When the SPAC King starts asking if your sector is a bubble, it's not the question that kills you—it's who's listening.
The Summary
- CoreWeave's bonds hit distressed territory (1,000+ basis points over government bonds), shares down 20%+ since Chamath Palihapitiya questioned whether AI infrastructure is a private credit bubble
- Contagion spreading: other AI infrastructure companies saw collateral damage in the market
- CoreWeave's debt problem is becoming the industry's credibility problem—private credit has funded the GPU arms race, and now investors are asking what happens when the music stops
The Signal
CoreWeave went public on the premise that AI compute is the new oil drilling. The company raised billions in debt to buy Nvidia GPUs, then leased that compute to companies building foundation models. Clean arbitrage: borrow cheap, charge premium rates for scarce H100 clusters, collect the spread. Except the spread is narrowing, and the debt is coming due.
The 1,000 basis point threshold matters because institutional investors have mandate limits. Once bonds cross into distressed territory, certain funds are required to sell. It becomes a mechanical spiral, not just a sentiment problem.
"When the SPAC King echoes your concerns, the smart money has already left the building."
Here's what makes this different from typical tech volatility: CoreWeave's debt structure represents how the entire AI infrastructure layer got built. Private credit funds couldn't get exposure to OpenAI or Anthropic directly, so they financed the picks-and-shovels layer. CoreWeave, Lambda Labs, Crusoe Energy—they all raised debt against future compute revenue that assumed:
- GPU scarcity would persist
- Foundation model companies would keep raising capital
- Inference costs wouldn't drop
- Hyperscalers wouldn't flood the market with their own capacity
At least two of those assumptions are now broken:
- Nvidia supply constraints have eased significantly
- Google, Microsoft, and Amazon are offering compute at cost to strategic partners
- Inference optimization techniques have cut costs 10x in 18 months
The Implication
Watch the credit markets, not the equity markets. If CoreWeave's distressed debt spreads to other GPU-backed borrowers, it reveals how much of the AI boom was financed on assumptions about persistent scarcity. The companies that survive will be the ones that own their infrastructure outright or have locked in long-term contracts at prices that still make sense when GPUs become commoditized.
For everyone building on rented compute: your landlord might be getting margin calls. That's a different kind of dependency risk than most startups price in.