Hedge Fund Deleveraging, Not an AI Bubble

FinanceThe AI Bubble Didn’t Pop. Hedge Fund Leverage Did.

TL;DR

The recent market turmoil stems from hedge fund deleveraging, not an AI bubble burst; the S&P 500 is only 2.7% off all-time highs while AI infrastructure demand and revenue growth remain strong through 2030.

Key Takeaways

1Hedge fund leverage has peaked and is in 'complete and utter dismantling'—the speaker predicts it will never exceed current levels due to poor risk-adjusted returns (Sharpe ratios) despite leverage more than doubling from 2018 to 2023.
2Multi-strat hedge funds lost historic amounts in July and face ongoing deleveraging through September, with Vara down 44% in one month and Lone Pine erasing entire year-to-date gains (55% gross drawdown) in a single quarter—LTCM-like behavior.
3CoreWeave trades near all-time highs only 5 times in its existence, but its equity is down far less than bubbles that actually popped (50–60% drawdowns); clients include Microsoft, Meta, and Anthropic with signed supply agreements through 2030.
4GPU prices remain elevated or rising (H100s higher than a year ago per Zuckerberg); Nvidia guided 70% revenue growth vs. 40% expected; ASML, Lumentum, and Amazon all report AI chip demand sold out through 2028–2030, contradicting 'bubble' narratives.
5Anthropic revenue grew 10x year-over-year for three consecutive years (9B to 90–100B run-rate); OpenAI and Anthropic combined expected to reach 600–700B, meaning only a 3x deceleration from 10x prior years, not a collapse.
6Agent workflows currently adopted by only 2% of companies; consumer agents and agent swarms launched in the last 2 months, representing step-up in GPU demand; robotics breakthrough via GPT-6 expected next year, unlocking new revenue pools beyond current token pricing.
7Micron has signed agreements representing >35% of revenue through 2030; Gartner's 3.5T impairment forecast (out to 2030) is dismissed as linear-thinking forecasting while real business already sold out through 2028, invalidating bear case.

Notable Quotes

“This is not a bubble popping. It might be in the future. It's not now.”
— Speaker

“Hedge funds are the problem, guys. Mark my words. They are the problems.”
— Speaker

“There is no end to this build-out. There is no end. And because the progress is driving the demand and the progress keeps accelerating, I cannot make it any clearer.”
— Speaker (referencing Brad Gerstner)

“Compounding is your friend. And the S&P 500 and all investments, effectively equities represent innovation.”
— Speaker

Chapters

1. The AI Bubble Did Not Pop—Hedge Fund Leverage Did

Opens with clarification that panic emails comparing current market to 2008 and dot-com are misguided; the real problem is hedge fund deleveraging, not equity or economic weakness. S&P 500 is 2.7% off all-time highs—far from historic 50–60% drawdowns seen in true bubbles.

2. Hedge Fund Leverage at All-Time Highs, Now Collapsing

Fed data shows hedge fund leverage more than doubled from 2018 to 2023; multi-strat leverage peaked and declined; prime broker leverage at all-time highs through Q2. CoreWeave $45B positions with $33B borrowed capital highlight the problem: as markets move against crowded AI positions, forced deleveraging cascades.

3. Catastrophic Hedge Fund Losses and LTCM Parallels

Vara down 44% in August; Lone Pine (one of the greatest funds ever) erased 44% year-to-date gains in Q3, equivalent to a 55% gross drawdown—LTCM-like volatility never seen before. Drawdown levels trigger investor redemptions and force funds to liquidate positions regardless of fundamentals.

4. Crowded AI Longs Unwind, Spreading to Credit and Converts

Goldman Sachs crowded hedge fund longs (dominated by AI, ~25% of S&P 500) down month-to-date in a straight line; credit spreads widened; convertible index waterfall (ex-CoreWeave, MicroStrategy). Hedge funds short-stacked after July losses cannot absorb more volatility, triggering contagion.

5. CoreWeave and AI Infrastructure: Not a Bubble, Built-Out Reality

CoreWeave's 5th visit to current equity price level; historically, bubbles fall much further (AIG, Sienna). Clients (Microsoft, Meta, OpenAI, Anthropic) have best-in-world balance sheets; GPU prices rising per Zuckerberg; ASML, Lumentum, Amazon all report sold-out supply through 2028–2030.

6. Revenue Forecasts and the Exponential Growth Paradox

Barclays desk claims 3.5T impairment by 2030; Gartner forecasts 800B shortfall by 2031. Speaker dismisses as linear-thinking: Anthropic 10x three years running; OpenAI + Anthropic expected 600–700B (3x deceleration, not failure); Micron has signed agreements for >35% revenue through 2030. Agent swarms, robotics, and frontier labs add new revenue streams not yet monetized.

