The AI trade now runs on borrowed money, and the lenders are repricing it.
The most consequential development in the data this week is that one sector's financing decisions have become visible across three vectors at once. Bloomberg, reporting on the wave of AI-related bond issuance, captured the concern in a single clause: a disappointment around AI capital spending would produce a reaction that "won't be concentrated in just one market." That transmission has started. The Investment Grade Options Adjusted Spread signal reached 27 this week, its widest reading in seventy-six days, after a ten-point single-session move against an average daily move of roughly one point across the window. The session was Monday, July 27, when the S&P 500 closed within a point of Friday's close of 7,412 on news of a pause in Middle East hostilities, and oil retreated. It landed on a market with little cushion. The Shiller CAPE signal score sits at near peaks of 93 and investors remain highly levered as FINRA margin debt levels reached $1.5T in June. So while equity market valuations are historically expensive and boosted by leverage as the starting condition, the decline in the S&P 500 Breadth signal to 60 from 82 a month ago would suggest that the closing of the equal-weighted index on the cap-weighted index is healthy participation. Rather, it's attributable to the mechanical result of megacap technology being sold while the rest of the index holds, and the indicator sits at its best level precisely because the leadership is breaking.
What that leadership was selling into was a bond market. Morgan Stanley forecasts roughly $570 billion of AI-related borrowing globally in 2026, spanning bonds and loans across every rating tier, with about $236 billion priced by May 31 at four times the year-earlier pace. The US investment grade bond market faces an increase of its own on a different measure, with net issuance running toward $1 trillion this year against roughly $625 billion in 2025, on Apollo's estimates. Absorption is where the strain shows. Apollo Chief Economist Torsten Slok tracks the cover ratio on new deals, the dollars of investor orders received for every dollar of bonds sold, and reports it falling for hyperscalers from nearly five times in February to below two in July, evidence that investors "may need wider spreads to absorb additional hyperscaler supply." Amazon's $25 billion offering this month drew orders at 2.5 times the amount available against 3.2 times in March, and priced with eighteen to twenty-one basis points of extra yield on its longest maturities. JPMorgan strategists including Nathaniel Rosenbaum, in a note reported by Bloomberg, called the resulting widening the high-grade market working to "rationally price in an accelerating pace of issuance." Goldman Sachs and Barclays reached similar conclusions the same week, and the strain has reached past the hyperscalers, with data-center operator Prime Data Centers postponing a planned bond.
The cost of that debt has two parts, and both rose. One part is the yield on a Treasury of matching maturity, the return available without taking corporate credit risk at all. The other is the spread, the additional compensation investors require to hold a company's paper instead. A borrower pays the sum. The Treasury component climbed from a weekly average near 4.44 percent at the end of June through five consecutive higher sessions from July 17 to July 23 that printed 4.55, 4.60, 4.63, 4.67 and 4.71 percent (Federal Reserve data H.15 release). The spread component reached its window high across the same weeks. Option-adjusted spread measures only the compensation for credit risk, stripped of the Treasury yield underneath it, so a widening spread sits on top of the rate move rather than reflecting it. Monday shows why that distinction matters: the ten-year fell to roughly 4.65 percent while investment grade spreads made their largest single-session move of the window, so corporate bonds underperformed duration-matched Treasuries on a day when the risk-free rate was moving in their favor. With competitive yields for Treasuries rising, the premium for owning stocks instead of bonds has grown thin as the Equity Risk Premium signal reads 91. The consequence reached equity prices this month: Alphabet fell about 6 percent in the session following an earnings report that disclosed its first quarter of negative free cash flow since going public in 2004, largely as a result of AI spending. That closes the loop. Capital expenditure now consumes about 80 percent of operating cash flow across Amazon, Google, Microsoft, Meta and Oracle, (Apollo calculation from Bloomberg data), and the shortfall gets funded with debt that enters the market described above, where both components of its cost have been rising at once.
None of this yet resembles a credit cycle turning, and the signals that would show one are quiet. Large Banks Underperformance reads 13, Regional Bank Drawdown 12, and Leveraged Loan ETF Drawdown 14 after falling thirty points over thirty days, all three at Baseline, with the Bank Stress vector at 34. What the data describes is closer to the opposite of a contraction: a credit expansion absorbing record supply and charging progressively more for it. The pricing runs in a ladder. Investment grade widening tracks hyperscaler issuance. High yield widening tracks the tier below it, where speculative-grade borrowers finance the same buildout. CoreWeave, which rents Nvidia chips by the hour to Microsoft and OpenAI, issued unsecured notes at 9.25 percent, and developers have raised more than $8 billion in high-yield bonds for projects leased to that one tenant, per Bloomberg's data. Demand in that tier has swung hard within a single year. A North Dakota campus that required a 10 percent yield in November sold a further tranche at 7 percent in June, oversubscribed five times. High Yield OAS now sits at a seventy-six-day high, though still at Baseline because the absolute level remains historically unremarkable.
Where the ladder breaks is at the bottom. CCC and Lower Option Adjusted Spread signal rose twenty-two points over thirty days to 88 and sits at Critical, markedly higher than the Investment Grade or High Yield spread signal levels, and AI paper does not price at CCC. Private Credit Stress sits at 96, up thirty-one points, consistent with reporting data center financing has moved toward private credit and off-balance-sheet structures where the ultimate holder is harder to identify. What separates an expansion from a contraction is not the level of spreads but whether new issuance keeps clearing. An expansion growing more expensive still places its paper; a contraction is when deals stop pricing at any spread. The market is still clearing. It is clearing at a price that has moved in one direction.
