Morgan Stanley Reveals $3T in Hidden AI Liabilities: Purchase Commitments, Unapproved Leases, Delayed Payments

According to Morgan Stanley's latest research, as of end-May 2026, the global artificial intelligence infrastructure investment boom has accumulated over $3 trillion in off-balance-sheet liabilities not captured in traditional financial statements, posing risks comparable to the subprime mortgage crisis.

The three major hidden liabilities include: approximately $1 trillion in long-term purchasing commitments with Nvidia; over $822 billion in signed but unapproved lease obligations; and roughly $110 billion in unpaid capital expenditures embedded in accounts payable. Morgan Stanley estimates that when including finance leases, Microsoft's capital expenditure as a percentage of revenue could jump from 33%/50% to 44%/64% in fiscal 2026/2027, while Oracle's could surge from 76%/115% to 101%/189%.

AI-related debt bond issuance surged 357% year-over-year, reaching $236 billion through May 2026, with Morgan Stanley projecting full-year issuance to exceed $570 billion. The gross leverage ratio of hyperscale cloud providers has climbed from 0.9x in Q3 2025 to 1.8x currently, exceeding energy sector leverage levels.