An in-depth analysis reveals that the capital expenditures reported by Alphabet, Meta, Microsoft, and Amazon in their latest earnings significantly understate their true financial commitments to future AI infrastructure.
Nine leading technology firms collectively hold approximately $3 trillion in off-balance-sheet obligations, primarily consisting of data center leases and chip purchase agreements. This figure is roughly three times their reported lease liabilities and long-term debt combined, and it is growing at a pace that far outstrips traditional capital spending.
Breaking down the off-balance-sheet commitments
According to an analysis of footnotes in the most recent quarterly reports, lease commitments that have not yet commenced total approximately $1.2 trillion, a year-over-year increase of about 400%. Purchase commitments, mainly covering chips, equipment, and energy agreements, account for roughly $1.9 trillion. Combined, these off-balance-sheet obligations reach approximately $3 trillion, dwarfing the roughly $600 billion in on-balance-sheet capital expenditures recorded over the past year.
Take Meta's Hyperion data center project in Louisiana as a case in point. The facility spans an area equivalent to about 1,700 football fields. Its construction is backed by roughly $27 billion in debt, none of which appears on Meta's balance sheet. The project is owned by a joint venture controlled by a fund managed by Blue Owl Capital. Meta holds a minority stake and acts as the tenant, supporting bond cash flows through future rent payments and providing residual value guarantees. Due to accounting rules, these obligations remain off the books until rent payments begin. Meta has disclosed total not-yet-started lease commitments of $347 billion.
Meanwhile, Alphabet's purchase and contractual obligations reached $811 billion as of the end of June. These commitments primarily involve technology infrastructure, inventory, and data center energy agreements, with some energy contracts extending as far out as 2054.
Accounting treatment and potential risks
Under current accounting standards, leases that have not commenced and undelivered purchase commitments typically do not appear on the balance sheet until actual payment or delivery occurs. This keeps reported leverage looking modest, even as future mandatory payment obligations loom large.
Optimists argue that surging AI demand will generate sufficient revenue to cover these commitments. Pessimists, however, point to recent negative free cash flow at both Alphabet and Amazon, where capital expenditures have exceeded operating cash flow. Should AI demand or hardware supply assumptions falter, these leases and purchase agreements, which cannot be easily cancelled, could become a heavy burden, potentially forcing companies to take on additional debt.
Accounting analysts at Morgan Stanley warn that as off-balance-sheet commitments grow in scale and complexity, it becomes significantly harder for investors to assess the true leverage of these companies.
What this means for the market
These figures reveal the hidden leverage underlying the AI arms race. The robust capital expenditure numbers visible on the surface are merely the tip of the iceberg; the actual financial exposure is far greater than what the books suggest.
If AI commercialization progresses smoothly, these commitments will translate into competitive advantages. However, if demand softens or technology paths shift, they could intensify financial pressure on the tech giants and trigger ripple effects across the entire AI supply chain's valuations and financing environment.
Investors would be wise to scrutinize the long-term commitments buried in financial statement footnotes, rather than focusing solely on traditional capital expenditures and on-book debt.