/decisions · 2026-05-17 · investing · power-layer-v1

Deploy power layer v1 paper trade — go / revise / defer / archive?

Decision logged from the power-cycle thesis. Click an action to fire an iMessage back to the Ray session with the decision encoded.

Context

Parallel-v1 paper-trade thesis to the memory bucket. Same operating shape (tranche-accumulate / no per-trade stops / phase-marker exits / R-units), different fundamentals. Power is the Diamandis-named bottleneck layer — $156B of US data-center projects are blocked on power-infrastructure constraints, interconnection-queue wait times are 36–48 months in data-center growth zones, and hyperscalers are now contracting for decade-plus nuclear baseload at premium pricing (Talen-Amazon $18B / 17yr / 1.92GW; Constellation-Microsoft 20yr / 835MW Crane restart).

Full thesis lives at theses/2026-05-17-power-cycle-v1.md.

Locked parameters

Ticker	Layer	Max position	Initial tranche
TLN	Pure-play nuclear IPP (primary)	2.0 R	0.5 R ($2,500)
CEG	Diversified IPP (proxy)	1.5 R	0.5 R ($2,500)
CCJ	Uranium + Westinghouse reactor tech	0.5 R	0.25 R ($1,250)

1 R = $5,000 paper. Bucket cap = 4.0 R (matches memory). Initial deployment = 1.25 R / $6,250.

Phase markers (exit triggers):

1. PJM + ERCOT annual peak-load growth vs forecast — bearish flip if materially below baseline 2 consecutive annual cycles
2. Hyperscaler PPA cadence (≥2 deals/yr at >500MW / >10yr) — bearish flip if zero deals in any rolling 4-quarter window
3. Interconnection-queue backlog (DC growth zones) — bearish flip if wait times drop below 18mo across PJM + ERCOT

Bear-case headline: if a training-efficiency breakthrough drops compute-per-token 10x (a la DeepSeek-R1 propagating to training), hyperscaler capex can flatten without end-user demand dropping — that's what funds the power bucket. Capex line is the load-bearing assumption.

Lock in