Big Tech's Off-Balance-Sheet AI Obligations Hit $3.2 Trillion, Up 90% in Six Months
Off-balance-sheet obligations tied to AI at five US tech giants have risen about 90% in six months to roughly $3.2 trillion, a Nikkei compilation published November 4 shows. The same five companies (Alphabet, Meta, Microsoft, Amazon and Oracle) stood at an estimated $1.65 trillion when Nikkei ran its first version of the study in July. Nikkei says the commitments now equal about 80% of the companies’ assets.
What is in the number
The $3.2 trillion is not borrowed money. It is contracted future spending that sits in SEC footnotes rather than on the balance sheet: purchase commitments for chips, servers, construction and power, plus data-centre leases that have been signed but not yet begun. Existing loans and leases already on the balance sheet are excluded.
Filing-level figures from the latest disclosures show how uneven the stack is:
| Company | Disclosed item | Amount |
|---|---|---|
| Alphabet | Purchase obligations | $924.4B ($213.4B due within 12 months) |
| Meta | Non-cancellable contractual obligations | $454.5B |
| Meta | Leases not yet commenced | $355.7B |
| Amazon | Purchase and construction obligations | $254.4B |
| Amazon | Leases not yet commenced | $164.9B |
| Oracle | Future lease payments | $288B (terms of 15 to 19 years) |
Oracle’s profile is the outlier. Its exposure is mostly long-dated leases rather than supplier contracts, so the bill depends on data centres being filled for up to two decades.
Why the total is soft
Other tallies of the same five companies land at $3.3 trillion, and per-company splits differ between them. Companies report on different dates and with different definitions: Microsoft’s newer lease data from September is combined with its purchase-commitment disclosure from June. Alphabet’s purchase obligations include energy and content, not only AI infrastructure, and some are calculated at maximum contract prices and volumes with room to cut. Lease schedules are undiscounted future payments, so they cannot simply be added to a balance-sheet debt figure.
That makes the headline an upper-bound estimate of AI-linked contractual exposure, not a measure of leverage. The direction is clearer than the level: Alphabet’s total alone was reported to have risen nearly 700% in a year.
What to watch
The near-term pressure is the payment calendar. Amazon has about $74.1B of its purchase and construction obligations scheduled for 2027. Meta has $121.4B of non-cancellable purchase commitments due in 2027. Alphabet’s $213.4B falls inside 12 months. Those are the amounts that have to be funded from operating cash flow or new debt regardless of whether AI revenue keeps pace.
The structures also feed the bond market. Meta was reported earlier this week to be weighing a $20 billion-plus cross-border bond offering, and its Hyperion data-centre project is held 80% by Blue Owl Capital and 20% by Meta, a split that keeps most of the financing off Meta’s own books.