AI Debt Issuance Already Tops $489 Billion in 2026 as Goldman and Morgan Stanley Flag a Credit Market Splitting in Two

The numbers are bigger than last year's forecast

Goldman Sachs said in 2025 that AI-related debt issuance for 2026 would land around $322 billion. It has already blown past that, hitting $489 billion so far this year, according to Goldman's own analysis. Morgan Stanley projects the full-year global total will reach nearly $570 billion. RD World Online reported the same $489 billion figure had already been reached by early August, meaning the pace has only accelerated since.

The riskier end of the market is growing even faster. Goldman put AI-linked leveraged finance issuance at $88 billion year-to-date in 2026, up from $20 billion over the same stretch in 2025. High-yield AI infrastructure supply hit $40 billion this year, already more than the $12 billion issued in all of 2025, per Goldman's figures as reported by Crypto Briefing.

Leverage has doubled in six months

Aggregate gross leverage among hyperscalers, the ratio of debt to earnings, surged from roughly 0.9x to 1.8x over the past six months, according to the Goldman and Morgan Stanley analyses. On top of that on-balance-sheet debt, the banks estimate roughly $3 trillion in off-balance-sheet liabilities tied to leases and purchase commitments.

Columbia University economist Stijn van Nieuwerburgh, whose paper was published by the Brookings Institution and cited by both the Wall Street Journal and Breitbart, said tech companies often route this borrowing through off-balance-sheet entities involving banks and private-credit firms. These structures come with little public reporting, making the real scale of financial risk hard for outsiders to gauge.

A two-speed credit market

Both banks describe shrinking risk appetite concentrated at the bottom of the credit scale. Investors are crowding into investment-grade paper from top-rated hyperscalers, who can still borrow enormous sums easily. Smaller, more leveraged AI infrastructure players are facing harder questions and higher yields, creating what Goldman and Morgan Stanley both characterize as a bifurcated market.

ZeroHedge framed this more starkly, arguing the "credit party" has already ended, "just not where most people are looking," and cited leveraged finance issuance climbing from $20 billion to $420 billion over three years. That figure is harder to verify against Goldman's own $88 billion year-to-date leveraged finance number reported elsewhere, and it is not clear the two outlets are measuring the identical category over the identical window. The $420 billion figure represents ZeroHedge's own framing rather than a confirmed Goldman data point.

The spending behind the debt

The borrowing is funding a buildout that Van Nieuwerburgh estimates will total $10.3 trillion between 2025 and 2032, averaging 3.6 percent of GDP a year, according to the Wall Street Journal. That would exceed every prior American infrastructure boom by his measure, including railroads (2.24 percent of GDP from 1870-1890) and the interstate highway system (1.13 percent from 1956-1973). Goldman separately pegs 2026 AI investment at 1.9 percent of GDP, while JPMorgan estimates AI infrastructure spending alone will hit around $1 trillion this year, according to CNN. That exceeds annual U.S. military spending.

Five companies are doing most of the heavy lifting. FactSet projects Google, Amazon, Meta, Microsoft and Oracle will spend a combined $4.2 trillion through 2029. Alphabet's debt has climbed to about $100 billion after it raised its 2026 capex outlook to $195 billion to $205 billion. Amazon expects roughly $220 billion in 2026 cash capex, citing rising memory prices. Oracle has racked up about $137 billion in debt and leases, with S&P estimating half of its $638 billion backlog traces to OpenAI's reported $300 billion, five-year cloud deal. OpenAI itself reportedly carried no debt as of March 31, 2026.

Who's getting squeezed

The spending is already crowding out other industries. Commerce Department data show private data center construction spending hit $37 billion through July, up $9 billion from a year earlier, while private construction spending on everything else fell roughly $46 billion below year-ago levels. The Federal Reserve Bank of Richmond has reported data center construction straining regional labor supply.

In Mississippi, a proposed aluminum smelter that would have brought an estimated 1,000 permanent jobs went to Oklahoma instead. A data center near the Vicksburg site under consideration tied up the local electricity supply the smelter needed, according to a person familiar with the decision cited by the Wall Street Journal. Site-selection consultant Didi Caldwell put it bluntly: "It's crowding out manufacturing."

There's an honest counterpoint worth stating: this is private capital making private bets, not a taxpayer bailout. If hyperscalers with real cash flow want to lever up to build data centers, that's their shareholders' risk to take. The harder question, raised by Van Nieuwerburgh himself, is how much of that risk is actually visible to regulators and bondholders given how much sits off-balance-sheet.

CNN separately flagged a second-order concern: economists worry the sheer scale of AI capex, layered on top of an already-tight labor market, is adding to inflation pressure the Federal Reserve has to contend with. Whether the leverage buildup among hyperscalers becomes a genuine systemic risk, or simply a repricing exercise for weaker borrowers, is the open question both Goldman and Morgan Stanley say the rest of 2026 will answer.