Theses 14 / 20 WATCH published 05 Apr 2026 · updated 29 June 2026 · v22

The Efficiency Paradigm

AI compute demand is underestimated by 2-3 orders of magnitude. John's personal spend went from $50-70/month to Max 20x. Token spending at $500-$1000/day is the new normal for power users. GPU rental up 40% ($1.70 — $2.35/hr for H100). But this is NOT a bubble — unlike the late 90s, revenue is constrained by compute supply, not inflated by speculation. Mark's metaphor: "Black hole not a bubble" — tidal forces on large entities, "Long small. Short big." The group-drive problem

+6.8pp
Basket vs S&P 500since published — the verdict number
Basket absolute+22.4% · 1 instrument
S&P 500 same period+15.6%
Call directionWatch

No directional call — this basket is the thesis’s instrument panel.

InstrumentPriceTodaySince publishedvs SPY90d
NVDANVIDIA 217.44 -1.5% +22.4% +6.8pp
SPYS&P 500 — benchmark 761.78 -0.7% +15.6%
Basket: The moat test: efficiency vs scale plays out through NVDA

Recent evidence 283 signals filed

01 SEPT 2026
LeCun's scepticism and the evidence John and Mark say it lacks
31 AUG 2026
John and Mark argue over whether Apple should be selling inference racks

The research system PRO

  • Full analysis — the complete living document behind this call
  • Version history — watch the thesis evolve, v1 → v22, with change notes
  • Evidence archive — all 283 signals filed to this thesis
  • The knowledge graph — entities, concepts & comparisons behind the call
  • Your agent gets a key — full API access to the same system
Get NOOPS Pro
When we're wrong

2 theses are currently moving against us — Microsoft Has Lost the Plot, +34.4% since published; The SaaS Apocalypse, +54.7% since published. They stay on the board. We update priors on evidence, not on evidence we went looking for.