Goldman's public August analysis said the group had already issued $194 billion and expected debt to fund roughly one-third of 2026 capex. A September 17 summary of its latest research says the 2027 direct-issuance forecast rose to $420 billion and AI-related credits reached about 12% of U.S. dollar investment-grade notional.
AI infrastructure financing is already large enough to alter the composition of credit portfolios. In Goldman's public August discussion, the five hyperscalers in its framework had issued $194 billion of debt globally in 2026 by early August, after $108 billion in all of 2025. Goldman expected direct hyperscaler issuance to reach about $250 billion for 2026 and estimated that roughly one-third of capex would be debt-financed. It also tracked nearly $500 billion of AI-related debt across the wider ecosystem, with hyperscalers accounting for about 40%.
Goldman's public analysis focuses on market absorption. Amazon, Alphabet, Meta, Microsoft and Oracle retain substantial investment-grade funding capacity. Credit portfolios face a separate constraint: bondholders receive coupon and spread while their exposure becomes increasingly concentrated in one capital-spending cycle. Duration, issuer limits and market-saturation conventions can therefore affect pricing before borrower solvency becomes a concern.
The forecast has since moved higher. The Financial Times cited a Goldman estimate of about $400 billion of hyperscaler issuance in 2027 on September 16. A September 17 summary of the bank's latest Global Credit Trader says Goldman then raised the figure to $420 billion and expects debt to finance about 35% of 2027 hyperscaler capex. The same summary reports $578 billion of global AI-related debt issuance year to date through September 16 and puts AI-related credits at about 12% of eligible notional in the Bloomberg U.S. dollar investment-grade corporate index.
Supply is the pressure point
Goldman's public work says financing capacity should remain available across investment-grade bonds, private markets, regional debt markets and structured finance. The price of that capital is less certain. A larger share of new issuance tied to the same theme can force investors to accept more concentration, move across issuers and structures, or demand additional spread.
The requirement also extends beyond corporate bonds. Goldman has discussed roughly $300 billion of data-center and semiconductor project-finance issuance in 2027, while private infrastructure, real estate, private credit and structured markets are absorbing additional capital. The same AI investment cycle is therefore appearing across several lender balance sheets and security types.
The September $420 billion figure is an attributed update from a published summary; the underlying September 17 Goldman note is not publicly accessible in the records reviewed here. The primary August record already establishes the core result: hyperscaler debt supply is rising fast enough to create a portfolio-concentration and market-absorption problem even while the issuers themselves retain strong access to capital. The September update indicates that the supply burden is increasing.
Sources
https://www.goldmansachs.com/insights/goldman-sachs-exchanges/how-ai-debt-is-reshaping-the-credit-market https://www.ft.com/content/033b21e7-b5d5-4511-ae01-53edaa4e051e https://finvaulta.com/research/goldman-sachs/an-update-on-ai-related-debt-issuance-2026-09-17 https://www.goldmansachs.com/insights/articles/private-markets-expected-to-have-growing-role-in-data-center-financing