Big tech holds $3 trillion in off-balance-sheet AI commitments, filings show

Uncommenced-lease obligations quadruple to $1.2 trillion in a year

Nine major technology companies carry roughly $3 trillion in off-balance-sheet commitments tied to artificial-intelligence infrastructure, according to a Wall Street Journal analysis of footnotes in their most recent securities filings. The obligations are growing faster than the $600 billion in traditional AI capital expenditures the same firms reported over the trailing twelve months, and total about three times the companies’ combined outstanding leases and long-term borrowings.

The Journal reviewed obligations disclosed by Alphabet, Amazon.com, Microsoft, Oracle, Nvidia, Broadcom, SpaceX and Advanced Micro Devices, alongside Meta Platforms. Across the nine companies, uncommenced-lease promises totaled $1.2 trillion — about four times more than the figure disclosed a year earlier — and purchase obligations stood at $1.9 trillion. The data are reported as of June for most of the firms, with Nvidia’s figures as of April and Meta’s as of July.

The obligations remain hidden from headline balance-sheet totals because of how the deals are structured and how accounting rules treat them. Purchase commitments — long-term contracts with chip suppliers and other vendors signed in advance to lock in production — typically stay off the balance sheet until a product or service is delivered. Lease obligations do not appear until rent payments begin. Together those two categories account for nearly all of the $3 trillion total.

Meta’s “Hyperion” data center in Louisiana, a campus roughly the size of 1,700 football fields, shows how the structure works in practice. Funds managed by Blue Owl Capital own the majority of a joint venture that in turn owns the campus, and a holding company called Beignet Investor raised the $27 billion in construction financing through a bond sale. None of that debt appears on Meta’s balance sheet.

Meta is Hyperion’s minority partner and tenant. The company initially agreed to lease Hyperion for a four-year term beginning in 2029, with options to renew for up to 20 years, and guaranteed that it would make bondholders whole if it did not stay the full two decades. Meta said it does not view payments under that guarantee as probable, so it has not recorded any related liability. Meta disclosed an aggregate initial lease commitment of about $12.3 billion for Hyperion and reported $347 billion in total obligations for leases that have not yet kicked in as of June.

Alphabet’s purchase commitments and contractual obligations rose sharply, reaching $811 billion as of June 30, up from $332 billion three months earlier. Alphabet said the commitments primarily relate to “technical infrastructure and inventory” and “agreements to secure energy for data center usage,” and that some obligations under its energy agreements extend as far out as 2054. The company did not detail why the obligations increased so substantially over the quarter.

Nvidia, the primary supplier of AI chips, committed to $27 billion in equity investments between April 26 and the end of its fiscal year in January 2027. Off-balance-sheet exposures at other companies include agreements to buy other companies’ stock in the future or to backstop leases for other tenants.

Wall Street is split on what the commitments portend. Optimists cite skyrocketing demand for AI tools and shortages of key hardware as evidence that demand will stay strong for years and that revenue will materialize to pay the bills. More cautious observers note that Alphabet and Amazon recently posted negative free cash flow — meaning capital spending exceeded operating cash — even before the off-balance-sheet obligations are counted, and that some companies once seen as having fortress balance sheets have had to tap the capital markets frequently.

Morgan Stanley accounting analysts wrote in April that the trend is making it harder for investors to evaluate the companies. “As these off-balance sheet commitments become more frequent, larger, and more complex, it is becoming increasingly difficult for investors to assess companies’ total potential leverage,” they said.

Because purchase commitments and signed leases generally cannot be canceled, the financial exposure is largely locked in regardless of how AI demand develops. If revenue assumptions prove wrong, the companies could end up paying for infrastructure they cannot profitably use and may have to take on more debt to service the obligations.