AI Debt Binge Meets Cool Bond Market as Long-Term Risk Premiums Hit Investment-Grade Highs
0xBroomberg
Bonds issued by the five largest hyperscalers now carry a yield premium wider than any other investment-grade sector, and long-dated AI paper fell further this week — the bond market is pricing a question equity investors have yet to confront: can AI spending ever pay for itself?
How big a discount is the bond market demanding?
Bonds from Amazon, Google, Meta, Microsoft and Oracle yield roughly 0.6 percentage points more than blue-chip debt of the same rating and maturity.
This means → investors need extra compensation to hold these names — the premium is now the widest across all investment-grade sectors, per Bank of America Global Research.
Bonds with maturities of ten years or longer tied to AI fell again this week, ranking among the worst performers in the investment-grade universe.
Amazon raised $25 billion — why did long-dated tranches struggle?
Amazon sold $25 billion of bonds this week, but demand split sharply by maturity: orders for five-year paper ran roughly 20% higher than for the 30-year tranche.
A SpaceX 30-year bond saw its yield climb from 6.7% at issue to 7.3% — a 60-basis-point move in under two weeks.
In plain terms = investors will lend short-term money to tech giants readily, but the price of lending for thirty years is rising fast.
How is supply overload crushing the long end?
Cross-currency issuance of high-grade AI-related bonds has reached $270 billion year-to-date — nearly double last year's full-year total.
John Lloyd, global multi-asset credit head at Janus Henderson, notes that many portfolios are already loaded with hyperscaler debt; to buy Amazon's new deal, managers had to sell existing holdings first.
His words: "You have to offer a big enough concession to get us to participate."
This reflects a saturation problem — demand exists, but the market cannot absorb this volume of similar paper.
What is the fundamental doubt about AI's long-term payoff?
DoubleLine portfolio manager Mariya Entina frames it simply: buying a 30-year bond requires very stable prospects and clear returns — AI capital spending currently offers neither.
Capital Group's Pramod Atluri agrees: technology evolves so fast that the industry landscape a decade out is impossible to predict, making long-term lending a higher-risk proposition.
In plain terms = no one doubts these companies can repay debt in the near term, but betting they will still profit from AI in thirty years is a wager bond investors refuse to make.
Why are the "natural buyers" of long bonds also hesitating?
Entina adds that the main buyers of long-dated bonds are insurance companies and pension funds — institutions that match long-term liabilities and invest conservatively.
This means → these buyers have the lowest tolerance for uncertainty — and AI's payoff timeline is among the biggest unknowns in markets today.
High rates compound the problem: short-dated Treasuries already offer attractive yields, so there is little incentive to reach further out the curve.
What does the bond-equity divergence on AI really tell us?
Short-term borrowing costs remain stable, signalling no concern about near-term solvency for these companies.
But sustained long-end pressure shows bond investors are asking a question equity markets have not seriously answered: when will AI spending translate into quantifiable profit?
This reflects a fundamental split — equities price AI's upside imagination, bonds price AI's repayment capacity — and the two markets are reaching very different conclusions.
Content is for reference only, not financial advice.