Goldman Sachs: AI Debt Is Becoming the Defining Supply Risk in Investment-Grade Credit

AI is no longer only an equity-market story. Goldman Sachs’ credit-market message is that AI infrastructure financing has become a major source of new bond supply—and that investors may need to distinguish a temporary improvement in spreads from a lasting resolution of the supply pressure. 24

AI has become a dominant credit-market theme

Goldman raised its forecast for 2026 U.S.-dollar investment-grade gross issuance to $2.3 trillion, from $2.1 trillion, and increased its net-supply forecast to $1 trillion from $850 billion. It attributed a meaningful part of the revision to AI-related financing: such issuers represented about 24% of year-to-date U.S. investment-grade issuance. 4

A separate Goldman trading-desk view described roughly $300 billion of AI-related bond issuance already completed in 2026 and called AI the dominant credit-market theme. That figure should be read as a point-in-time, defined universe rather than a universal measure of all AI-linked borrowing: Goldman Research had earlier estimated nearly $500 billion of AI-related debt issuance in 2026 across a broader global ecosystem that included data-center financing and other AI-linked sectors. 212

The common conclusion is clear: AI buildout is materially changing the volume, composition, and maturity profile of corporate credit supply.

Why Goldman expects another funding wave in 2027

The Goldman trading-desk scenario reported in September projected about $340 billion of 2027 issuance by hyperscale cloud companies and chipmakers, approximately 40% more year over year. The calculation assumes about $930 billion of capital expenditures and debt financing rising from roughly 30% to 37.5% of that spending. 2

The underlying mechanism is straightforward. If capital spending rises faster than internally generated cash flow, issuers must use more external financing. For large cloud platforms and semiconductor companies building data centers, compute capacity, and supporting infrastructure, bond issuance is one of the main ways to close that gap.

Goldman has separately published a broader forecast that hyperscalers could finance 35% of 2027 capital expenditure with debt, implying around $400 billion of global debt issuance. Differences between the $340 billion and $400 billion figures reflect differing scenarios and issuance definitions, but both point to continued heavy borrowing rather than a quick normalization. 67

Why refinancing demand is not the whole story

New borrowing tends to be easier for credit markets to absorb when it mainly refinances bonds that are maturing. In that case, investors receive principal repayments that can be redeployed into new deals.

Goldman’s trading-desk analysis instead estimated that only about $45 billion of relevant hyperscaler and chipmaker debt matures during 2027–28. That means much of the prospective borrowing would be incremental net supply, not replacement debt. 2

That distinction matters because the market needs additional capacity to buy the bonds—through new cash inflows, investor reallocations, or sales of other securities. It can place pressure on credit spreads even when the issuers themselves remain highly rated and financially strong.

Why a fourth-quarter lull may be tactical, not structural

The reported Goldman view was that AI-linked issuance could fall by roughly 50% in the fourth quarter, leaving a relatively light near-term deal calendar. With AI Credit Basket spreads near their widest levels, reduced immediate supply could support a short-term tightening in spreads. 2

But Goldman characterized that possibility as the “eye of the storm,” not evidence that the underlying supply issue has ended. Following a period in which AI-credit spreads had widened by more than 50 basis points, the reported recommendation was to use a rebound to reduce exposure rather than assume a durable recovery. 2

In practical terms, this is a tactical-versus-structural distinction:

• Tactically, fewer imminent deals can improve technical conditions and support spread tightening.
• Structurally, the expected 2027 financing requirement could bring another large wave of bonds to market.

The duration and concentration challenge

The concern is not just the total amount of debt. It is also where that supply sits on the maturity curve and how concentrated the issuers are.

JPMorgan estimated that five major hyperscalers plus Nvidia had issued around $320 billion of debt in 2026, including structures in which companies ultimately support data-center lease obligations. In 10-year-equivalent terms, JPMorgan put the long-duration component at approximately $303 billion, or about 68% of new long-duration U.S. Treasury borrowing for the year. 5

That comparison does not mean corporate AI debt and Treasuries are interchangeable. It does illustrate the potential competition for investors willing to hold long-dated fixed-income assets. As long-duration Treasury supply, interest-rate volatility, and AI-related corporate issuance converge, buyers may demand more yield or wider spreads to absorb the added duration risk.

Can investment-grade markets absorb the debt without further widening?

Possibly—but it is not automatic. Goldman has noted that credit spreads remained very tight and credit volatility very low despite the scale of AI-related borrowing earlier in 2026. 42 Market absorption is more likely if AI issuers’ earnings and operating cash flow validate the capital spending, interest rates are stable or lower, and non-AI corporate supply remains manageable.

Further spread widening becomes more plausible if large deals arrive quickly, Treasury yields rise, investors become less willing to add long-duration exposure, or the expected returns on AI infrastructure take longer to materialize. A supply surge does not by itself imply a financial-stability event; it does, however, create a valuation and market-technical risk that can differentiate even strong issuers.

Keep the long-term forecasts in perspective

Two widely cited AI-financing numbers require careful attribution:

• The forecast of nearly $570 billion in global AI-related debt issuance for 2026 came from Morgan Stanley, not Goldman Sachs. 19
• PwC’s $31.6 trillion estimate is a central-scenario projection for global data-center capital expenditure through 2050, not a committed financing plan or a Goldman forecast. 24

Those forecasts underscore the possible scale of the infrastructure cycle, but they do not determine near-term credit-market outcomes. Goldman chief economist Jan Hatzius has also cautioned that the AI investment boom “will not go on forever” and that some investments could prove unproductive. 26

For credit investors, that is the essential test: debt can be absorbed while the buildout is supported by credible cash flows and demand. If spending shifts from construction to monetization before the economics are proven, the market is likely to become more selective about who can borrow cheaply—and at what maturity.