On September 14, 2026, Goldman Sachs and the Wall Street Journal reported roughly $500 billion in AI-related debt issued this year and nearly $3 trillion in debt at the nine biggest tech companies, most of it off balance sheet. Reporting since then has moved to the borrowing itself: hyperscalers could raise around $300 billion in the investment-grade market this year, and force majeure has been implied in a couple of recent instances. Global credit spreads have since hit a 17-month high, with special purpose vehicles backed by Broadcom and Nvidia concentrating risk in about 10 companies. What remains open is how much sits off the books and what the terms mean for the companies carrying it.