There’s a signal that doesn’t show up on the Nasdaq chart, or in earnings per share, or in the headlines about artificial intelligence. It’s in the debt market. While the stock market keeps pricing in how much Big Tech can grow thanks to AI, bond investors are paying more and more to protect themselves against something going wrong. And when stocks and credit start telling different stories, it’s worth paying attention.
What a CDS is
That signal is called a CDS, a Credit Default Swap. It sounds complicated, but the idea is simple: a CDS works, in simplified terms, like a kind of insurance on a company’s debt. The more expensive that insurance gets, the more worried the market is about a possible default, a credit rating downgrade, or even a drop in the price of its bonds.
In late July, Oracle’s 5-year CDS hit 215 basis points. At the start of 2026 it was at 144 — an increase of roughly 49%. In practice, insuring $10 million of Oracle debt now costs $215,000 a year. And it’s not just Oracle: Nvidia’s and Alphabet’s CDS also touched recent highs, around 79 and 67 basis points respectively.
Why this is happening: the bill for AI
For years, Big Tech companies were relatively asset-light businesses: they built software, sold advertising or cloud services, and generated enormous amounts of cash. But that’s changing. Artificial intelligence needs data centers, chips, memory, electricity, cooling systems, land, and very long-term supply contracts.
- Moody’s estimates that capital investment from six major US operators could get close to $785 billion in 2026
- Fitch has warned that a prolonged AI-related correction is becoming a significant credit risk
The problem isn’t spending a lot. The problem shows up when spending grows far faster than the cash being generated — and nobody knows exactly when that return will show up.
The Oracle example and S&P’s downgrade
Oracle announced an investment of roughly $70 billion to expand its infrastructure. Shortly after, Standard & Poor’s downgraded its rating to BBB−: the lowest rung within investment grade, just one notch above bonds considered speculative (so-called “junk bonds”).
S&P pointed to increased business risk: uncertainty over profitability, weaker free cash flow, and leverage that could stay above 4x over the next two years.
Two investors, two different stories
This is where the divergence between stocks and credit shows up:
- The equity investor looks at the upside — new customers, cloud growth, future AI revenue. If the story plays out well, they can multiply their investment
- The bond investor looks at something else — how much is being spent today, how much will need refinancing, whether cash flow will arrive before the debt gets too expensive. The bondholder barely gets their coupon and little more
That’s why bondholders tend to be much less tolerant of uncertainty than shareholders: they don’t have the same upside if everything goes well, so they demand more assurance that everything actually will.
Is this a sign of crisis?
This credit signal doesn’t mean Oracle is about to go bankrupt, or that some supposed AI bubble is about to burst. In fact, the spread across the broader US High Yield bond market was at 2.87 in late July — there’s no widespread fire across the whole credit market, at least not yet. The warning is fairly concentrated in a handful of companies with enormous exposure to AI investment.
On top of that, CDS can also rise simply because investors are hedging their portfolios, because there’s less liquidity, or because a lot of bonds have been issued. It doesn’t always mean someone expects an imminent default.
What institutional investors are watching
Three signals big bondholders track
Whether CDS keep rising even after the company reports good results
Whether new bond issuances need to offer higher and higher yields to find buyers
Whether free cash flow improves before new rating downgrades arrive from the agencies
The contradiction that sums it all up
In Bank of America’s July fund manager survey, 45% named a potential AI bubble as the market’s top tail risk. And at the same time, 82% of those same managers were long semiconductors. Big investors are afraid of the risk, yes — but they’re still in the trade.
Conclusion
This signal isn’t saying artificial intelligence is fake or a bubble. It’s saying that financing it is no longer free, and that the market is starting to demand proof that those hundreds of billions being spent will produce enough cash to justify it. Next time Oracle, Nvidia, Alphabet, Amazon or Meta report earnings, don’t just look at revenue growth or earnings per share — look at what’s happening with their debt too. Sometimes the first crack doesn’t show up in the stock price — it shows up in the price of the insurance. It’s the same principle we already saw in why the stock market and the economy don’t always tell the same story: you need to look past the first layer before drawing conclusions.
