Chip Stocks Crash, $20B Fund Margin Called, Frontier Labs: SLOW DOWN AI, Mamdani's Grocery Stores

The episode opens as a hangout and closes its first two minutes as a risk lecture. There is housekeeping about chart position, a forgotten Starlink terminal, five enterprise leads described in Glengarry Glen Ross terms, and then Jason asks Sacks for enterprise-sales advice on Chamath's behalf. Sacks's answer is two sentences: "No leverage. Don't put on leverage." Jason immediately converts the joke into the episode's thesis statement: "Leverage equals risk of ruin."

The placement matters. The audience is told the moral before it is told the story, so everything that follows about Leopold Aschenbrenner is read through a frame the hosts have already built. Chamath's own contribution to the bit is the sharpest setup of the hour: he jokes that if you name your fund "situational awareness" you are welcome on the pod anytime. That is not an aside. Situational Awareness is the title of the essay that made Aschenbrenner a public figure, and the fund that just got liquidated was the hedge-fund version of that essay. The show knows this, and the joke is doing structural work.

There is also a small, telling exchange about what "situational awareness" is supposed to mean in practice. Chamath says he has the situational awareness not to lever up. Nobody disagrees. The implicit argument of the opening is that the failure being described is not an information failure. The people involved knew the risks, could describe the risks, and were rewarded for taking them anyway.

The reported facts, as the hosts lay them out: Aschenbrenner is 25, left OpenAI two years ago, started the fund with roughly $225 million in 2024, grew it about 100x to $20 billion, ran it up to $45 billion, and was at "200 x" earlier this month. CNBC is cited for a fourfold, 450% gain year to date. On the Thursday of taping he was margin called and sold the entire public book; Citadel bought it, though the hosts concede they do not know whether Ken Griffin was personally involved. A reported sale of a large Anthropic stake is disputed by the Wall Street Journal, and the show says so.

Chamath supplies the mechanics rather than the gossip, and this is the most useful passage in the segment. If the leverage was about three and a half turns, then a 34% move in the underlying is amplified twelve- or thirteen-fold, and the 25% move of the last three days is amplified to 75%. At that point the position does not fail slowly; it is closed. "It's sort of an automatic one way ratchet." The prime brokers start calling around, and the owner of the book has no meaningful choice. Chamath's closing questions are the right ones: what AUM is left, what is the high-water mark, and can he dig out.

The market backdrop is presented with a lot of numbers, and several of them should be flagged as 数字存疑，不可对账 — doubtful, do not reconcile. The Philadelphia Semiconductor Index is said to be down over 20% in a month, which is a clean bear-market definition, and to have bounced 7% on the day of taping. But then Samsung is "down 38% over last month", SK Hynix "down 14% since going public three weeks ago", and the KOSPI "down over 40% in the last forty days". SK Hynix has been listed for years, so the IPO framing is a transcription or date error. A 40% index decline in forty days is not consistent with a 38% decline in the largest constituent and 14% in the second largest. The five-year return figures (Micron 850%, Nvidia 875%, Broadcom 663%) are directionally plausible and internally consistent; the monthly and weekly ones are not. Treat the second group as colour, not as data.

What survives the audit is the shape of the event rather than its magnitudes: a violent, fast, leverage-driven liquidation of a concentrated AI and semiconductor book, absorbed by a deep-pocketed buyer, with the principal's remaining capital and incentive fee structure now in question. That is enough to carry the rest of the discussion, and it is the part the hosts actually reason about.

Sacks poses the question the episode keeps returning to: is this correction driven by fundamentals or by momentum? His answer is momentum. Memory chips ran roughly 10x in a year, everything adjacent to the AI capital-expenditure boom rose with them, and a pullback was inevitable. The Nasdaq fell about 10% from its peak, but the momentum trade fell 30% to 40%, because that trade was the most exposed part of the index. Korea and the Aschenbrenner fund are presented as two views of the same leverage sitting behind the same trade.

He then separates price from substance. The real question is whether the capex is sound: is it an investment the hyperscalers should be making, and will it deliver a return? His answer is that it is real, and that the hyperscalers have spent essentially all of their free cash flow "and then some" on it. He is explicit that he is not predicting stocks or telling people when to buy. That caveat is doing a lot of work, because the claim that the capex will pay off is asserted rather than evidenced — there is no payback period, no marginal revenue per GPU, no utilisation figure anywhere in the segment.

The bridge between the two halves of the argument is the Buffett-or-Munger line that leverage is the only way smart people go broke. Sacks walks the counterfactual: unlevered, the same portfolio is down 30% this month, up 7% today, down "20 something percent" for the month after rising 10x over the year. Levered three or four times, it is gone. The arithmetic is unanswerable, and it is the cleanest thing said in the whole hour.

Sacks knew Aschenbrenner before the fund existed and describes him as an interesting thinker who laid out the bull case cleanly. The essay's structure, as reconstructed here, is three multiplicative orders-of-magnitude improvements. Raw compute gets about 3x better per year, roughly 10x every two years. Algorithmic efficiency — reinforcement learning and similar techniques — also improves about 3x per year, again 10x every two years. And "unhobbling", which Sacks translates into modern terms as the harness and the connectors, the ways of actually putting the model's decisions to work, improves on a similar curve.

Multiply those and you get 100x over four years and 1000x over six. Sacks's reason for liking the argument is methodological rather than empirical: "his thinking in Oom's always appealed to me because I think most people just don't think in exponentials". Chamath offers virality and the PayPal era as the historical analogy, where the team could see an exponential curve and extrapolate from it. Jason supplies the intuition pump about how large the gap between a billion and a trillion is.

The critique that nobody in the room makes is the one that matters. The three drivers are treated as independent rate constants that can be multiplied, but unhobbling is not a rate constant — it is a bag of product and integration decisions whose returns are lumpy and eventually bounded by the workflow being automated. More importantly, the same exponential that justifies holding the position also justifies the position size. If you believe the underlying compounds at 10x every two years and you are confident in the direction, leverage looks less like recklessness and more like optimal sizing. The error is not in the thesis; it is in the inference from a correct long-horizon view to a survivable short-horizon position.

Chamath adds the structural point that most coverage of the blowup misses: fund growth changes who owns the return. Early investors are up 10x or more; the investors who arrived in the last few months, after the fund became the obvious place to pile into, are wiped out. If $10 billion came in recently, then the headline "up 10x" describes almost none of the current capital base. The high-water mark question follows directly: a manager who has lost most of the recent gains has to earn it all back before he earns a fee again, which changes behaviour.