are mid-curve (signed at 2025/2026 peak pricing) when the queue clears. The winners will be the equipment makers whose backlog is locked in regardless (PWR, GEV, ETN) and the upstream fuel-cycle players whose volume scales independent of which IPP wins (CCJ, LEU).

Implication for basket construction: Tilt toward picks-and-shovels and upstream fuel over pure IPPs as we add. The IPP exposure is sufficient at 3.5R (TLN 2.0 + CEG 1.5); adding more nuclear IPPs increases correlation without adding diversification.

Recommended basket additions to power thesis v1.1

Three additions at 0.5R each (total +1.5R brings power bucket to 5.5R; founder may want to keep the 4R cap and trim TLN/CEG instead — flag for /decisions/).

Alternative if founder wants to hold 4R cap: trim CEG from 1.5R to 1.0R (valuation re-rate risk), trim CCJ from 0.5R to 0.25R, free 0.75R; add VST 0.5R + GEV 0.25R. Skip KTOS at the formal basket but keep on watchlist.

Honest sizing call: I'd argue for the 5.5R version. The bear case is mostly about pricing not structure; basket-wide drawdown is what kills R-unit theses, and adding GEV + VST is genuine diversification (different point in supply chain, different smart-money validation), not just more nuclear beta. KTOS at 0.5R is the speculative leg; comfortable cutting it from formal basket if founder prefers cleaner exposure.

Caveats

• Smart-money lag. 13F filings reflect end-Q1 2026 positions filed mid-May. We don't see Q2 movements until mid-Aug. If Tepper or Druckenmiller trimmed VST after Meta deal closed, we won't know for 3 months.
• No private-market signal on Crusoe. Crusoe's 29-turbine Boom order is a strong demand signal but Crusoe is private; we can't cross-verify via 13F.
• EIA numbers are conservative. The 1.3% 2026 / 3.1% 2027 EIA forecasts assume current data-center load materializes on schedule. ERCOT itself flagged in Dec 2025 ("ERCOT Warns Texas AI Power Boom May Not Materialize") that the 233 GW queue may overstate actual builds.
• SMR timelines could slip further. Even GEV-BWRX-300 at 2029 is "estimated"; nuclear new-build slips historically.
• KTOS exposure is partial. Most of Kratos is defense drones not power; analytic risk of overweighting the Boom story.
• No alternative-data on PPA pipeline beyond announced deals. We can't see PPAs in negotiation; the cadence forecast assumes announced cadence continues, which it may or may not.

Mapping against Ray Data Co

The power-cycle thesis is a textbook instance of the founder's "AI infrastructure as the picks-and-shovels of the AI buildout" surface. The pressure-test sharpens it: where the thesis has been most contagious in retail/social commentary (TLN, CEG, CCJ pure-plays) is where the alpha has been most compressed. The non-obvious tier (transmission EPC, electrical equipment, fuel cycle, asymmetric mobile-power optionality via Boom) is where the founder's "instrument the bottleneck not the headline" reading actually lives. The v1.1 basket additions push the bucket in that direction.

This is also the cross-cycle precedent the founder wants to embed: the power-cycle thesis is the SAME shape as memory-cycle (supply lead-time longer than demand surge), but unlike memory, the bottleneck has a hard physical floor (nuclear construction = 8-12 years, not 18-24 months). That asymmetry is the structural alpha. The risk is paying for it twice through valuation.

Related

• [[2026-05-17-power-cycle-v1]] - parent thesis being pressure-tested
• [[2026-05-17-aggregate-2yr-backfill-summary]] - smart-money 13F cross-ref source
• [[2026-05-17-memory-cycle-v1]] - sibling thesis (same operating shape)
• [[2026-05-18-memory-cycle-v1.1]] - just-updated sibling (mechanical-exits-kill-alpha learning)
• [[2026-05-17-power-anchors-backfill-notes]] - prior anchor-data session
• [[06-reference/2026-05-12-diamandis-innermost-loop-ai-infrastructure-thesis]] - parent reference
• [[06-reference/2026-04-30-not-boring-scarce-assets-abundance-driven-scarcity]] - Packy McCormick framework