The five largest U.S. technology companies collectively hold roughly $1.65 trillion in debt that does not appear on their balance sheets. That is more than the debt they officially report. The money is financing the construction of data centers for artificial intelligence, and it is all legal. This is according to an extensive analysis by the Japanese newspaper Nikkei.
The scale of the debt and where it is hidden
When you look at what Alphabet, Microsoft, Amazon, Meta, and Oracle officially owe, the figures are enormous but manageable. Yet outside their balance sheets lies another, even larger pile of debt. Nikkei put it at $1.65 trillion. Over four years, this amount has grown roughly eightfold. And it exceeds the $1.35 trillion in debt that the five companies officially report in total.
To put it in Czech terms, this is an amount many times greater than the annual output of the entire Czech economy. Yet an ordinary shareholder will not find it among the headline figures. It is hidden in the notes to the financial statements, which most people never read.
A company packages debt for chips, servers, and electricity into a separate legal entity, often a joint venture with another party. The cost therefore never passes through its own accounts. A prime example is the Hyperion data center that Meta is building in Louisiana. Meta and Blue Owl Capital invested capital in a separate structure that took on $27 billion in debt. Meta is the sole tenant of the entire facility. Yet it claims it does not have to record the debt because it would not be responsible for finding replacement tenants if the project failed.
Oracle has contracted future lease payments totaling $260 billion, which will eventually appear on its balance sheet. Nvidia has committed to purchases worth $119 billion. Alphabet and Microsoft keep their structures off their balance sheets in the same way.
Meta and Oracle are cautionary examples
Meta's off-balance-sheet debt alone amounts to roughly $420 billion, nearly three times what the company officially reports. According to calculations derived from Nikkei's analysis, Meta reports around $140 billion on its balance sheet, while the hidden portion is three times higher.
Oracle is not much better off. Its off-balance-sheet liabilities have increased approximately thirtyfold over four years and now stand at around $273 billion. That is a jump of roughly 2,900 percent compared with 2022.
It all serves a single purpose. The industry is expected to spend more than $3 trillion by 2028 on building and equipping data centers for artificial intelligence. Much of this money is secured against the very chips that are intended to run inside the centers.
The shadow of Enron
The money is tied up in off-balance-sheet structures, the same kind of arrangements Enron used to hide debt. Enron was a U.S. energy company that collapsed in 2001 following one of the largest accounting frauds in U.S. history. Back then it was fraud, but today it is legal thanks to stricter rules and more detailed disclosures.
The tools, however, remain the same. Analyst Gil Luria summarized it for Bloomberg Law by saying that Enron's crime was not having special-purpose entities, but concealing them. That is the entire difference. Companies today do not hide their figures; they merely relegate them to places where almost no one looks.
The risks involved
Four of the five companies will publish their financial results over the next two weeks, and the debt they report will look tidy. The $1.65 trillion hidden in the notes will not make the headlines.
The catch is what comes next. Once a data center begins operating, its lease suddenly appears in the company's accounts as official debt. If demand for artificial intelligence computing power fails to reach the expected level, the value of the center will be written down, and the loss will fall on the lenders and insurers that financed its construction.
There is also risk in the structure of the deals themselves. Companies such as Alphabet, Amazon, and Microsoft have a backlog of cloud service orders worth $1.45 trillion. These are services they have yet to provide and be paid for. To deliver them, they enter into long-term contracts with data centers and secure computing capacity in advance. But if demand ultimately fails to materialize, they will be left with excess capacity and no one to sell it to. On top of that, these companies' capital expenditures already exceed their profits, so they are relying increasingly on corporate bonds and the issuance of new shares.
Some have already taken notice. S&P downgraded Oracle's credit rating because of its excessive debt. Morgan Stanley and Moody's have also raised concerns about the issue. Morgan Stanley estimates that hidden debt linked to artificial intelligence is even higher across the industry, at around $1.8 trillion. What would happen if one of these companies were a house of cards held together solely by this accounting practice? No one would rather imagine it.
The dispute over whether it is debt at all
Not all experts agree that this is debt in the true sense of the word. Some analysts say that a contract to purchase something is not debt. Both sides agree that the agreements are real and disclosed. They disagree on what to call them and how much they matter.
There are two key questions. Whether conventional debt metrics capture the true risk at all. And whether actual demand for artificial intelligence will keep pace with the capacity companies are financing in advance. Nikkei draws a parallel with the telecommunications crash at the beginning of the millennium, when equipment manufacturers financed the very customers purchasing their technology. It was circular financing that held together only as long as demand continued to outpace the debt being accumulated.
The companies insist that future profits will easily cover the bills. None of this is illegal, and all the figures can be found if one is willing to dig through the detailed notes to the financial statements. But it also means that shareholders reading this week's results will see less than half of the actual financial leverage.
