ThinkPatternGet the app
Perspective
BUSINESS · NOV 22, 2026

The bond market priced AI's risk before the stock market did

The $8 trillion selloff wasn't the moment the buildout's risk arrived — it was the stock market finally catching up to a repricing the bond market had been doing all year.

TSMC's bonds now yield less than Meta's. SK Hynix paper due in 2031 trades nine basis points above Amazon's — nine hundredths of a percentage point — and the premium investors once demanded to hold emerging-market corporate debt has shrunk from a historical 30 to 50 basis points to a handful. [1] That price printed on September 10, two months before the $8 trillion stock-market wipeout everyone treated as the event. [2] The bond market had already delivered its verdict; the stock market was still arguing about valuation. The migration began in July. When the retail trade broke — the leveraged funds and speculative crowds it had drawn in — the leverage did not disappear. Analysts flagged it moving instead onto the balance sheets of companies like Alphabet and Nvidia. [3] Those balance sheets were already stretched. By the second quarter, the hyperscalers' capital spending had crossed above their operating cash flow — the cash the businesses generate from running themselves — leaving free cash flow negative. [4] In August, Goldman Sachs counted a trillion dollars of data center lease commitments the hyperscalers had signed but not begun, obligations living in footnotes rather than on the balance sheet, up from $725 billion a month earlier. [5] Their borrowing had reached 30% of all net investment-grade debt issued this year — investment-grade meaning borrowing by companies whose credit is rated safe enough for conservative buyers — on its way toward a trillion dollars. [6] The instruments were built to stay hard to see: 144A bonds, which skip the usual public registration; special purpose vehicles; joint ventures like Meta's $50 billion Hyperion; unrated loans. Nearly $500 billion of AI-related debt was raised through such structures this year. [7] In September, Apollo's chief economist Torsten Slok named what the bond market had begun to price. [8]

What the market is repricing is hyperscaler credit fundamentals, namely a debt-financed AI capex cycle with rising leverage, negative free cash flow and uncertain payback on depreciating assets. — Torsten Slok

The crash was still two months away. By October the credit market was marking individual names. Credit-default swaps — insurance against a borrower failing to pay — widened across Nvidia, Broadcom, Oracle, AMD and Meta. Broadcom's swaps tripled while its shares had fallen only 25%; Oracle's had dropped more than 40%. The bond market was treating these as riskier credit while equity investors still debated whether the stocks were cheap. [9] In November the equity market marked down the same risk — $8 trillion erased in 45 days. [2] The same week, the European Central Bank gave the warning institutional weight: US tech giants lean on more than $4 trillion in opaque off-balance-sheet commitments, an echo of the dotcom and subprime eras, with circular financing in which chipmakers lend to their own customers. [10] Michael Burry put a number on the special-purpose share of that pile: up to $3 trillion of AI debt hidden in special purpose vehicles. [2] By then the risk had left the tickers entirely. Jefferies downgraded whole countries — South Korea to underweight, Taiwan to neutral — citing higher financing costs and weaker data center spending, in a week when AI names made up nearly half the MSCI Emerging Markets Index. [11] The ECB named France the sovereign weak link, debt at 119% of GDP. [10] Even the machines make doubtful collateral: an Indian state lender planning to finance GPU clusters flagged that the chips lose their usefulness in 18 to 24 months — faster than the loans against them pay down. [12] And Aon's chief executive warned that clustering billions of dollars of equipment in a few buildings creates a concentration insurers won't underwrite — which, he said, closes the door to the debt market itself. [13] None of this has shut the window. Goldman Sachs, publishing during the crash week, still forecasts the hyperscalers will finance 35% of next year's capital spending with roughly $440 billion of new debt, after this year's borrowing beat estimates. [14] Strategists at Morgan Stanley, Goldman and Federated Hermes kept 8,700 as their targets for the S&P 500, reading the selloff as froth removal rather than a credit event. [15] The market's own logic is blunt.

the lesson in capital markets, and we've seen this time and time again, is corporates issue when they can, not when they have to. — Goldman Sachs Private Wealth Management

The test has a date. $13.4 billion of data center developers' fixed-rate dollar debt matures by 2027, and refinancing it at October's Treasury yields adds 2.85 percentage points — at least $380 million in added interest before credit spreads widen at all. [16] Either the bond market was early, and those maturities hit a buildout already spending past its own cash flow; or the froth-removal reading holds and the borrowing rolls over on the old terms. 2027 is when it comes due.


Sources
  1. 1. US Tech AI Spending Shifts Global Bond Risk Hierarchy
  2. 2. AI Stock Selloff Wipes $8 Trillion From Markets
  3. 3. AI Trade Leverage Shifts to Major Tech Balance Sheets
  4. 4. AI Spending Surge Shifts Investors Toward Free Cash Flow
  5. 5. Hyperscalers Accumulate $1 Trillion in Data Center Lease Commitments
  6. 6. Hyperscaler Debt Issuance Rises Toward 1 Trillion Dollar Mark
  7. 7. Wall Street Develops Novel Financing for AI Infrastructure Boom
  8. 8. Apollo Global Management Warns of Rising Hyperscaler Debt Risk
  9. 9. AI Giants Face Widening Credit-Default Swap Spreads
  10. 10. European Central Bank Warns of AI Bubble and French Debt
  11. 11. Jefferies Downgrades South Korea and Taiwan Over AI Risks
  12. 12. Union Bank of India Forecasts Shift to AI Infrastructure Debt
  13. 13. Aon CEO Warns AI Infrastructure Risks Threaten Data Center Financing
  14. 14. Goldman Sachs Forecasts $440 Billion AI Debt for 2027
  15. 15. Wall Street Divided as S&P 500 Enters Correction
  16. 16. AI Data Center Developers Face $380 Million Interest Surge

Keep reading in the app

The full perspective, free in the app.