Moody's projects the six companies' combined capital expenditures will reach $785 billion in 2026 and approach $1 trillion in 2027. CNBC reported the update on July 24, five months after Moody's first flagged the issue in February with smaller figures: $969 billion in total future lease commitments, of which $662 billion had not yet commenced. The growth in both numbers in five months shows how quickly lease financing is scaling alongside the buildout itself, a pace that fits the broader pattern of capital finding new structures to reach infrastructure projects even as traditional financing channels get more selective. An increasing share of AI infrastructure is being financed through long-term lease structures alongside traditional debt and equity, not replacing them, and because a lease that has not yet commenced is not a current liability under U.S. GAAP, the bulk of that $1.2 trillion sits outside the standard debt metrics investors and lenders use to assess these companies. Moody's treats the obligations as debt-equivalent commitments the companies will eventually have to pay regardless.
Why the Leases Don't Show Up as Debt
The mechanism is rooted in how AI infrastructure differs from the data centers hyperscalers leased a decade ago. U.S. data center leases historically ran 10 to 15 years. The semiconductors inside a modern AI facility have a useful life closer to four to six years, so hyperscalers are negotiating shorter initial lease terms with renewal options instead, according to Moody's analysts David Gonzales and Alastair Drake. Under GAAP, a lease's residual value guarantee only has to be recorded as a liability once payment becomes probable, and most of these guarantees haven't crossed that threshold yet. That is not a workaround, Gonzales told Fortune: "It's not as if these hyperscalers have avoided a liability through structuring. More accurately, they have not yet received the services to trigger this liability at this time, but they will."
One Company's Disclosure Nearly Doubled in a Single Quarter
The pace of the shift shows up in the companies' own SEC filings. Alphabet disclosed in its second-quarter 2025 filing that it had entered leases for data centers not yet commenced with future payments of $23.9 billion, not recorded on its balance sheet. By the third quarter, that figure had grown to $42.6 billion, with lease terms extending between 2025 and 2031 and noncancelable periods of one to 25 years. Meta's Louisiana data center, tied to a $27 billion private financing arrangement that has already drawn state-level scrutiny over ratepayer exposure, carries a reported $28 billion residual value guarantee that similarly does not appear on Meta's balance sheet.
Moody's is not arguing that these companies are overleveraged today. Amazon, Alphabet, Meta, Microsoft, Oracle and CoreWeave continue to hold some of the strongest investment-grade credit profiles in the market. What is changing is how those balance sheets are evaluated. As AI infrastructure commitments move from future obligations to active leases over the coming years, rating agencies are increasingly looking beyond reported debt to assess the cash commitments companies have already made, the same shift already underway in how capital treats power, permitting, and physical constraints as a single underwriting question rather than separate line items. For investors, lenders, and finance leaders, the implication is straightforward: reported leverage alone no longer captures the full scale of AI infrastructure obligations.
