NEW YORK (Sept 10): The AI boom has unleashed an avalanche of corporate debt issuance by tech giants, upending some long-held valuation rules. With the debt binge showing no signs of slowing, these anomalies could deepen.

In the past year, roughly $220 billion of debt has been issued by U.S. hyperscalers, including Alphabet, Amazon, Meta, Microsoft and Oracle. These gargantuan cloud computing companies ​operate huge, globally distributed data centers that they've been expanding rapidly to keep up in the AI arms race.

Many financial pundits have recently argued that this new supply of bonds has been ‌so massive that it has contributed to this year’s sharp rise in long-term U.S. interest rates.

That point remains debatable. However, an analysis at the individual bond level suggests this huge volume of borrowing is having one notable impact: it's challenging some of the credit market’s valuation norms.

In a well-behaved market, two bonds with equivalent risk would be expected to carry essentially the same yield and spread — the additional yield that investors demand for holding corporate bonds instead of U.S. Treasuries.

But now that AI is reshaping the debt landscape, this ​rule isn’t always holding up.

Consider Oracle. The scale of its debt issuance has skyrocketed as it has joined the race to capitalize on the AI revolution.

Oracle currently has two bonds that mature in 2065. Both ​are senior unsecured issues with $1 billion outstanding. They each give the issuer the option to redeem them at 100 cents on the dollar, beginning three months prior to ⁠maturity.

These two bonds had the exact same yield, of 8.00%, as of August 21. Their option-adjusted spreads were also almost indistinguishable, at 280 and 281 basis points, respectively. In other words, the market was pricing them as expected.

However, it’s ​a very different story for two other Oracle bonds that mature in the same year — 2055 in this instance — which also feature identical seniority and early redemption provisions.

A bond with a coupon of 4.375% maturing in April 2055 yielded ​7.67% as of August 21, while a bond with a coupon of 5.95% due at the end of September that year yielded 19 basis points more, at 7.86%. In addition, the 5.95% issue’s spread exceeded that of the 4.375% bond by 22 bps.

In this case, there was one difference between the two securities that affected their relative pricing: the size of each issuance.

Oracle's 4.375% bond, which came to market in 2015, was “only” a $1 billion issue, while the 5.95% issue from late 2025 raised $3.5 billion.

As a general rule, if two otherwise ​similar bonds differ materially in their amounts outstanding, the larger one is considered less risky. That’s because it will tend to have a deeper secondary trading market, making it easier for holders to sell it without incurring a ​significant loss, as might happen with a less liquid issue.

So if anything, the larger Oracle issue would be expected to have a lower yield and smaller spread, yet we observe the opposite.

The chart below shows similar anomalies in bonds of two other hyperscalers — ‌Alphabet and Meta ⁠Platforms — and chipmaker Nvidia, which isn't building data centers but is rapidly scaling its own production capacity.

In each pair, the bond floated more recently has a far greater amount outstanding than the older issue, along with a yield and spread that are 15 to 22 bps greater.

In the credit universe, these differences are far from trivial. The gap was twice as big as the spread differential between the average AA-rated and A-rated corporate bonds on the same date, according to ICE Indices.

While this pricing gap was not seen at fellow hyperscalers Microsoft and Amazon, that's because there were no comparable matched pairs of bonds. Where the conditions were present, the pricing anomaly was consistent.

What accounts for this? The ​sheer size of the hyperscalers’ financings appears to make ​them difficult for the market to absorb.

Bonds with $3.5 billion ⁠to $4.0 billion outstanding are very rare. They rank among the top 1.5% in the investment-grade corporate universe. Having so many of these giant issuances arrive in a short period of time is bound to cause market indigestion.

That’s particularly true because institutional investors’ diversification requirements mean they must avoid becoming overly concentrated in a single sector, especially one that faces ​the serious risk that its colossal investments in AI will fail to pay off with adequate returns.

Moreover, with hyperscalers expected to continue making mammoth additions to debt ​supply, these anomalies could intensify.

The net ⁠result is that conventional assumptions about debt valuation have been challenged. Essentially identical risks are now being priced differently, subverting a fundamental assumption held by participants in the $10 trillion investment-grade U.S. corporate bond market.

That matters beyond the trading desk. Credit markets exist to allocate capital efficiently by pricing borrowing costs in line with risk. When that mechanism breaks down, capital can flow to the wrong places, resulting in a suboptimal economy.

Economists have long noted AI's potential to radically restructure the labor market, but ⁠that's not all ​that the powerful new technology is disrupting.

Uploaded by Siow Chen Ming