Leopold Aschenbrenner says 'No More Stocks!'

ELI5/TLDR

Two podcast hosts spend half an hour decoding the trades of Leopold Aschenbrenner, a 24-year-old ex-OpenAI researcher who runs an AI-focused hedge fund that has supposedly ballooned to a $20 billion notional position. The headline: he is short Nvidia — betting against the most valuable company in the world — while loading up on power, memory, optical fiber, and a large private stake in Anthropic. The hosts read Nvidia’s surprise $25 billion bond raise as a possible crack in the AI boom, then mostly talk themselves back out of it. Their conclusion is less “no more stocks” and more “stop buying the obvious AI stock and chase the next bottleneck.”

The Full Story

The setup: a 24-year-old who keeps being right

The whole episode orbits one person. Aschenbrenner left OpenAI, raised a fund of roughly $200 million about eighteen months ago, and — per the hosts — has grown it to a $13.7 billion liquid book, plus another ~$7 billion in a private Anthropic stake, for a claimed $20 billion total. For scale, they note that Bill Ackman’s Pershing Square, built over decades, is around that same size.

“This guy is 24 years old. He has zero and I must emphasize zero investing experience… but he’s made some of the most amazing calls ever.”

That framing matters, because almost everything the hosts say flows from a prior belief that this person is close to infallible. The phrase “copy trade the boy” recurs without irony.

The short that confused everyone

The puzzle the episode opens on: Aschenbrenner’s last 13F filing showed roughly $9 billion in put positions against Nvidia, ASML, and Oracle — the “picks and shovels” of the AI buildout. (A 13F is the quarterly disclosure US funds above a size threshold must file, listing their public stock positions. A put is a bet a price will fall.) Betting against Nvidia, demand booked out to 2029, looked to the hosts like “the worst take ever.”

Then a data point arrived that they read as vindication: Nvidia raised $25 billion in a bond offering — borrowing from outside investors rather than spending its own cash. On the surface that is strange. Why would the company with the fattest margins on earth borrow?

The bond raise, both ways

Credit to the guest (EJ) for actually arguing both sides. The benign reading: it is just cheap money. The analogy he reaches for is a homebuyer who could pay cash but takes a mortgage anyway, because borrowing at a near-government rate frees your own capital for other things.

“It’s effectively free money that they’re raising at. The interest rates are almost nothing.”

The offering was reportedly four times oversubscribed — about $85 billion chasing $25 billion — so Nvidia could pick its terms. Nvidia’s stated reason is plain refinancing, and Google did the same thing weeks earlier. The damning detail that undercuts any “borrowing out of need” story: in the same month, Nvidia’s board authorized $80 billion in buybacks and raised the dividend 25x. Companies in distress do not hand $80 billion back to shareholders. So the honest verdict the episode lands on is: this is opportunistic cheap financing, not a distress signal — though the fact that several hyperscalers all tapped external debt in the same six weeks is at least worth noticing.

The actual thesis: rotation, not collapse

Here is the real signal, and it is more modest than the title suggests. The hosts conclude Aschenbrenner is not calling a top on AI. He thinks Nvidia is fine. What he believes is that the picks-and-shovels trade is overcrowded — too many people own the obvious chip names — and that money will rotate to the next bottleneck.

Where? Down and across the stack: power and energy (his largest category), memory, neoclouds (Coreweave, Iron — companies that rent out GPU infrastructure the way AWS rents servers), and optical fiber (Coherent, Lumentum). The optics logic is the one genuinely educational stretch: as you pack thousands of GPUs together, the copper wiring connecting them overheats and wastes energy, so above a certain density you switch to light-over-fiber, which moves data faster and cheaper.

The Anthropic wildcard

The piece of news the hosts found most surprising: roughly 20% of the fund is reportedly in Anthropic equity, held privately and therefore invisible on the 13F. He allegedly bought in March 2025 at a $60 billion valuation; the company is now cited at $965 billion — a ~15x markup. This is the “invest in the mines, not the picks and shovels” move: rather than betting on who sells AI infrastructure, bet on a company actually building frontier models.

The victory lap

The back third loosens into vibes — a coming SpaceX episode (described, implausibly, as a $3 trillion company that “closed the Cursor acquisition”), Jensen Huang touting Marvell on stage shortly after Nvidia invested in it, a riff about copper futures and visiting mines to “front-run” the next thesis. The one durable idea buried in the noise is from the guest’s bull-vs-bubble checklist: unlike 2008 or the dotcom bust, there are real customers paying for real products, and the buildout is constrained by physical reality — you cannot pour money in and conjure data centers, power lines, and trained workers overnight. That human-and-physics bottleneck is, in his view, what keeps this from being a purely levered bubble.

Key Takeaways

The “no more stocks” framing is clickbait. The actual claim is “rotate out of the crowded chip trade,” not “exit equities.”
Aschenbrenner is short Nvidia/ASML/Oracle (~$9B in puts) but long the rest of the AI stack — power, memory, neoclouds, optical fiber. He is bearish on one layer, not on AI.
Nvidia’s $25B bond raise is almost certainly opportunistic, not distress — proven by simultaneous $80B buyback and a 25x dividend hike. Cheap money taken because it was on offer.
A 13F only shows public positions. Aschenbrenner’s ~20% private Anthropic stake is invisible there, which is why estimates of his fund size jumped from ~$13.7B to ~$20B.
The optics point is real: past a certain GPU density, copper interconnects overheat and you move to optical fiber — a genuine, less-hyped bottleneck.
The bull case the guest actually argues: real paying customers (unlike dotcom) and a hard physical/human-capital ceiling on how fast you can build — limits that prevent the kind of leverage that broke 2008.
Treat the hosts’ hero-worship as the dominant bias. “Copy trade the boy” is the operating principle, which is the opposite of analysis.

Claude’s Take

Strip the worship and there is one legitimate idea here: in any boom, the most obvious beneficiary gets crowded and bid up first, so the marginal opportunity moves to adjacent, less-glamorous bottlenecks. Capital rotating from Nvidia toward power, cooling, interconnect, and the firms actually building the models is a coherent, even sensible, story. The guest deserves credit for genuinely arguing the bond raise both ways and landing on the unexciting-but-correct answer: cheap money, not distress.

Everything else is froth. The episode’s epistemics rest entirely on the premise that a 24-year-old with eighteen months of track record is “very clearly the best AI investor in the world” — a claim made of a survivorship bias and a bull market, not a process. A fund that “doubles every quarter” during the single hottest sector run in a generation tells you almost nothing about skill; it tells you he was long the right theme early and levered. The repeated “copy trade the boy,” the Jensen-says-buy-it-so-buy-it logic (which they half-acknowledge edges toward market manipulation), and the casual $3 trillion SpaceX-bought-Cursor aside all signal a show optimizing for excitement over rigor. The “no more stocks” title is not in the conversation at all.

Score: 4/10. Useful as a map of where AI capital is supposedly rotating and for two or three real mechanics (the 13F blind spot, the copper-to-fiber bottleneck, the buyback-vs-bond tell). Discounted heavily for hero-worship dressed as analysis and a headline the content doesn’t support. Read it for the rotation thesis; ignore the verdict.