All In Podcast: CoreWeave Crash, AI Trade Cracks & Anthropic IPO Red Flags

SPEAKER_1 (0:01)

So the All In Podcast just hit number one, and they brought on Ben Horowitz to talk about what might be the biggest question in tech right now: is the AI infrastructure trade cracking, or are we just seeing normal volatility? And CoreWeave is right at the center of it. Their debt and stock are both under pressure. Bond yields are soaring, credit default swap pricing is climbing, and Elon Musk literally just tweeted "Yikes" about the selloff. CoreWeave, Iron, Nebius— they're all getting hammered. The panel's asking whether this is the canary in the coal mine for the whole AI infrastructure boom. What's interesting is how much of the debate centers on CoreWeave's financing model itself. Right, Michael Intrator had described this "box" structure where client contracts, GPUs, and data centers are bundled together so the cash flow services the debt. That sounded elegant when compute was scarce, but now the panel's worried about rising risk-free rates, heavier borrowing costs, and customer concentration. A huge share of revenue comes from just a few companies. Chamath and the others compare this to earlier infrastructure bubbles, especially LoudCloud during the dot-com era. But here's the key difference they point out—AI doesn't have a demand problem. Compute is still scarce, power is still scarce, data center space is still scarce.

So the real threat isn't that customers disappear. It's that financing gets too expensive before the economics have time to work. That's a critical distinction. Which leads to the bigger question: is this just market volatility or the start of something deeper?

Sacks says, "I would argue that this is just volatility."

He thinks the AI buildout still has strong underlying demand, and the market is simply repricing the cost of capital. The panel keeps returning to this idea—if debt stays manageable, the model works. If rates climb too far, the math breaks. From there, they shift to Anthropic and the scale of frontier AI labs. And they make a fascinating point: Anthropic and OpenAI aren't normal venture companies anymore. They're described as a "compute duopoly," raising massive sums and operating more like public-market giants than startup bets. On Anthropic's possible IPO, the panel notes the revenue is impressive, but the future may depend less on growth and more on securing enormous war chests for compute.

Then the conversation gets ideological. They criticize Anthropic's warnings about open-source models, model distillation, and AI worms on IoT devices—calling much of it fear-mongering. They argue small models are often safer and more practical, and that better technical defenses matter more than government crackdowns. The final theme is power, literally. Electricity, land, and infrastructure may become the ultimate moat in AI as U.S. energy demand climbs.

They close with a warning against what they call AI doom rhetoric from figures like Bill Gates, calling it an "epistemic hall of mirrors" that distracts from building useful technology and real businesses.