US tech off-balance-sheet AI obligations hit $3.2T, up 90% in six months

Nikkei analysis of Google, Meta, Amazon and Microsoft disclosures shows AI infrastructure obligations now dominate future cash flows. These include data-center leases, GPU purchase commitments and power contracts not recorded on balance sheets. Primary data drawn from 10-Q filings and earnings footnotes places cumulative capex plus contractual liabilities at levels last seen during 2000-era telecom buildouts.

Cross-referencing the same period's SEC filings reveals actual 2026 capex run rates already 35% above 2025 guidance for Alphabet and Meta. Off-balance-sheet treatment follows ASC 842 lease rules and vendor financing structures that defer recognition. This accounting choice masks the speed at which fixed obligations are growing relative to current AI revenue, which remains below 15% of total sales for all four firms.

Sustainability hinges on whether inference and training workloads generate cash before 2028 debt maturities. If utilization stays below 60% on new clusters, refinancing costs will rise once interest coverage ratios drop under 8x. Regulators reviewing climate-related disclosures may also require expanded lease and power-contract reporting.

Operational impact appears in 2027 guidance cycles: any slowdown in AI monetization will force either balance-sheet recognition or reduced expansion, directly affecting next-year capex budgets already projected above $200 billion combined.