I’ve spent more than 15 years staring at bond and credit-derivative screens, and there’s one rule that never fails: when a company’s stock and its credit start telling different stories, believe the credit. This week that rule got tested with Oracle, and almost nobody in the headlines noticed.
What’s actually happening with AI’s debt
While the Nasdaq hit yet another record high —powered, once again, by enthusiasm for Nvidia, Oracle, and the rest of the AI infrastructure buildout—, the cost of insuring Oracle’s debt (its five-year CDS) jumped to 2.03 percentage points, the highest level since the 2008 financial crisis. It started the year at 144 basis points (KuCoin). S&P already cut the company’s rating to BBB-, one notch above junk, after Oracle reported a negative free cash flow of $23.7 billion against roughly $130 billion in debt (TechSpot). Oracle isn’t an isolated case: total default protection on the biggest AI tech names hit a record $12.5 billion, up 500% since mid-2025.
What’s real: AI demand isn’t an illusion
Demand for AI infrastructure —data centers, chips, power— is genuine and massive. That’s not an illusion. Companies like Amazon, Google, Meta, Microsoft, and Oracle have issued more than $200 billion in bonds this year just to fund that build-out. The problem isn’t that AI won’t create value. The problem is how that bet is being financed.
What nobody’s telling you
Much of that financing was designed for a world where the Fed kept cutting rates. Instead, in September the Fed did the opposite: it hiked for the first time since 2023, to 3.75%-4.00%, because inflation —pushed higher by an energy shock tied to the Iran conflict— refuses to cool off. That reversal makes exactly the kind of debt AI hyperscalers need to keep issuing more expensive. When Amazon priced $15 billion in bonds recently, demand fell 40% once pricing was set —double the usual drop-off for a deal that size. That’s the kind of signal a stock chart will never show you.
Jamie Dimon, who’s lived through several of these cycles from JPMorgan’s chair, put it better than any analyst could: “There’s a lot of exuberance out there. But it was in 1972, 1986, 2000, 2007. That doesn’t give me comfort.” He didn’t call AI a bubble —he actually thinks it isn’t one—, but he did warn it could be the surprise stress point of the next credit cycle.
The real opportunity — and the question I’ll leave you with
It’s not “sell everything AI-related,” and it’s not “ignore the noise and keep buying records” either. It’s simpler, and more boring: if you have exposure —direct, through funds, or indirectly through your pension— stop watching only the stock price and start watching the cost of that same company’s debt. Credit moves before the stock does, not after. And if you’re a CFO or treasurer at a Colombian company weighing dollar-denominated financing abroad, this is the worst moment to assume conditions from six months ago still hold: the Colombian peso has had one of its best years against the dollar among emerging-market currencies, but that strength also depends on global capital flows that can reverse quickly if U.S. credit stress stops being an analyst’s footnote and becomes an actual correction.
This is exactly what the CRITERIO Method is about: not reacting to the stock-market headline, but reasoning from the fundamentals the headline hasn’t caught up to yet. The question I’ll leave you with isn’t whether AI will reshape the economy —it already is—, but this: are you measuring your exposure to that bet by the stock price, or by the real price of the debt holding it up?
If you want to keep following this story —the credit, not just the stock— every day, I cover it in the ViSFI Daily Briefing.