CoreWeave bond yield tops 16% as AI credit stress rises

CoreWeave’s bond yield spiking above 16% is being read as a warning sign for heavily indebted “neocloud” AI infrastructure providers, whose junk-rated debt now carries borrowing costs comparable to credit cards. Commenters contrast this with far lower yields for giants like Google and U.S. Treasuries, arguing that easy money for Big Tech and massive government issuance are crowding out weaker borrowers. Beneath the surface is a broader worry that AI demand is being propped up by subsidized tokens and optimistic growth narratives, while power, data center capacity, and sustainable profitability for firms like CoreWeave, Anthropic, and OpenAI remain uncertain.

CoreWeave bond stress

  • CoreWeave’s bond coupon was ~9.625% at issuance (already junk-rated, around BB-); secondary-market yields have climbed above 16%, implying markets now see substantially higher default risk.
  • Commenters question whether the bonds are secured (e.g., by GPUs) and where they sit in the bankruptcy hierarchy, noting this will matter in a restructuring.
  • Some compare CoreWeave to dot-com-era datacenter failures: overbuilt capacity, high leverage, and never reaching profitability.

Comparison to other debt (Google, Treasuries, credit cards)

  • Alphabet bonds with yields around 5.7–6.1% and US Treasuries around ~4.6–4.9% are seen as anchors; Google’s spread over Treasuries is under 100 bps, reflecting AA+ credit.
  • Debate over risk: some argue Treasuries are “safer” due to money-printing, others note corporates can sometimes look safer in practice depending on governance.
  • Inflation (~3.7% YoY) means real returns are modest; some say bonds are “fairly priced” in the sense that excess real profit is competed away.
  • 16% CoreWeave yields are compared to credit-card rates, highlighting how extreme that level is for a corporate borrower.

AI infrastructure, GPUs, and neocloud viability

  • Strong consensus that GPU and RAM supply remains extremely tight; secondary market for enterprise GPUs is “white hot.”
  • Disagreement on what happens if neoclouds fail:
    • Some expect bankrupt assets (GPUs) to flood the market and get snapped up by hobbyists and enterprises.
    • Others think lenders will recapitalize and keep clusters intact, since power and datacenter capacity are the real bottlenecks.
  • Several argue CoreWeave’s business model (capital‑intensive, narrow moat, AI-only revenue, high capex/debt) looks structurally fragile.

Anthropic / OpenAI economics and accounting

  • Intense debate over whether AI inference is genuinely profitable or subsidized:
    • One side cites reported high gross margins on API inference and recent quarters of positive operating cash flow.
    • Critics argue: training costs are enormous and recurring; generous flat‑fee subscriptions massively subsidize token usage; and non‑cancellable compute commitments approach the scale of their revenues or larger.
  • Long back‑and‑forth on accounting concepts:
    • Distinction between gross margin vs operating income vs cash flow.
    • Training treated as R&D (expensed) vs inference compute as COGS.
    • Whether high gross margins on inference can ever recoup escalating training and capex before models obsolete.
  • Disagreement over how much compute is being sold at below‑market rates by cloud partners; acknowledged as opaque and “unclear” from public filings.

Macro, politics, and bubble risk

  • Some see this as early evidence of an “AI credit stress” phase: junk yields rising, VC funding cooling, and large raises getting harder.
  • Others argue demand for tokens and compute remains enormous; the core issue is financing and power/datacenter constraints, not lack of usage.
  • Political headwinds: local voters and some politicians are increasingly hostile to datacenter build‑out (power, water, subsidies), which could keep compute scarce even if chip prices fall.
  • Several commenters draw explicit analogies to past bubbles (dot‑com hosting, dark fiber), arguing that even transformative tech can be massively over‑invested in before the economics settle.