Thinking in Bets: Why the AI Bubble Fears Are Wrong—Again
In this week's video, I tackle the latest wave of panic—S&P down just 3.5% from all-time highs while my portfolio is down 35%—and explain why this feels identical to every other correction we've climbed over the past two years. Despite economist covers, CoreWeave CDS fears, and institutional hand-wringing about AI credit stress, the data tells a different story: Nvidia just posted $57 billion in quarterly revenue (up 62%), S&P profit margins hit all-time highs, and earnings revisions remain elevated globally. The AI buildout isn't slowing—it's accelerating into physical infrastructure constraints that most investors still don't understand.
The real shift: we're moving from GPU scarcity to power scarcity, from code-driven margin expansion to grid-driven capital rotation. Google's Gemini 3 launch underscores that AI capabilities continue advancing at pace, yet junk spreads remain tight, bank stocks are up 13% year-over-year, and leading economic indicators show none of the credit stress that precedes actual bubbles. Meanwhile, Bitcoin's 36% drawdown from recent highs has pushed sentiment to Liberation Day levels, creating what I view as a compelling asymmetric entry point heading into a year where PMI expansion, Fed rate cuts, and tokenization tailwinds converge.
I walk through Annie Duke's Thinking in Bets framework as a mental model for navigating this volatility—every decision is a probabilistic wager on incomplete information, and right now the crowd is assigning far too much probability to bubble scenarios while ignoring structural drivers. Small-cap earnings growth is forecast near 60%, international equities show the largest relative growth differential in a decade, and the K-shaped economy continues separating AI-enabled winners from legacy laggards without triggering a traditional recession. This is exactly when alpha gets made.
Timestamps
(00:00–03:45) Market panic context: S&P down 3.5% feels like capitulation, retail fear at Liberation Day levels, why this drawdown (5.6% S&P, 8% NDX) pales versus historical corrections
(03:45–08:20) Thinking in bets: Annie Duke's framework for decision-making under uncertainty, why emotions dominate when you believe you know things you don't
(08:20–14:30) AI bubble fears debunked: Economist covers, CoreWeave CDS, Oracle credit stress—compared to dot-com (800% 5-year NDX gain vs. 95% today), junk spreads tight, bank stocks up 13% YoY vs. down 20% at prior peaks
(14:30–21:15) The data that matters: Nvidia $57B revenue (+62%), Palantir +63% revenue growth, S&P profit margins at all-time highs, AI mentions across every S&P sector, DRAM prices rising (not falling)
(21:15–28:40) Why this isn't 5G telecom: Hyperscalers are high-margin, low-leverage cash-flow machines vs. levered telcos; AI capex looks more like 1990s semiconductor boom or early AWS buildout
(28:40–35:50) Recession that isn't: Leading indicators negative since 2022, PMI 20-month average in historical recession territory, temp jobs collapsing, yet S&P earnings climbing—K-shaped economy confirmed
(35:50–42:10) Bitcoin drawdown: 36% from highs, fear/greed at 6, daily sentiment index at futures-era lows, RSI below 21—historical buying zone despite being down 9% YTD after back-to-back 100%+ years
(42:10–48:55) Earnings revisions and PMI setup: 20-week moving average of revisions still elevated, international earnings growth accelerating faster than US, small-cap/mid-cap EPS growth near 60%—all pointing to 2025 PMI expansion
(48:55–52:30) Investment takeaway: Craig Shapiro's insight on AI demand colliding with physical limits—hyperscalers can pay legacy industrials to go offline and use their grid allocations; the next moat is control of land, power, water, and grid access