Goldman Sachs Warning: AI Hyperscaler Bond Issuance Saturating Investment-Grade Credit Market
Alina Collins
Goldman Sachs and Morgan Stanley have issued parallel warnings: the five major hyperscalers are on track to issue over $250 billion in bonds in 2026, pushing investment-grade absorption capacity to its limit and forcing capital into private credit.
How much are these companies actually borrowing?
Goldman tracks five hyperscalers — Amazon, Alphabet, Microsoft, Meta, and Oracle. They issued $108 billion globally in 2025; year-to-date 2026 already stands at $194 billion, with the full year expected to exceed $250 billion.
Debt financing as a share of capex keeps climbing: 35% in 2027, the cycle peak, before easing to 25% by 2030.
This means → even after the peak fades, cumulative financing needs from 2026 to 2030 could reach $1.3 trillion. This is not a single deal — it is a multi-year issuance cycle.
How crowded has the investment-grade market become?
AI-related supply as a share of U.S. investment-grade issuance jumped from 1% in 2024 to 18% in 2026. Among bonds with maturities of 15 years or longer, 40% now come from AI-themed issuers.
Amazon is now the single largest duration contributor in the entire investment-grade index — up from 20th place last year. Google leapt from 86th to 18th.
In plain terms = if you own an investment-grade bond fund, your portfolio is already heavily loaded with long-dated AI debt — whether or not you chose that exposure.
Are buyers already voting with their feet?
Goldman credit salesman Zach Ablon described real-time signs of retreat: six months ago, large insurers typically bid over $50 million per ticket on hyperscaler 30-year bonds. That figure has now roughly halved.
Spreads on Goldman's AI-leader credit basket have nearly doubled from a low of 74 basis points. New-issue concessions — previously 2–3 bps — have surged to as high as 20 bps on large deals.
Ablon noted: "Clients face concentration limits on single names, and the AI issuance wave has made that quite tricky." This reflects not a loss of faith in AI itself, but a supply pace that has outrun portfolio managers' capacity to absorb it.
Can these companies not afford to repay?
Goldman estimates the hyperscalers could push net leverage to 2× or gross leverage to 3× and still hold investment-grade ratings — roughly $2 trillion in additional debt capacity.
The bottleneck is not the issuers' balance sheets — it is the public market's ability to absorb. 2026 operating cash flow ($778 billion) still exceeds capex ($750 billion), but the gap has narrowed sharply.
This means → these companies *can* borrow; the problem is the market *can't keep buying*. Management teams broadly acknowledge a lag of several months to two years between spending and monetization. Whether public markets can digest this unprecedented multi-year cycle is the credit market's defining stress test for years ahead.
Where will the money flow instead?
Including data-center, chip-related, and project-finance vehicles, total AI-linked bond issuance this year is approaching $500 billion. Hyperscalers account for only 40% of that.
Goldman expects capital to spill into private credit, project finance, and regional bond markets. Morgan Stanley's June report flagged $77 billion in AI-related supply that month alone — the busiest single month of the year.
In plain terms = the public-market "pipe" is nearly full; the overflow has to run through private credit and project finance "side channels." For ordinary investors, that means more AI credit risk will sink into opaque private markets rather than sit in publicly trackable bond indices.
Content is for reference only, not financial advice.