Ben Horowitz: CoreWeave Crash, AI Trade Cracks, Anthropic IPO Red Flags, Bill Gates Doomer

The episode opens on a five-year milestone: the show finally reached number one on Apple Podcasts, a fact the hosts spend several minutes claiming credit for. David Sacks is back, and Ben Horowitz fills in for Friedberg, who is described as busy “saving the world from famines” and due back the following week. The banter sets the register for the rest of the hour — competitive, self-aware, and quick to pivot from jokes to balance-sheet arithmetic.

The first substantive turn is the week's market action, framed as a possible deflation of the AI bubble. The “Neo Clouds” — CoreWeave, Iren and Nebius — were all down roughly 20% as public companies after CoreWeave's bonds started to sell off. The mechanism is described in sequence: on Tuesday, posters on X, including Chamath, flagged that the yield on some CoreWeave debt had reached 15%, because bond traders doubted the company could generate enough cash to service it.

Elon Musk then replied with a single five-letter word, which the panel treats as the proximate cause of the headline cycle that followed. By Wednesday, CoreWeave, Iren and Nebius had fallen between 10% and 15%, and the damage spread to Oracle, Dell and Nvidia, the last down about 3%.

CoreWeave is the tip of the spear. They've raised over $35,000,000,000.

Chamath's analytical frame is that the informative price is not the equity but the credit derivative. For companies that must borrow heavily to fund capacity, the smart money trades credit default swaps — default insurance meant to protect bondholders. He is careful to distinguish his own view from the market's, a distinction the segment returns to more than once.

The numbers he cites: the risk spiked this week to roughly 953 basis points, which he converts into an implied probability of default of about 16% in any given year, and about 55% over a five-year duration — “a coin flip”. He attributes the move to the insurance market rather than to his own forecast.

That's not what I'm saying. That's what the insurance market was saying about their debt.

The charts that follow are a market check rather than a company check. The risk-free rate has been pushed higher by central bankers and bond vigilantes demanding more compensation for long-term inflation and other unresolved risks, and as the ten-year rises, everything else rises with it. Against that backdrop the average yield on the junkiest junk debt climbs, and CoreWeave's own cost of debt has moved up “relatively violently”.

The segment leans on a clip of CoreWeave founder Michael Intrator explaining his structure. The company signs a compute contract with a customer such as Microsoft, then creates a “box” containing three things: the customer contract, the GPUs bought from Jensen, and the data center contract. Cash flows into the box and out through a fixed waterfall.

The waterfall, as Intrator describes it: the customer pays the box rather than the company; the box first pays the data center, then the power bill, then interest and principal; whatever remains flows back to CoreWeave.

Intrator's claim is that each box is independent and does not cross-pollinate or create contagion, and that within a five-year deal the economics are repaid inside two and a half years. Chamath accepts that the explanation is reasonable on its face, then argues that something changed in the last six months.

within two and a half years of a five year deal, we have paid for everything

The problem he identifies is that the contracts are fixed-price. If spot pricing rises, CoreWeave does not capture that revenue, because the capacity was pre-sold at a set number. At the midpoint of his chart, the company still has to raise another $35 billion of debt to finance everything it has signed, on top of the $35 billion already outstanding, and the marginal rate is rising for reasons that have nothing to do with management.

Today the blended rate across the funding pools is about 8.5%, but the GPU loan facility he cites, DDTL 1.0, spiked to 15%, which theoretically could exceed the rate the company must pay on everything still unfinanced. Layer on extreme customer concentration — 75% of revenue from three customers, some of them walking down forward expectations — and Chamath's team estimates the company sits roughly $5 billion, plus or minus, from breakeven.

His conclusion is a reflexive loop rather than an insolvency call: higher risk-free rates raise borrowing costs, which erodes margin, which pushes the company toward breakeven, which moves CDS, which moves bond yields. He concedes the company is probably not in real danger, but notes that this is a confidence game and that the market must absorb another $35 billion from these issuers alone.

Ben Horowitz's first move is to widen the frame rather than dispute the arithmetic. CoreWeave and its peers are “Rube Goldberg financial machines,” exceptionally complex, and the complexity is not only financial: data center contracts of uncertain length, memory purchases in the middle of a memory shortage, and GPUs whose duration and price have both been rising.