The Crossover

The instrument.

Five-name fixture RAR, coverage, SpaceXAI 2.7 / 0.4, NBIS caption. Live five-name median is not computed. Not a trade signal.

← Research desk

The instrument

Fixture RAR · not a live capex median · not a trade signal

META 0.82 / 1.41 / 0.63 is the published fixture triple — do not recompute.

Snapshot of the live board. Figures update on the host.

Open live board

Board legend →

Five notes

Substack
The worker is not 2.32 kilowatt-hours
Two-thousand-calorie label is ingested energy, not delivered work. The unpaid loop is what capex has to replace.

Sep 8, 2026
Substack
Coverage is not duration
Remaining performance obligation is booked work, not chip life. Three issuers. Three ways the ratio lies.

Sep 8, 2026
Substack
The leased watt vs the internal robot
SpaceXAI pair 2.7 vs 0.4. Leased watt and internal robot are not the same unit.

Sep 8, 2026
Substack
The same test, five filers
One fixture, five names. META / ORCL / CRWV / MSFT / GOOGL. Not a live median.

Sep 8, 2026
Substack
The residual
What is left after coverage and RAR. Series close, not a price target.

Sep 8, 2026

Notes

Substack Draft
The instrument
Board legend. Fixture RAR, empty cells, NBIS caption. Live median not computed.

Sep 10, 2026
Substack
As of 8 September 2026
Standing falsifiable test through year-end 2029. Series coda.

Sep 8, 2026

Standing test as of 8 September 2026: the series is false if, using only signed multi-year named-counterparty revenue and the Article 2 parameter table, the median return on a disclosed gigawatt-scale training or inference fleet remains at or above one through year-end 2029 inside the published sensitivity ranges.