Goldman Sachs: AI giants’ bond issuance exceeds full-year forecast as off-balance-sheet commitments increase funding needs

Why has debt issuance by AI giants exceeded expectations?

In the third quarter of 2026, the five largest cloud computing companies’ combined free cash flow turned negative: capital expenditure exceeded operating cash flow, increasing the role of debt financing in AI infrastructure development.
As of 2 November, Alphabet, Amazon, Microsoft, Meta and Oracle had issued a combined 271 billion dollars in debt during the year, exceeding Goldman Sachs’ earlier full-year forecast of 250 billion dollars; some of the issuance brought forward financing originally planned for 2027.
Funding needs extend beyond the five companies. Goldman Sachs estimates that total issuance in the global AI-related syndicated credit market exceeded 657 billion dollars over the same period, with direct issuance by the five giants accounting for 41%. As the financing cycle lengthens, investors’ overall exposure to AI and market capacity will affect future supply.

The five giants have a combined 1.7 trillion dollars in operating and finance lease commitments, more than 75% of which have not yet commenced; under lease accounting rules, the uncommenced portion has not yet been recorded as a liability on balance sheets.
Based on company filings, Goldman Sachs estimates that the five companies have another 1.9 trillion dollars in non-lease, unconditional purchase commitments, of which 685 billion dollars are scheduled for payment by the end of 2027. The commitments cover cloud infrastructure components, servers and networks, power and data center construction, among other items.
Some commitments are already included in consensus capital expenditure forecasts, but they still represent future uses of cash. Goldman Sachs believes faster monetization of AI investments could reduce debt issuance needs; if monetization is delayed and purchase commitments continue to grow, funding needs could rise, depending also on market conditions.

Rating agencies are paying closer attention to off-balance-sheet obligations. In late October, S&P lowered its rating outlooks for Broadcom and Meta, citing liabilities backed by the residual value of AI computing capacity contracts and rising leverage from leased data centers, respectively.
Goldman Sachs believes evidence of monetization of AI investments by 2028 is key to easing downward rating pressure across the ecosystem. However, most relevant companies currently have debt ratings at the high end of investment grade, and the bank believes it is highly unlikely that their access to debt market financing would be restricted even if most were downgraded by several notches.

Goldman Sachs expects private capital to play a larger role in AI financing if AI investments generate insufficient returns in the near term, as the multiyear financing cycle continues and capacity in the traditional syndicated credit market is gradually used up; individual issuers’ access to capital markets will still depend on market sentiment.
According to Preqin, as of March 2026, global private markets held approximately 4.6 trillion dollars in uninvested capital across strategies including private equity, private credit, real estate and infrastructure; this is a pool of capital spanning multiple strategies. Since the start of 2025, private markets have completed more than 225 billion dollars in data center financing, excluding chip and GPU financing.