← Feed📈 Board🎯 Predictions🚧 Bottlenecks📐 Calibration📊 AnalysisOwn-the-Bottleneck ThesisOwn-the-Bottleneck

Own-the-Bottleneck — Own-the-Bottleneck

Theme thesis — tracked as a cross-cutting view, no single ticker.

Running conviction (P)
0.64prior 0.76

Investment Read as of 2026-10-22

The Read

MIXED investment. While hyperscaler demand remains strong, regulatory friction and local power grid constraints are creating significant execution risk, dampening immediate conviction despite the core thesis holding up. (P=0.64).

Bull case

Bear case / what breaks it

What the latest signal says

The signals show a clear tension: massive, committed demand (Microsoft, Google) is running headlong into severe local power and regulatory resistance (ISO-NE withdrawal, Arbol insurance). This confirms that while the *need* for bottlenecks remains high, the *ability* to build out supply is uneven and politically volatile.

Posterior history

DatePΔCallDriver
2026-10-220.64+0.02UP/LOWSame ISO-NE story: when connecting costs $100M+, most new supply gives up. Power is still
2026-10-180.62-0.03DOWN/LOWThe data-center backlash itself isn't new; we've seen it for 9+ days. What's new is the si
2026-10-160.65-0.07DOWN/MEDThis is the week's sharpest point against us. For copper, the glut is at the processing st
2026-10-150.72-0.07DOWN/MEDA Virginia state senator has proposed a three-year statewide pause on new data centers. Vi
2026-10-140.79-0.07DOWN/MEDOregon's governor asked the state environmental regulator to pause permits. That works as
2026-10-020.86-0.07DOWN/MEDFour substitution/new-supply items in one week — rare-earth-free magnets, DOE transformer
2026-10-010.93+0.05UP/MEDFour jurisdictions in one day moved the binding constraint from steel to politics: PJM del
2026-09-300.88+0.02UP/LOWThree independent jurisdictions moved the binding constraint from power to permission in o
2026-09-290.86+0.05UP/MEDPentagon put $450M into Elmet for domestic tungsten, and the WSJ reports the broader US mi
2026-09-250.81-0.07DOWN/MEDThree NEW items all cut the same way: Escondida restarted (a bottleneck easing), Hormuz re
2026-09-170.88-0.07DOWN/MEDGoogle, Nvidia, Anthropic and RWE formed a coalition with Emerald AI to free up ~100 GW of
2026-09-150.95+0.05UP/MEDPJM-commissioned study models loss-of-load "over 100x worse than planning criterion" by 20

Thesis detail

Core thesis

AI compute is a demand shock cascading down a power supply chain that can't keep

up. The durable money isn't the headline names — it's whoever controls the

scarce inputs and conversion steps everyone depends on. Own the bottleneck,

not the company behind it. Processors beat miners.

Pillars (with priors)

1. Power is the binding constraint on AI (DC load 183 TWh '24 → 325–580 '28) · P = 0.85

2. Value concentrates at the conversion step (HALEU not uranium; refining not mining) · P = 0.80

3. 6 of 11 top bottlenecks are single-country-controlled → supply security = pricing power · P = 0.75

4. Energy *addition*, not transition — every source scales at once · P = 0.65

Expected news (the prior)

Thesis-breaking triggers (→ set relevant P near 0)

Leading vs lagging indicators

Key metrics

Cross-arena sensors

B1 (demand/efficiency), B2 (the constraint), B7 (supply security), B4 (magnets).

The biggest signals are *cross-bucket connections*.

Component theses

CLF (GOES) · LEU (HALEU) · CEG (nuclear) · MP (rare earths) · FCX/SCCO (copper)

Posterior log