The growing jitters over hyperscaler debt
Add Axios as your preferred source to
see more of our stories on Google.


Bond market investors are getting slightly twitchy over the scale of borrowing and spending on data centers.
Why it matters: Markets run on confidence. When investors feel sure, they lend and invest relatively cheaply and freely — sometimes too freely.
- When they're worried, borrowers face higher borrowing costs and more difficulty getting hold of capital.
The latest: Recently, a bond market credit gauge associated with Oracle hit a record high, as traders tried to grok how the arrival of low-cost, open-source Chinese AI models might change the potential profitability of providing computing capacity.
- Five-year credit default swaps on Oracle — a kind of insurance that investors can buy to protect them against a company defaulting on its debt — jumped to 212 basis points (or 2.12 percentage points) for every $100,000 of Oracle debt to be insured.
- That means it costs about $212,000 a year to insure $10 million of Oracle bonds against default.
Zoom out: It's not just Oracle.
- Even tech giants considered more creditworthy have seen their CDS creep higher, suggesting a relatively small but growing concern about whether they've bitten off more than they can financially chew with their capex blitz.


Case in point: Apple is the exception that proves the rule.
- Its CDS price is the lowest among its tech titan peers, a reflection of the fact that Wall Street has learned to love its strategic approach to AI, which rests on its dominance in devices and not spending hundreds of billions of dollars.
Reality check: In the grand scheme of things, credit worries about the giant tech companies are relatively small. But they're important to watch.
- The AI building boom is now highly dependent on the willingness of financial markets to fund the spending binge.
What they're saying: "Credit metrics are still very strong for most of these companies," Moody's Ratings bond market analysts wrote in a report last week. "But a material shift in the structure of their balance sheets is becoming evident. While these are among the most cash-rich companies in history, the current level of spending has prompted significant borrowing."


State of play: Through July 22, five companies — Alphabet, Amazon, Meta, Microsoft and Oracle — had raised nearly $302 billion in the markets by selling equity (creating new shares of stock) and debt (issuing bonds), according to data from S&P Global Market Intelligence.
- A separate report from Goldman Sachs looked at AI-related issuance in global corporate bond and loan markets, finding "$489 billion of AI-related supply so far this year — already well above our full-year 2025 estimate of $322 billion."
Yes, but: Oracle's CDS rise suggests that at least some investors may be thinking a bit harder about the risks of such investments as new information on the scale and potential payoff from the boom emerges daily.
- Nikkei published a story last week spotlighting that the major hyperscalers may have some $1.65 trillion in what the Japanese publication called "hidden debts."
- These are essentially off-balance-sheet obligations — often certain lease commitments — which, under prevailing accounting rules, can be kept off a balance sheet.
- Alphabet, meanwhile, rattled some investors by reporting its first quarter of negative free cash flow since going public in 2004, largely as a result of its rampant AI spending.
What's next: Companies borrowing for AI infrastructure are likely to face rising interest costs.
- In part, that's because of the credit concerns.
- But even if investors had zero concern about the financial footing of these companies, borrowing costs would still be rising because of the recent increase in U.S. Treasury bond yields.
The bottom line: AI buildout costs are already massive and clearly rising. Now, borrowing costs are likely to be higher as well.
- At the same time, cheap and effective Chinese AI offerings make the ultimate payoff on these bets increasingly uncertain.
- That sounds like a recipe for some market "excitement" at some point.