7. Agent Workflows, Robotics, and Demand Acceleration

Only 2% of companies use agent workflows currently; consumer agents and agent swarms launched in last 2 months; GPT-6 expected to enable robotics breakthrough next year. Unitree CEO (Chinese company close to government) says OpenAI advances accelerated GPT moment in embodied intelligence—robotics = huge new revenue pool.

8. Market Structure Has Changed; Tokenization Is the Answer

Hedge funds trade on short-term arbitrage; AI requires certainty gone, forcing leverage down but not returns down. Public markets face disintermediation; SemiAnalysis grew 200M to ~2B revenue in ~1 year with 90 employees (10x stock price). Tokenization removes covariance matrix and margin constraints, enabling winners to scale without prime broker limits.

9. Consumer Confidence Crisis and Jobs Weakness Hide Behind Stock Gains

Consumer confidence at lowest since 2013 (below 2020 pandemic low); jobs plentiful vs. hard to get inverted; financials and tech payrolls shrinking for 18 months (ex-healthcare). Unemployment shadow rate ~6% vs. headline 4%. Agent adoption will intensify labor pressure, margin expansion trade.

10. Crypto As Hedge Against Deleveraging; Bitcoin Hangs in There

Bitcoin down 2.3% week vs. Nasdaq down 5.5%, S&P down 1.95%; crypto up significantly since June despite AI pain. Speaker moved majority of portfolio to crypto; expects Bitcoin to meander toward 75k until hedge fund liquidation stabilizes. Tokenized digital economy (100M+ revenue companies) to dwarf public equity returns.

Key People & Entities

Brad Gerstner: Investor; consistent predictor of AI capex and GPU demand; argues AI CapEx cycle has no visible end—permanent re-architecture of global compute.
Jensen Huang: CEO of Nvidia; guided 70% revenue growth for next year; one-word answer ('forever') to question of whether AI CapEx cycle ends.
Dylan Patel: Founder of SemiAnalysis; 30 years old; vertically integrated AI research covering data centers, semiconductors, and models; company grew 5x in one year from $20M to $100M revenue.
Andrej Karpathy: Former Tesla AI leader; clarified that most people have little AI exposure, few are frontier users; argued exponential innovation cannot be predicted.
Mark Zuckerberg: CEO of Meta; discussed GPU value remaining high, agent adoption, and that old GPU prices are up vs. a year ago despite supply concerns.
Gavin Baker: Investor who has navigated multiple bubbles; provided revenue forecasts for frontier labs ($600–700B combined for OpenAI and Anthropic by ~2026).
Steve Mandel: Founder of Lone Pine Capital; one of the greatest hedge fund managers; fund down 55% gross in Q3 after erasing year-to-date gains.
Aaron Levie: CEO of Box; noted that agent swarms (launched in last 2 months) will consume majority of future token demand.

Glossary

Multi-strat: Multi-strategy hedge fund using hundreds of different trading strategies across asset classes, seeking low correlation diversification.
Prime broker: Bank or broker providing leverage, lending, and settlement services to hedge funds; supplies the majority of hedge fund capital.
Leverage: Borrowed capital used to amplify returns; hedge funds borrow to increase position sizes beyond their own capital, multiplying both gains and losses.
Deleveraging: Forced or voluntary reduction of borrowed positions; occurs when volatility rises, margins are called, or investors redeem; cascades through markets as many funds must sell simultaneously.
Crowded trade: Investment position held by many market participants (e.g., long AI names); when crowds exit together, liquidity evaporates and prices fall sharply.
Sharpe ratio: Risk-adjusted return metric (excess return per unit of volatility); hedge funds must deliver attractive Sharpe ratios to justify leverage; declining ratios force leverage reduction.
Drawdown: Peak-to-trough decline in a portfolio's value; historical bubbles saw 50–60% drawdowns; current AI/equity decline is ~2.7%.
LTCM: Long-Term Capital Management, a multi-billion-dollar hedge fund that collapsed in 1998 due to leverage and correlation breakdown during the emerging markets crisis; required Federal Reserve bailout.
Agent workflows: Use of AI agents to automate business processes and decision-making; currently adopted by ~2% of companies, expected to drive future AI revenue and labor disruption.
Tokenization: Conversion of assets (equity, debt, revenue streams) into blockchain-based tokens; removes traditional prime broker middlemen and covariance matrix constraints on leverage.

Explore