Tech companies are flooding bond markets to raise billions of dollars to build the infrastructure needed to power artificial intelligence.
U.S. companies have issued more than $200 billion worth of investment grade (IG) corporate bonds this year to fund AI-related infrastructure projects — about 13% of total IG issuance, as of end-October.
AI debt is “reshaping credit markets,” Janus Henderson analysts wrote last week.
Five companies account for most of the borrowing: Amazon, Google, Meta, Microsoft, and Oracle. These firms — known as hyperscalers — are building the data centers needed to power AI models.
Collectively, they’ve raised $121 billion of IG debt this year, Bank of America analysts wrote in a November 17 note. Around $75 billion of this was issued in September and October alone — amid two interest rate cuts — which is more than double the sector’s average annual issuance during the prior decade, BofA data shows. The borrowing is expected to push total U.S. corporate bond issuance to a record $1.8 trillion next year, according to JP Morgan.
Yet, concerns are building that hyperscalers are spending too much. Magnificent Seven stocks have fallen about 7% in November, according to an exchange-traded fund that tracks their performance.
Now, with debt-financed data centers thrown the mix, frankly, bond markets aren’t sure how to react. Hyperscaler auctions started off oversubscribed. Cut to Monday, however, when Amazon sold of $15 billion of notes, and orders fell an unusually steep 40% once pricing was revealed. That’s double the typical dropoff rate.
But should the dreaded AI bubble exist, will it burst over debt markets too?
Spreads have widened — a sign bonds are perceived to be riskier, as investors are demanding an additional yield. In September, amid the deluge of new debt, bonds issued by Amazon, Microsoft, Meta, and Alphabet saw their spread over Treasuries widen towards 80 basis points from around 50, according to BofA data.
But Oracle appears to be in league of its own. The spread on its bonds have widened by 48 basis points since September. Its $3.25 billion, 5-year bond, is trading 104 basis points over Treasuries — doubling since September 15.
Demand for credit default swaps (CDS) — insurance-like contracts which protect investors against a company defaulting — on Oracle bonds has also surged, widening spreads and making it more expensive for investors to insure their debt. The spread on Oracle’s five-year CDS has more than doubled since September, according to ICE Data Services, to the highest cost since 2023.
But analysts say widening spreads doesn’t mean that hyperscaler bonds are risky assets.
Robert Schiffman, senior credit analyst at Bloomberg Intelligence, refers to hyperscalers as the “Mount Rushmore” of credits with “massive financial flexibility” to add hundreds of billions of dollars of debt due to their large balance sheets and historically tight margins. The recent widening in spreads reflects minor weakness in the technicals — a surge in supply — rather than in the fundamentals of the companies themselves, says Schiffman.
Meta, Microsoft, Alphabet, and Amazon have credit ratings ranging from AA- to AAA. In their third quarter earnings for 2025, these companies’ trailing twelve-month cash flow from operating activities ranged from $30 billion to almost $340 billion.
So, while Big Tech’s AI-related capex has been jaw-dropping — hitting $113.4 billion in the third quarter, up 75% year-over-year — cash flow has remained strong.
However, Oracle — the hyperscaler that’s been giving AI credit somewhat of a bad name — tells a different story. With a BBB credit rating, the cloud company has more than $104 billion in outstanding debt, while its free cash flow turned negative in 2025, falling to its lowest level in over two decades. But Oracle aside, no other hyperscaler has had its credit ratings be downgraded or outlooks change to negative.
“Some of the individual numbers may be staggering (think, for example, Oracle). But we’re not too concerned. For example, well over half of U.S. tech IG issuance in H2 has come from issuers rated A and above,” HSBC multi-asset analysts wrote in an investor note on Monday. “Aggregate capex to free cash flow is nowhere near the levels of the late 1990s yet,” they said in reference to the dot-com bubble that burst in 2000.
What’s more, this year, U.S. IG spreads have touched their narrowest levels in 15 years. So, the market’s ability to absorb these large deals while staying at historically tight spreads is “impressive,” and speaks to the “exceptional fundamentals in this cohort,” Janus Henderson analysts wrote last week.
While cash flow may be robust — for the most part — should bond investors also be sharing Wall Street’s worries about AI’s future returns?
Schiffman thinks not: “The bubble that people are talking about, quite frankly, is equity valuations,”
Proponents of the AI bubble theory contends that investor enthusiasm about the technology’s future capability is driving up stock prices faster than the companies’ current fundamentals: earnings and revenue.
Stocks in hyperscalers Meta, Alphabet and Microsoft are up more than 347%, 230% and 109% since 2023, respectively. But some argue that there’s not yet enough evidence of future revenue to justify these sky-high valuations. J.P. Morgan’s estimates that AI needs $650 billion in annual revenue “to make the math work.”
However, demand for compute is no longer hypothetical.
For example, Nvidia’s data center revenue is now running around $39-41 billion annually, more than doubling year-over-year in some recent quarters. AWS has reached $123 billion in annualized revenue. Citi Research forecasts $780 billion in annual AI revenues by 2030.
“All we’ve heard from everybody is that demand has outstripped supply. If you’re really talking about a bubble, there’s a bubble of demand,” says Schiffman.
Morgan Stanley research estimates a $1.5 trillion funding shortfall for AI data centers by 2028. Tracy Chen, a portfolio manager at Brandywine Global wrote last week that credit markets — including corporate bonds — will be “critical” in plugging this gap.
Looking ahead, Henderson analysts noted that hyperscaler spreads may continue to fluctuate, but only until the “winners” of this supercycle appear, as opposed to signaling a deeper unease about the liability of the bonds themselves.
Leave a Comment
Your email address will not be published. Required fields are marked *