Not a Blank Cheque: The $220 Billion AI Debt Wave Meets Bond-Market Fatigue

For most of the AI boom, the money question has been an equity story: valuations, revenue multiples, IPO windows. On August 21, 2026, Reuters moved a story about the other side of the balance sheet — and it may be the more consequential one. The wave of debt issuance funding the AI buildout has grown so large that some of the biggest bond investors in the world are now openly warning about indigestion. The AI capex race, it turns out, is being financed to a significant degree on borrowed money, and the lenders are starting to ask for more.

The numbers behind the fatigue

The headline figure comes from BNP Paribas: AI hyperscalers — the handful of giant technology companies building out AI infrastructure — have issued roughly $220 billion of debt in 2026 through August 10. In the comparable period of 2025, the same group issued $12.5 billion. That is not a gradual increase; it is a seventeen-fold explosion in borrowing, compressed into a single year, and it has happened while overall US investment-grade corporate bond issuance hit a record $1.68 trillion.

Individual deals tell the same story at higher resolution. Amazon’s $25 billion eight-part bond sale in early July — one of the largest corporate bond offerings of the year — had to be priced with roughly 18 to 21 basis points of extra yield on its longest tranches to get the deal done, according to Fortune’s account of the transaction. The 20-year tranche was indicated around 120 basis points over Treasuries, and the 30-year around 130 — levels that analysts noted were roughly double what an equivalent spread would have been a year earlier. Alphabet’s more recent offering was, by all accounts, well received — and it still required a new-issue concession of roughly 10 to 15 basis points relative to its existing bonds.

“You’ve started to see the indigestion show up in tech spreads in particular,” Neil Sutherland, head of US fixed income at Schroders, told Reuters. “It’s not really a credit issue” with highly rated issuers like Amazon and Alphabet, he noted — but the more often the same giant borrowers come to market, the larger the premium investors demand. Per Capital Group’s Karen Choi, technology-sector spreads now stand at around 89 basis points, roughly 9 basis points wider than the average across the investment-grade market. That is the quiet regime change buried in this story: tech debt, which for years traded tighter than the market (investors accepted less yield for it), has flipped to trading wider than the market.

Why spreads, not defaults, are the story

It is worth being precise about what is and isn’t being priced in. Nobody in the Reuters piece questions whether Amazon or Alphabet can pay their debts — these remain among the most cash-generative companies on earth, with strong investment-grade ratings. The strain is showing up as spread: the extra yield investors demand over risk-free Treasuries to hold the paper. Wider spreads signal higher perceived risk or, in this case, simple supply saturation.

George Catrambone, head of fixed income for the Americas at DWS, drew the line sharply: the investment-grade market has undergone a structural shift, with companies that once had modest funding needs and issued mostly shorter-dated debt now taking on much larger borrowings and longer maturities to finance AI spending. “The issuance in January versus August looks different,” he said — fatigue is setting in. “It’s not a blank cheque. If these companies keep tapping the market over and over again, concessions are going to get larger and spreads are going to get wider.”

The most acute stress isn’t at the mega-cap names at all — it’s at Oracle, whose five-year credit default swaps, the market’s real-time gauge of perceived credit risk, hit a near 18-year high of about 203 basis points in July, up from roughly 144 basis points at the start of the year, according to Bloomberg and FT reporting. Oracle has emerged as the sector’s marginal credit risk: its AI-cloud contracts demand enormous upfront capex against revenue recognized over decades, making it the hyperscaler-adjacent name where debt investors’ anxieties concentrate first.

The 2-3% rule and the arithmetic of absorption

Perhaps the most underappreciated point in the Reuters analysis comes from Capital Group’s Choi, and it is about plumbing rather than sentiment. Pension funds and insurance companies — the natural buyers of long-dated investment-grade paper — typically cap exposure to a single issuer at around 2% to 3% of assets. When the same handful of AI companies issues debt again and again, those caps start to bind. “It really depends on how much debt this market will take,” Choi said, noting that clients “don’t want to open a statement and find they own 10% of one bond.”

This is the mechanism that turns a demand story into a supply problem. It isn’t that investors have soured on AI; it’s that the buyer base for corporate credit is structurally concentrated, and front-loaded borrowing from five or six repeat issuers collides with portfolio limits that exist for good fiduciary reasons. Foreign investors, pensions and insurers have absorbed a good deal of the issuance so far — helped by an investment-grade index still yielding around 5.4%, in line with long-term averages — but each incremental deal eats into that capacity.

Why the debt market’s mood matters for AI

The scale of what’s being financed is the context. Combined capital expenditure from Alphabet, Microsoft, Meta and Amazon is expected to approach $700 billion in 2026 — Amazon alone around $200 billion, Alphabet in the $175-185 billion range, Meta $115-135 billion. Free cash flow, once the pride of big tech, has already turned negative at several hyperscalers. What equity markets cheer as “AI investment,” credit markets increasingly see as leverage, and the two views are converging on the same balance sheets.

The transmission channel matters because debt is the marginal source of AI funding. If concessions keep widening, every future data center gets financed at a higher hurdle rate — which means either slower buildout, thinner returns on AI capex, or more expensive AI products downstream. A credit market that demands 120 basis points where it once demanded 60 is, in effect, voting on the economics of the AI buildout more honestly than any earnings call.

What to watch

Three signals will tell you whether this is indigestion or something worse. First, new-issue concessions on the next mega-deal: if the follow-on offerings from the hyperscalers price with 20-plus basis points of concession as standard, fatigue is confirmed. Second, Oracle’s CDS — the sector’s canary — and whether other AI-adjacent credits start trading in sympathy. Third, whether companies pivot financing toward bank loans, private credit, or off-balance-sheet structures (SPVs and joint ventures) to route around public bond-market scrutiny; Reuters’ reporting notes off-balance-sheet vehicles are already part of the picture, and opacity rarely improves outcomes.

None of this means the AI buildout stops. It means the era of free financing for it is ending. The bond market spent fifteen years treating big-tech credit as a safe asset with a tech-company yield; in 2026 it has started pricing that debt like what it has become — the funding instrument of a capital-intensive, cyclical bet on uncertain future revenue. As Catrambone put it, the cheque is still being written. It’s just no longer blank.