Goldman Sachs: Hyperscale Cloud Providers to Issue $296B in Bonds This Year, Credit Spreads Biased Wider
nashnova research
Google, Amazon, Microsoft, Meta and Oracle have issued $296 billion in bonds so far in 2026 — well past Goldman's full-year forecast of $250 billion — as AI capex turns the credit-supply balance toward wider spreads.
Why are five cloud giants suddenly borrowing this much?
In Q3 2026, the five companies' combined free cash flow — cash left after running the business — turned negative for the first time. Spending on servers, data centres and power contracts now exceeds operating cash generation.
This means → to keep expanding, they have three options: spend reserves, issue equity, or borrow. For these firms, bonds are the cheapest and fastest route.
In under a week from October 29, all five completed 5 deals totalling $67 billion — and not just in dollars. They tapped euro, sterling, yen, Canadian dollar, Swiss franc and Australian dollar markets simultaneously to spread demand across investor bases.
How much capex does this debt cover?
Goldman estimates the $296 billion covers about 36% of 2026 capex, up from the 33% assumed at year-start. Working backward, that implies total 2026 capex of roughly $820 billion — nearly double 2025's ~$450 billion.
In plain terms = for every $100 spent building AI infrastructure, $36 comes from borrowed money, not from cash the businesses earn.
Goldman believes part of this is pull-forward — bonds originally planned for 2027, issued early to lock in historically tight spreads and secure capacity before more AI issuers crowd in.
Will 2027 supply pressure be even heavier?
Goldman projects 2027 AI-related issuance in two buckets: ~$440 billion from hyperscaler direct issuance, plus ~$300 billion in project finance across investment-grade and high-yield markets — roughly $740 billion combined.
In plain terms = this is not a one-off spike. Goldman expects annual issuance to stay at $440–470 billion through 2027–2028.
This reflects a multi-year, sustained cash burn phase for AI infrastructure — credit markets will need to absorb massive new supply year after year.
How large is AI's footprint in the credit market?
Including semiconductors, memory, software, data-centre REITs and project finance, Goldman estimates total global AI-related debt issuance in 2026 exceeds $687 billion. The five hyperscalers' direct issuance is only 43% of that; the other 57% comes from the broader supply chain.
In the US dollar investment-grade market — the highest-rated corporate bond market — AI-related supply is ~$450 billion, or about 22% of total supply. On a net-issuance basis (total minus maturities and redemptions — the portion that actually needs new investor money), AI accounts for ~$375 billion, or 41% of the dollar market's net issuance.
This means → for every $10 investors put into new US dollar bonds, roughly $4 goes to AI-related issuers. AI financing has evolved from a corporate-treasury issue into a structural supply problem for the entire credit market.
Where are spreads heading?
The US dollar IG market has seen 27 mega-deals above $10 billion in 2026, totalling $501 billion. 14 of those — 66% by dollar amount — are AI-related, far surpassing 2025's full-year record of ~12 deals and ~$200 billion.
Larger deals require bigger new-issue concessions to attract enough buyers. Goldman's data: AI deals average ~7.7 basis points of concession versus ~4.4 bp for non-AI deals — nearly double.
Goldman's base case for year-end spreads is still slightly tighter than current levels: IG at 82 bp (currently ~89 bp), high yield at 295 bp (currently ~324 bp). But the bank explicitly flags that between now and early December, risks are biased toward wider spreads — its downside scenario puts IG above the March high of 93 bp and high yield above 335 bp.
What does this mean for the AI competitive landscape?
Goldman identifies sustained access to debt financing as a structural advantage in the AI race — wider and cheaper funding windows let a company lock in GPUs, power and land earlier.
US dollar IG spreads currently sit at roughly the 15th percentile since 2000. This means → spreads have been wider 85% of the time over the past 26 years. In plain terms = spreads are already very tight, and the room to widen far exceeds the room to tighten further.
This reflects a widening divide: differences in credit quality will increasingly translate into different issuance terms — and the gap between cheap and expensive borrowing may determine who can sustain the AI arms race the longest.
市场有风险，内容仅供研究参考，不构成投资建议。