🔮 Leopold & exponential markets; transformative GLP-1s; runaway AI & the future of safety++ #595
Exponential View ● Covered by 6 sources
Leopold Aschenbrenner's $45B AI-bet fund got forced to liquidate this week. Even a right thesis can't survive 4x leverage when chip stocks slide.
Based on reporting by Exponential View — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Leopold Aschenbrenner spent the last year making one of the boldest bets in finance: that the race to superintelligence would pour trillions into compute, chips, and power, and that betting on that build-out with real conviction would pay off. He was an AI researcher, not a hedge fund veteran, when he launched Situational Awareness LP in late 2024. He leveraged the trade roughly four times over. At its peak the fund managed $45 billion. This week it was forced to liquidate.
The unwind wasn't gradual. The Philadelphia Semiconductor Index dropped 28.6% from its June high, while software stocks moved the other way and gained. That's the kind of reversal that turns a leveraged bet from genius to margin call in a matter of weeks, no matter how sound the underlying logic looked six months earlier.
And that's the important distinction here. Situational Awareness collapsing doesn't mean Aschenbrenner's thesis about AI's compute demands was wrong. It means a highly leveraged, single-direction bet on that thesis couldn't survive the market's mood swinging the other way, even briefly. Being right and staying solvent are two very different skills.
There's a parallel playing out inside companies right now, just slower and quieter. A model comparing three companies adopting AI, all starting from the same economics and the same 5% hit rate but different habits around learning, shows something uncomfortable: two years in, all three look like they're losing similar amounts of money. By year five, the company that will eventually lose the race actually looks like the front-runner. It takes eight years before the approach that compounds learning pulls decisively ahead. Most CEOs deciding how much to spend on AI right now have no way of knowing, from where they're standing, whether they're funding compounding advantage or just burning cash.
Meta's Mark Zuckerberg is held up as the counterexample, dubbed by the FT as running a portfolio of side quests. He doesn't need any single project to pay off on its own. What matters is whether the experiments keep deepening the company's infrastructure and shaping the next move. That's the system-builder mentality: patient, diversified, built to compound rather than to bet everything on one trade going the right way in one direction.
My take — AI-written commentary, not fact-checked reporting
Leverage is what killed this fund, not the idea behind it, and that distinction gets lost every time a big bet blows up in public. People will point to Aschenbrenner's liquidation as proof AI infrastructure spending is a bubble, when really it's proof that betting four times your money on any single narrative is reckless regardless of whether the narrative is true. The companies and investors who'll actually win this cycle are the boring ones treating it as an eight-year compounding game, not the ones trying to prove they called it first.
Read more about this at: Exponential View