As U.S. companies raise billions of dollars to fund their AI dreams, they're primarily tapping debt markets. But the burden of debt can hurt stock investors, too.
Nearly 70% of the $456 billion cash raised for AI from the public market in 2026 has come from the investment-grade debt market, according to Bank of America Global Research. At $309 billion, that amount is more than double the $136 billion in AI-related investment-grade debt issued in all of 2025.
Much of that debt supply is coming from just five companies: Alphabet, Amazon.com, Meta Platforms, Microsoft, and Oracle. Collectively known as the hyperscalers, these five companies operate data centers that account for 71% of the world's cumulative AI compute, according to analysis firm Epoch AI. Nvidia has also issued billions in AI-related investment grade debt.
The supply of debt is growing as "demand for compute continues to outstrip supply," resulting in capital expenditure expectations that exceed cash flow, write Morgan Stanley strategists.
So far, bond investors have shown a willingness to keep buying, though they're asking for a bit more yield in return. Hyperscalers' outstanding bonds maturing in 2035 carry an average yield of 5.7%, 1.05 percentage points above a 10-year US Treasury bond. By contrast, a year ago, those bonds were yielding 4.75% on average, at a spread of 0.44 percentage points, according to LSEG data.
Meanwhile, credit default swaps (CDS), which effectively serve as insurance on defaults, have risen. A buyer looking to protect their investment against an Oracle default was paying 2.13% of the face value of the bond annually over a five-year period on Wednesday, in contrast to 0.43% a year ago. Insurance against an Alphabet default would cost 0.56%, versus 0.33% a year ago.
No one seriously thinks Alphabet is going to default- less than a percent of the investment-grade market has defaulted in 26 years-but higher CDS reflects investors' worry about the immediate costs of the AI buildout, and the uncertainty about the investments' ultimate returns.
Some argue that the growing supply of bonds to fund AI is one factor driving a broader global bond selloff that has brought the 30-year U.S. Treasury yield to its highest level since 2007. Either way, it is certainly the case that higher overall yields and higher spreads make tech companies' bonds more attractive in relation to their stocks.
"If I'm getting 7% on my bond investment, I would need at least 10% plus to make equity investment more attractive," PIMCO's CIO of Core Strategies Mohit Mittal said earlier this month. "A lot of things need to go right for hyperscaler equities to deliver a return above 7% annualized for the next 10 years. Whereas a lot of things need to go wrong for debt investors to have poorer returns."
Still, equities remain in demand because companies keep booking strong profits. In the second quarter, about 84% of companies in the S&P 500 index exceeded Wall Street's earnings per share (EPS) expectations, while most hyperscalers have posted solid growth in net income. Google's profits are up 300% compared to the second quarter of 2025, while Microsoft's income is up 31%.
However, given the companies' growing debt loads, earnings may no longer be the best barometer. In fact, according to Manish Kabra, multi-asset strategist at Société Générale, investors should "Watch CDS, Not EPS, for Hyperscalers."
Kabra points out that there has been a strengthening inverse correlation between CDS spreads and hyperscaler stock valuation-meaning that as one rose, the other fell. Over roughly the last five-years, the correlation had been low of -0.29, but taking a smaller sample size starting from October when Oracle's CDS first spiked, the correlation has jumped to -0.49, he found. (The correlation compars the change in 5-year CDS with changes in hyperscaler forward price-earnings ratio; a correlation of -1 denotes a perfectly inverse relationship.)
The correlations have shifted because "the market is now myopically focused on free-cash of hyperscalers and funding needs," says Kabra.
Given that the hyperscalers represent about 20% of the S&P 500 by market value, a problem for hyperscalers is a problem for the stock market as a whole. The importance of the AI debt load can hardly be overstated.
Comments