J.P. Morgan's OpenAI loan is strange (marketunpack.com)

In October OpenAI secured a $4 billion revolving credit facility arranged by J.P. Morgan and other banks at roughly SOFR+100bp — about a 5% interest rate. That looks odd because OpenAI is unprofitable and high-risk: a simple expected-value thought experiment shows why. Using a toy investor model (90% failure, 9% 10x, 1% 100x) a $1,000 equity bet has EV ≈ +$900, but the same $1,000 lent at 5% yields EV ≈ −$895 for the lender. Lending only breaks even at 5% if the borrower's chance of survival exceeds ~95% — far more optimistic than public signals for OpenAI today. Market signals back this puzzlement: the 3‑month Treasury yield was ~3.94%, so OpenAI’s effective default spread is only ≈1% — consistent with A−/BBB corporates — yet OpenAI’s interest coverage is negative and Reuters estimates ~$3.6B revenue this year with >$5B losses. The likely explanation is a structural backstop: Microsoft owns ~28% of OpenAI and would have strong incentive and capacity to avoid a creditor seizure, so banks are effectively lending against Microsoft’s support rather than pure OpenAI cashflows. For the AI/ML community this matters because financing terms for strategically important but unprofitable AI firms can reflect corporate sponsorship and systemic importance, not conventional credit fundamentals, changing how risk and runway are priced.