Limitless Podcast

What Leopold Actually Got Wrong (FULL STORY)

Aug 4, 2026

·

26m

·

24 min read

·

gpt-5.4-mini · v7 (default)

mixed

The video’s core argument is that Leopold Aschenbrenner did not suddenly become “wrong” on his AI/infrastructure thesis; he got crushed by 4x leverage on a $45 billion book once memory stocks, rate fears, and public scrutiny turned the trade against him. The hosts say his Situational Awareness fund ran from a few hundred million dollars to $45 billion, then went from legend to fully liquidated in about 20 days, with the public book effectively sold in one block to Ken Griffin/Citadel. They also say Anthropic saved what remained of the fund, keeping year-to-date performance positive even after a 67% drawdown in a month. By the end, the panel’s view is that the thesis still stands for AI infrastructure, but leverage turned a right call into a brutal lesson—especially on Leopold’s wedding weekend.

  • The opening claim is that Leopold Aschenbrenner had been “dethroned” and “wrecked,” and that every one of his public positions was closed at the time of the segment.
  • The hosts say Leopold was in his mid-20s, had left or been fired from OpenAI’s super-alignment team in 2024 at age 23, and then wrote a 165-page AI essay that people thought was essentially dead-on.
  • They say his hedge fund took roughly $225 million and, over two years, turned it into $45 billion, a 1,600% gain, making it the best-performing hedge fund on Earth.
  • The fund’s main bet was two-part: long physical AI infrastructure like compute, GPUs, and memory, and short software/applications such as Microsoft-style software that Leopold thought would be hollowed out by AI.
  • The market started unwinding the thesis about 6 weeks before the episode, with a memory selloff, “the global war,” and oversupply in funding pressures hitting the trade.
  • The speakers say July 10, marked by SK Hynix’s bell-ringing IPO moment, was the top tick for Leopold’s memory trade and the start of the run toward liquidation.
  • They say Leopold was 4x leveraged on the entire fund, so roughly a 17% move against the portfolio could erase about two-thirds of equity and a 25% move could wipe the book.
  • According to the notes, prime brokers and lenders—named as Bank of America, Goldman Sachs, and J.P. Morgan—started demanding money back as the stocks they had lent against fell, which forced cascading liquidations.
  • The speakers say rumors in early July that Leopold was raising more capital were followed by a July 24 investor letter admitting damage, and then by July 30 the whole book was sold in one block and he was fully out of the market.
  • They say the market sold him off so hard that he went from “absolute legend” to someone people said had lost all his money, although the speaker adds that maybe not all of it, but a lot of the portfolio was wiped out.
  • The fund’s public book was reportedly down 67% in a month, yet still up 440% year-to-date at one point; later in the discussion Josh says the same drawdown was still leaving the fund up 80% YTD because of Anthropic.
  • Ken Griffin/Citadel is described as the buyer of Leopold’s entire public equity book, after six hours of work into the early morning before Monday’s open, at 40 to 80 cents on the dollar.
  • The hosts claim the purchase immediately sparked a Monday open relief rally, and that Ken made $4 billion to $6 billion in a couple of hours by buying Leopold’s book if the reported numbers are right.
  • The remaining survivor was Anthropic: the notes say Leopold got in early, reportedly in Series H at a $60 billion valuation, sold a huge amount of Anthropic shares, and the fund still seemed to have about $5 billion of Anthropic exposure.
  • Josh says the fund had effectively become a holding company for Anthropic, and that this single position was what kept year-to-date performance from collapsing completely.
  • Josh argues the post-liquidation surge in names like Novias (+27%), Iron (+26.5%), Bloom Energy (+25%), SK Hynix, and even Biden midday (+16%) showed the market had been hunting Leopold’s positions and ripped once they were gone.
  • The panel closes by saying Leopold was “right” on the thesis but wrong on leverage and execution, that AI infrastructure and memory demand still look structurally strong, and that China is moving closer by turning tokens into commodities, while the sponsored segment pitches Ledger as a security layer where agents propose, humans approve, and Ledger signers enforce.