Key Takeaways
- • Jordi Visser argues the October 2026 selloff reflects an unwind of record hedge-fund leverage rather than the bursting of an AI bubble, with the S&P 500 only 2.7% below its all-time high at the close of the week of October 16, 2026.
- • Hedge-fund leverage hit an all-time record in Q1 2026 per Fed data, having more than doubled from roughly flat levels between 2018 and 2023, and Visser says it will never exceed that peak again.
- • Lone Pine, up 44% through July 2026, erased its entire gain by the end of Q3; Visser estimates an implied gross drawdown of over 55%, which he compares to the 2008 crisis drawdown of a hedge fund.
- • The unwind has spread from AI equities into credit and converts: a synthetic credit proxy (long HYG, short IEF) gave back effectively all of its summer returns in the week covered, while the convert index joined the selloff.
- • Visser rejects the bear case built on 2030 forecasts (Gartner's $3.5 trillion impairment estimate, Bain's $800bn shortfall), pointing instead to signed contracts, sold-out capacity through 2028-2029 and Nvidia's earnings beat despite a compressed forward P/E of about 15.
Summary
In an episode framed around the question of whether the AI bubble has burst, former Morgan Stanley and hedge-fund trader Jordi Visser argues that the AI equity story has not broken, and that what is unwinding is record hedge-fund leverage. As of the Friday close before the episode, the S&P 500 was 2.7% below its all-time high. Visser cites Fed data showing hedge-fund leverage reached an all-time record in Q1 2026 after being roughly flat from 2018 to 2023, and OFR data showing prime-broker-provided leverage also at record highs, with long-short equity near 2:1. He says net exposure across multi-strategy funds is very low while gross exposure remains high, and that there is no clean hedge for the AI trade because it cuts across sectors. He contends that hedge-fund leverage will never again exceed the Q1 2026 peak, attributing the problem to strategies that depend on hundreds of low-correlation positions and high trading activity.
The most striking example he cites is Lone Pine, which he describes as one of the ten best non-multi-strategy hedge funds of the century and says he does not believe is in trouble. The fund was up 44% through July 2026 but erased that gain by the end of Q3. Working back from a 20% performance fee, Visser estimates a gross drawdown of over 55%. He also points to VARA's $20bn flagship AI fund, which fell 44% in August after being up 65% year-to-date through July, and to Situational Awareness, which he says was rescued by Ken Griffin after losses of 60-70%. The Goldman Sachs crowded hedge-fund longs index, he notes, has fallen steadily since July, which he describes as roughly a 15% move over a rolling four-month period on an unlevered basis.
Visser says the pressure spread from equities into credit and converts in October. A synthetic credit proxy of long HYG against short IEF, which had risen steadily since February 2026, gave back effectively all of its summer gains in the week covered. He ties convert-market weakness to heavy AI capital raising, citing Broadcom's issuance of over $100bn this year and Google's raise of more than $100bn across equity and bonds. Other weekly internals he highlights include the S&P 500 down 1.95%, the Nasdaq down 5.5%, the VIX closing just below 20, the VVIX up 25 points, the MOVE index unchanged, and Bitcoin down 2.3%. He cites Mexico's market rising while the QQQ fell as evidence that the move is a factor unwind rather than an AI-specific story.
On the bear case, Visser disputes Gartner's estimate that more than $3.5 trillion of AI infrastructure spending could be exposed to impairment by 2030, and Bain's projection of an $800bn shortfall against $6 trillion of AI needs through 2031. He points to signed contracts, including Micron agreements representing more than 35% of revenue through 2030, ASML's view that AI chip demand extends into 2030, and Amazon CEO Andy Jassy's statement that AI compute is sold out beyond 2028. He also cites Anthropic's revenue growth, from $9bn last year to a pace of $90-100bn this year, and says CoreWeave's debt is spread over six to seven years with customers including Microsoft, OpenAI, Meta, Anthropic and Jane Street. He notes Nvidia's stock has been roughly unchanged since October 2025 while earnings beat consensus, leaving its forward P/E near 15, which he presents as a de-rating driven by a hedge-fund deleveraging cycle rather than a bubble. He concedes that CoreWeave's equity risk premium has risen and that bounces are getting shorter.
The most striking example he cites is Lone Pine, which he describes as one of the ten best non-multi-strategy hedge funds of the century and says he does not believe is in trouble. The fund was up 44% through July 2026 but erased that gain by the end of Q3. Working back from a 20% performance fee, Visser estimates a gross drawdown of over 55%. He also points to VARA's $20bn flagship AI fund, which fell 44% in August after being up 65% year-to-date through July, and to Situational Awareness, which he says was rescued by Ken Griffin after losses of 60-70%. The Goldman Sachs crowded hedge-fund longs index, he notes, has fallen steadily since July, which he describes as roughly a 15% move over a rolling four-month period on an unlevered basis.
Visser says the pressure spread from equities into credit and converts in October. A synthetic credit proxy of long HYG against short IEF, which had risen steadily since February 2026, gave back effectively all of its summer gains in the week covered. He ties convert-market weakness to heavy AI capital raising, citing Broadcom's issuance of over $100bn this year and Google's raise of more than $100bn across equity and bonds. Other weekly internals he highlights include the S&P 500 down 1.95%, the Nasdaq down 5.5%, the VIX closing just below 20, the VVIX up 25 points, the MOVE index unchanged, and Bitcoin down 2.3%. He cites Mexico's market rising while the QQQ fell as evidence that the move is a factor unwind rather than an AI-specific story.
On the bear case, Visser disputes Gartner's estimate that more than $3.5 trillion of AI infrastructure spending could be exposed to impairment by 2030, and Bain's projection of an $800bn shortfall against $6 trillion of AI needs through 2031. He points to signed contracts, including Micron agreements representing more than 35% of revenue through 2030, ASML's view that AI chip demand extends into 2030, and Amazon CEO Andy Jassy's statement that AI compute is sold out beyond 2028. He also cites Anthropic's revenue growth, from $9bn last year to a pace of $90-100bn this year, and says CoreWeave's debt is spread over six to seven years with customers including Microsoft, OpenAI, Meta, Anthropic and Jane Street. He notes Nvidia's stock has been roughly unchanged since October 2025 while earnings beat consensus, leaving its forward P/E near 15, which he presents as a de-rating driven by a hedge-fund deleveraging cycle rather than a bubble. He concedes that CoreWeave's equity risk premium has risen and that bounces are getting shorter.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.