Big tech’s AI investments are fueling rapid growth in off-balance sheet debts, with companies holding far more financial risks than their balance sheets suggest, according to multiple analyses.
What is off-balance sheet AI debt?
Off-balance sheet debts are future financial obligations that do not appear as liabilities on a company’s balance sheet. They can take various forms such as lease liabilities for data centers or future GPU supply commitments.
They are legal and common accounting practices in the U.S., because long-term leases are classified as future obligations, unless their renewal is deemed reasonably certain.
They appear in the footnotes of companies’ financial statements, and the total liabilities are significantly higher than the $1.65 trillion in AI-related obligations revealed this year.
How big of a problem is off-balance sheet debt for AI?
The amount of debt varies depending on the criteria used to define obligations.
According to a Nikkei Asia investigation of SEC filings, Alphabet, Microsoft, Amazon, Meta, and Oracle have about $1.65 trillion in off-balance-sheet debt tied to AI infrastructure — more than the $1.35 trillion in liabilities that technically belong to them as of the most recent filing period.
This is more than eight times higher than what these companies reported in 2020 and reflects the intense build-out of infrastructure capacity needed to finance the AI arms races.
A broader analysis published on Stocktwits revealed that nine tech giants including Nvidia, Apple, Alphabet, Amazon, Microsoft, Meta, and Tesla have about $3 trillion in off-balance-sheet commitments, made up of $1.2 trillion in unstarted leases and $1.9 trillion in purchase commitments, as of the most recent report.
Meanwhile, Bloomberg reports that the four largest data center bidders have already allocated $2.4 trillion in anticipated spending on this front.
Who has the most off-balance sheet debt liabilities?
It depends on which hyperscalers we are talking about.
According to the Nikkei analysis , Meta has the biggest exposure with off-balance sheet liabilities of about $420 billion as of last month, which is nearly three times higher than the company’s reported liabilities.
Meanwhile, Oracle’s off-balance sheet obligations jumped thirtyfold over the past four years and reached about $273.3 billion, as of the most recent report.
Microsoft, Amazon, and Alphabet accounts for the rest, with combined liabilities of about $1.45 trillion in cloud-related expenses as of March 10, 2026.
Why are companies signing these off-balance sheet obligations?
There are growing reports about how specific transactions take place, for instance, Meta and Blackrock arranged for a $14 billion joint venture to build a one-gigawatt AI data center in El Paso, Texas, with BlackRock funds owning 80% of the campus while Meta acts as construction manager and sole tenant. That structure keeps $12.5 billion of project debt, which is classified as expenses rather than capital assets on Meta’s balance sheet, reducing the company’s exposure to capital expenditures by about $12.5 billion in 2026, according to one example given by the Nikkei.
Otherwise, these obligations typically take the form of long-dated data center leases, joint venture financing, and special purpose financing for.gpu clusters or data centers. Because these items do not appear as ordinary debt on the balance sheet of the companies in question, the actual liabilities that these firms have incurred in relation to the AI infrastructure build-out have been difficult to capture in traditional financial analysis.
Concerns from Wall Street about the risks from these obligations are mounting.
Morgan Stanley has investigated the matter in detail in the context of a broader investor brief, while Moody’s warned that lease liabilities for future data centers were rising rapidly at the beginning of this year . The International Monetary Fund and the Basel-based Bank of International Settlements have also raised concerns about the systemic risks posed by off-balance sheet financing for AI.
Individual investors are also voicing concerns. For example, hedge-fund manager Michael Burry, who famously shorted the subprime mortgage market ahead of the 2008 crisis, has been one of the most vocal critics of the current market setup. According to his recent note, big tech firms are taking advantage of extended depreciation periods for their quickly outdated chips and servers, which understates their capital expenditures and he warns of a three-way compression of demand for AI, corporate profits, and financing . Additionally, Burry argues that circular financing between hyperscalers, AI labs, and GPU manufacturers will lead to a self-fulfilling demand boom for these companies’ shares, artificially inflating their profits . Depreciation practices alone, according to Burry’s calculations, could reduce expenses by $176 billion between 2026 and 2028, which would increase Oracle’s earnings by 26.9% in 2028 and Meta’s by 20.8% .
On the other hand, Nvidia founder Jensen Huang has dismissed the criticism, arguing that these liabilities are “ridiculous”
as a significant expense in the context of the $730 billion in capital expenditures that the big tech firms are set to raise this year .
Market implications: are investors losing confidence in the AI economy?
There are signs that investors are starting to reassess the allure of the AI infrastructure plays.
For one, four hyperscalers are set to spend a combined $730 billion on AI infrastructure in 2026, but the confidence of investors in their growth prospects is evaporating . Specifically, the 30-day correlation between the two largest AI spenders and the broader semiconductor market has turned negative and dropped to zero from +0.78, which is the lowest level in over four years .
What do investors need to know about off-balance sheet AI debts?
The main lesson for investors is that off-balance sheet financing is ubiquitous and legal, but it raises concerns when it fuels the rapid escalation of liabilities. As noted by Nikkei, most corporations rely on leases and similar structures to reduce their reported debts on the balance sheet .
However, when the obligations balloon eight times faster than the debts on the balance sheet within a four-year window, as is the case with Meta, and exceed them by 3X for Oracle, Wall Street is justifiably concerned.
For now, investors should be especially mindful about the discrepancies between the liabilities reported in the footnotes of cloud giants’ financial statements and the figures that appear on their balance sheets, particularly in relation to their leases, purchase commitments, and joint venture financing arrangements.
This information should help them better understand the actual financial position of the companies.
