Power-cycle thesis v1.1 — Druckenmiller doctrine + expanded basket
TL;DR
Backtest-v2 (2026-05-18) tested the v1 spec across 4 analog cycles + 1 live cycle (2024-AI-power). Two conclusions:
- Mechanical -8% per-trade stop is the alpha-killer. 6 of 10 stop-fires in the analog cycles hit names that became long-term winners. Same finding as memory v1.1. KILL the price-based stops.
- Basket expansion materially improves return. V2-no-mechanical with TLN/CEG/CCJ + VST/GEV/KTOS returned +276% on the live cycle vs +151% for the original-basket variant — a +125pp improvement. GEV alone was +730% (single-name outlier; flag for size discipline).
v1.1 = V2-no-mechanical + expanded basket. Same Druckenmiller shape as memory v1.1 (let winners run, exit only on thesis break), but applied to a cycle that DOES have a tradeable signal once the mechanical-rule layer is removed.
How power differs from memory (the honest read)
Memory v1.1 was archived because smart-money-mirror already covered the exposure (SNDK + INTC via 13F-following) — no separate thesis needed. Power is different:
- Smart-money signal is thinner. Only 3 of 7 tracked managers hold power names; the 2+ threshold for mirror inclusion was barely cleared by VST (Tepper + Druckenmiller). The mirror would NOT have caught GEV (only Tiger holds it) or KTOS (no smart-money holds it). So a dedicated power thesis adds names the mirror misses.
- Cycle phase is genuinely tradeable. Per the 2026-05-18 pressure-test, we're in phase 2 (capacity announcement velocity accelerating) heading into phase 3 (capacity online). Phase 2 → 3 historically returns +60-120pp in 12-18 months when caught with no mechanical exits. Memory's phase 2 already had smart-money saturation.
- Damage from mechanical exits is HALF as severe as memory. Memory v2 underperformed BH-univ by -27pp; power v2 by -14pp. Same direction, lower magnitude — because power has fewer false-signal events.
Spec changes vs v1
| Change | v1 | v1.1 |
|---|---|---|
| Mechanical -8% per-trade stop | Active | KILLED (6 of 10 stops in analogs hit winners) |
| Phase-marker mechanical exit | DRAM-spot-equivalent: PPA cadence reversal exit | KILLED; replaced by founder-review gate on 2-consecutive-quarter anchor reversal |
| Tranche-entry logic | 0.5R / 0.5R / 1R at -0/-5/-10% | Unchanged (value-additive) |
| Anchor-based position ADD | None | NEW — drawdowns >20% with bullish anchors = +0.25R add (Druckenmiller doctrine) |
| Max per name | 1R ($5k) | 1.0R for smart-money names, 0.75R for thesis-only names, 0.5R conviction-overlay |
| Bucket cap | 4R ($20k) | 4R ($20k) unchanged |
| Ticker universe | TLN + CEG + CCJ | TLN + CEG + CCJ + VST + GEV + KTOS |
| Profit trims | +50% trim 0.5R; +100% trim another | RETIRED — let winners run |
| Exit trigger | 2+ phase markers flip bearish for 2Q | Single anchor flip = founder review via /decisions/; 2+ HIGH anchors confirmed = bucket close |
R-sizing — weighted by smart-money corroboration + fundamental anchor
Total bucket: 4R = $20,000 paper.
| Name | R | $ | Why this weight |
|---|---|---|---|
| GEV (GE Vernova) | 1.0R | $5,000 | $163B backlog +67% YoY; Q1 electrification orders > all of 2025; only Western SMR with dirt moving (BWRX-300, 2029 target). Tiger Global holds. Strongest single-name signal in the basket. |
| TLN (Talen Energy) | 0.75R | $3,750 | First-mover Amazon nuclear-data-center deal (Susquehanna 960MW PPA). Already had its re-rate; ride the structural tailwind. |
| VST (Vistra) | 0.75R | $3,750 | 16.4x fwd P/E vs CEG 24.7x. New 2.2 GW Meta PPA. Tepper + Druckenmiller hold (2-manager smart-money — the only basket name passing the smart-money-mirror threshold). |
| KTOS (Kratos) | 0.75R | $3,750 | Conviction-overlay: owns Florida Turbine Technologies (Symphony engine designer for Boom Superpower). $1.25B Boom order book Dec 2025, $300M Series E. Priced as defense-drone; AI-power story unpriced. No smart-money signal — sized accordingly. |
| CEG (Constellation) | 0.5R | $2,500 | Flagship nuclear pure-play but premium-priced at 24.7x P/E. Half-weight of VST given cheaper alternative for similar exposure. |
| CCJ (Cameco) | 0.25R | $1,250 | Upstream uranium fuel-cycle pull-through. Zero smart-money corroboration; weakest fundamental link in the basket. Smallest position. |
Total: 4.0R = $20,000
Why not equal-weight: equal-weight ignores the smart-money signal. The weighted allocation tilts toward names with multiple convergent signals (smart-money + fundamental anchor) and underweights names with thinner signals (CCJ, where only the fuel-cycle pull-through argument exists).
