All-In PodcastCoreWeave Crash, AI Trade Cracks, Anthropic IPO Red Flags, Bill Gates Doomer
At a glance
WHAT IT’S REALLY ABOUT
AI credit stress tests neo-clouds as Anthropic IPO drama builds
- The episode dissects a sharp selloff in CoreWeave and other “neo-clouds,” focusing on how rising rates and credit repricing can stress highly levered AI infrastructure models even amid strong demand.
- Chamath argues credit derivatives and bond yields are signaling a materially higher perceived default risk due to CoreWeave’s need for substantial additional financing, fixed-price contracts, and customer concentration.
- Horowitz and Sacks contend AI infrastructure differs from the dot-com era because demand is massive and supply is structurally constrained by GPUs, data centers, and power, making today’s economics potentially more durable.
- The panel debates AI-driven software deflation, concluding that while unit pricing may fall, overall software creation and productivity will rise, shifting moats toward data, feedback loops, and especially energy/power access.
- They discuss Anthropic’s upcoming IPO and criticize its public safety messaging as “psyops” aimed at slowing open-source competition, then close with a segment on Bill Gates’ “AI doomer” rhetoric and incentives.
IDEAS WORTH REMEMBERING
5 ideasCoreWeave’s core innovation is asset-backed “boxes,” but complexity is the enemy.
CoreWeave’s “box” structure isolates project cash flows (contract + GPUs + data center costs) and is designed to be financeable as asset-backed/tenant-backed debt rather than pure corporate credit. The panel argues the structure can work, but it’s still a complex, multi-variable machine vulnerable to shocks.
Credit markets—not equity narratives—are driving the neo-cloud panic signal.
Chamath cites CDS implying ~16% annual default probability and ~55% over five years, interpreting it as the “smart money” repricing risk due to higher borrowing costs and future funding needs. Sacks counters that secondary bond pricing doesn’t change existing coupons, but it can raise the cost of future capital and alter lender psychology.
Unlike the internet bubble, AI’s bottleneck is supply scarcity, not demand scarcity.
A central debate is whether AI infrastructure resembles the dot-com buildout. Horowitz argues today’s environment differs because demand is real and supply is constrained (GPUs, power, cooling, labor, entitlements), whereas dot-com collapsed from oversupply and insufficient demand.
If rates stay high or move higher, leveraged AI infrastructure becomes a knife-edge business.
Multiple speakers identify the risk-free rate as the Achilles heel: as the 10-year rises, spread-dependent business models (levered infra, neo-clouds) get squeezed. Even with valuable collateral (GPUs/data centers), rising financing costs can erase margins and trigger reflexive confidence loops (CDS → yields → sentiment).
Watch payback period and delivery penalties—those are the real early-warning indicators.
The group frames the key operating metric as payback period: if capex pays back in ~1–2.5 years, debt and execution risk fall sharply; if it stretches, fragility rises. They also flag “liquidated damages” clauses and operational complexity (hotter chips, power density) as underappreciated risks for data-center operators.
WORDS WORTH SAVING
5 quotesSo it's a coin flip whether they default or not. That's not what I'm saying. That's what the insurance market was saying about their debt.
— Chamath Palihapitiya
These businesses, like CoreWeave, are Rube Goldberg financial machines. They're exceptionally complex.
— Ben Horowitz
In 1999, bandwidth was $1,500 a megabit... By 2001, it was $50 a megabit. So if you were holding bandwidth, you were... and you borrowed money to get it, you're dead.
— Ben Horowitz
Any business model that thrives on a financing spread gets cornholed.
— Chamath Palihapitiya
It's plain as the nose on your face. They're trying to create a panic. This is their latest panic against open source.
— David Sacks
High quality AI-generated summary created from speaker-labeled transcript.
