Borrowed Compute: Hyperscaler Debt Is Repeating Equity's AI Concentration Problem

Amazon, Alphabet, Meta, Microsoft and Oracle are increasingly funding the AI buildout with corporate bonds instead of cash flow. Credit markets are now showing the same concentration risk that has been flagged in AI equities for a year, just priced in spreads instead of multiples.

Amazon, Alphabet, Meta, Microsoft and Oracle, the hyperscalers building the data centers the AI industry runs on, are increasingly paying for that buildout with borrowed money rather than cash flow. Goldman Sachs estimates roughly $500 billion in AI-related debt has been issued so far in 2026, with hyperscalers accounting for about $200 billion of it, and expects more than $1 trillion in additional hyperscaler debt by 2030. The bank has already raised its own 2027 forecast once, to around $420 billion in hyperscaler issuance for that year alone. None of these companies is short of cash. The question worth asking is why they are borrowing anyway, and what that is starting to do to the market absorbing it.

Record Deals, One After Another

The individual transactions have been landmark events in their own right. Amazon's debut euro bond in March, €14.5 billion across eight tranches, was the largest-ever inaugural corporate issuance in euros by any company. Alphabet followed in February with its first-ever sterling deal, £5.5 billion including a 100-year tranche, the first century bond issued by a technology company since Motorola in 1997. These are not distressed companies reaching for unusual structures out of need. They are the opposite: investment-grade giants able to borrow in almost any currency, at almost any maturity, because investors want the paper. The deals keep getting bigger because the capital being asked for keeps getting bigger.

Why Spreads Are Widening Anyway

Spreads on hyperscaler bonds have widened through the year even as issuance has grown, and investors are explicit about why. It is not, for the most part, a credit-quality concern. Oracle is the one name discussed as a genuine exception, carrying more leverage relative to its cash generation than the others. For the rest, balance sheets are still seen as strong. What investors are pricing instead is capex uncertainty: every quarter, these companies revise their spending plans upward, which means the bond market cannot reliably size how much more debt is still coming. That uncertainty itself commands a premium, independent of whether any single issuer's credit actually deteriorates.

Is AI Debt Crowding Out the Treasury Market

There is a live, acknowledged debate over how much of this is spilling into government borrowing costs. Fed Chair Kevin Warsh has said hyperscaler borrowing is a direct contributor to rising costs across the board, framing it as real competition for capital, and Treasury Secretary Scott Bessent has pointed to the same dynamic. The 30-year Treasury yield hit a 19-year high in August, and the Treasury responded by doubling its long-bond buyback program starting September 9. The counterargument is a buyer-base one: the largest holders of Treasuries, reserve managers, insurers, and liability-driven investors, are not discretionary allocators toggling between government and corporate bonds, they have a Treasury-shaped obligation to fill regardless of what hyperscalers are offering elsewhere. Wall Street has not settled which effect dominates. What is not in dispute is that the two markets are now large enough relative to each other that the question has to be asked at all, which was not true even two years ago.

The Concentration Problem Credit Markets Share With Equities

The more structurally important finding is not about any single bond sale but about how narrow the pool issuing them has become. Research from JPMorgan and KKR found that 31 companies account for more than $500 billion in AI-related bonds outstanding, with five issuers making up more than half of that total. Total AI-related debt outstanding is now around $600 billion, roughly 6% of the entire US investment-grade bond market, a share KKR estimates could eventually reach 20%. KKR's own comparison is the useful one here: this is the credit-market version of the concentration already well documented in equities, where a handful of AI-linked names have come to dominate index weight. The Bank of England's July Financial Stability Report made the same point from the regulatory side, flagging rising AI-related debt issuance as a growing source of capital-markets exposure to AI sentiment specifically, and warning of a possible sharp correction if confidence in AI or in the rate outlook sours. Morgan Stanley separately put global AI-related debt issuance at roughly $450 billion, double 2025's pace.

What This Means for a Portfolio That Thinks Bonds and Stocks Are Separate Risks

An allocator who has already absorbed the lesson that AI exposure in equities is more concentrated than it looks, through a handful of mega-cap names carrying an outsized share of index weight, has not necessarily absorbed the same lesson in credit. A fixed income sleeve built around investment-grade exposure, diversified by issuer count and sector on paper, can still be carrying a meaningfully AI-concentrated book if five names account for most of its AI-adjacent paper and that paper's spread behavior is driven by one shared variable: how much capex guidance keeps rising. That is the same structural pattern covered in an earlier piece on this page about circular financing running between SoftBank, OpenAI, Nvidia and the hyperscalers, just one layer further into the capital structure, and it interacts directly with the rate backdrop described in a separate piece on what is and is not actually fixed by a Fed rate cut. A bond portfolio does not have to hold a single AI company's equity to be exposed to the AI trade. It only has to hold the debt that trade is now issuing at record pace.

The honest version of diversification here requires checking what is actually driving the spread on a given piece of investment-grade paper, not just which sector or rating bucket it sits in. Five issuers, one shared capex narrative, and a market now large enough to influence Treasury yields is not a diversified credit allocation. It is the same bet as the equity trade, priced in basis points instead of multiples.