Why GEV at 1R (highest single-name conviction): strongest convergent signal — backlog growth + Q1 order velocity + Tiger Global hold + only Western SMR with construction underway. Backtest's single-name +730% was on GEV; size discipline says don't chase that, but ALSO don't underweight the name with the strongest forward-looking signal.
Backtest evidence (honest)
Full report: [[../backtests/2026-05-18-power-cycle-v1-v2-honest-rerun.md]]
Headline numbers:
- V2-expanded / live cycle (2024-AI-power): +276% (beats BH-expanded by +18pp, beats BH-original by +107pp, beats BH-XLU by +227pp)
- V2 / original basket / live cycle: +151% (loses to BH-original by -18pp — buy-and-hold won the original basket)
- V1 mechanical / live cycle: +151% (identical to V2 — no exit fired in the live window because the AI-power cycle had no -8% drawdowns)
- All-cycle aggregate (4 analog + 1 live), V2 original basket: mean +48% with 95% CI [-10%, +107%]. V2 underperforms BH-univ by -14pp on aggregate.
- Leave-one-out (drop 2024-live): V2 collapses to +23% — same AI-supercycle dependence as memory v1.1.
Caveats (do not skip):
- N=1 for the actual thesis test. The 2024-AI-power cycle is the only cycle where this specific thesis (hyperscaler-driven structural power demand) applies. Analog cycles (1999-dotcom, 2007-commodity, 2010-shale, 2017-Texas-energy) are robustness checks, not primary evidence.
- GEV's +730% is a single-name outlier on the live cycle. Could be largely backlog-momentum that doesn't repeat. Sizing at 1R (not above) acknowledges this — strong conviction with size discipline.
- 1999-dotcom cycle excludes Enron / Calpine / Mirant (all bankrupt by 2003). Survivorship-biased UP. Live cycle has no survivorship gap (all 6 names tradeable today).
- Smart-money signal is fresh. VST + GEV positions are <12 months old per 13F filings. Could reverse on next quarter. The 4-anchor founder-review gate replaces 5 mechanical exit rules — manual review IS slower than autonomous exit.
Anchors to watch (founder-review triggers)
- ERCOT interconnection queue — currently 5.1x growth (68→346 GW). Reversal = phase 3→4 signal.
- Hyperscaler PPA cadence — Meta+Microsoft+Amazon+Google PPA announcement velocity. 2-consecutive-quarter slowdown = anchor flip.
- GEV order velocity — Q1 2026 electrification orders exceeded all of 2025. Q2 deceleration = single-name flag.
- Smart-money 13F deltas — VST or GEV exited by 2+ managers = bucket close trigger.
- PJM/CAISO power price normalization — peak-pricing reverts to pre-2023 levels = phase 4 confirmation.
What founder picks
See /decisions/2026-05-18-paper-trade-power-v1.1-go.html — 4 options (APPROVE / REVISE / ARCHIVE / DEFER) with click-back.
Related
- [[2026-05-17-power-cycle-v1.md]] — superseded
- [[../backtests/2026-05-18-power-cycle-v1-v2-honest-rerun.md]] — full backtest report
- [[../anchors/power-layer/2026-05-18-pressure-test-and-winners-survey.md]] — pressure-test that surfaced VST/GEV/KTOS
- [[../theses/2026-05-18-memory-cycle-v1.1.md]] — parallel thesis, archived in favor of smart-money-mirror coverage
- [[../strategies/2026-05-18-smart-money-mirror-v1.md]] — the auto-following strategy that covers MU-adjacent exposure
- [[../anchors/smart-money/2026-05-17-aggregate-2yr-backfill-summary.md]] — smart-money 13F backfill
- [[../../../06-reference/2026-05-18-stratechery-data-center-discontent.md]] — demand-side NIMBY risk update (Hill County TX moratorium, Gallup 71%, UBI-via-DC proposal)