The Bond Market Now Trusts the Chipmakers More Than Their Customers
Bond traders have started pricing TSMC and SK Hynix as safer credits than Meta and Amazon — while the stock market keeps betting the buyers' spending pays off first.
TSMC's bonds now yield less than Meta's. SK Hynix's yield sits just nine basis points above Amazon's — a basis point is one-hundredth of a percentage point, so the gap is a rounding error [1]. The companies that fabricate the chips are now cheaper to lend to than the companies with the deepest pockets on earth buying them. The hierarchy is running in reverse. The credit market has been absorbing a flood of borrowing from those buyers. Hyperscaler debt has gone from 1 to 3 percent of investment-grade supply — the bonds rated safe enough for most funds to hold — to 9 percent in three years [2]. Goldman Sachs is telling clients to move down in credit quality, away from the top-rated names, because the AI borrowing has crowded the top rungs [3]. Roughly $100 billion of investment-grade bonds are already trading at spreads — the extra yield over safe government debt — wider than the junk curve [4], and S&P cut Oracle to BBB-, the last rung before junk, over its AI spending [5]. AllianceBernstein's finding is the clarifying beat. The "emerging markets are outperforming" story is really an AI-supply-chain story: TSMC, Samsung, and SK Hynix make up nearly a third of the MSCI emerging-markets index and drove 60 percent of its second-quarter returns [6]. The bond market isn't saying emerging markets are safe. It's saying the companies that make the chips are safer credits than the companies buying them. The stock market is pricing the opposite bet. Credit default swaps — insurance against a borrower defaulting — have widened for the big AI names, yet Nvidia and Oracle shares have stayed elevated [7]. Risk premiums on the lowest-rated U.S. corporate bonds hit their worst levels in a year while the Russell 2000 and the S&P 500 High Beta index sat at or near record highs [8]. One market is pricing a risk the other hasn't noticed. The equity case is not a strawman. JPMorgan argues the market can absorb the issuance, and that the six largest hyperscalers still carry less leverage than the broader investment-grade index [9]. Amazon's record $53.8 billion sale in March drew $126 billion in demand [10]. But by July, Amazon's $25 billion deal drew the weakest demand for a hyperscaler bond since October 2025, and Bank of America put it plainly.
Investors are pushing back. — Bank of America
The two markets are counting down to different clocks. Trillions in take-or-pay compute contracts are in a teaser period — costs deferred, payments not yet due. In 2027 and 2028 those contracts flip from booked backlog to actual billing [11]. OpenAI's $1.2 trillion in commitments will have to meet real revenue, and Oracle's remaining performance obligations grew 363 percent in a single year [11]. The bond market has already started its clock on that date. The equity market is still pricing as if the clock doesn't exist. The disagreement is resolvable, and the resolution date is on the schedule.
- 1. US Tech AI Spending Shifts Global Bond Risk Hierarchy
- 2. Hyperscalers to Reach 9% of Investment-Grade Bond Supply
- 3. Goldman Sachs Advises Shift Toward Lower-Rated Corporate Debt
- 4. Investment Grade Bonds Face Potential Wave of Fallen Angels
- 5. AI Infrastructure Spending Pressures Big Tech Balance Sheets
- 6. AllianceBernstein Warns of AI Concentration in Emerging Markets
- 7. AI Tech Credit Risk Rises as CDS Spreads Widen
- 8. U.S. Credit and Equity Markets Show Growing Divergence
- 9. AI Hyperscalers Issue Billions in Debt to Fund Infrastructure
- 10. Amazon.com Inc. Raises Record $53.8 Billion for AI Infrastructure
- 11. AI Credit Cycle Risks Compare to 2008 Subprime Crisis