Big Tech is concealing $1.65 trillion in AI liabilities that are off the balance sheet.

      When examining the official debts of Alphabet, Microsoft, Amazon, Meta, and Oracle, the figures appear significant yet manageable. However, looking beyond the surface reveals a much larger and concealed amount of debt. A study conducted by Nikkei estimated this hidden total to be $1.65 trillion, which represents an increase of approximately eight times within four years, exceeding the $1.35 trillion that the five companies openly declare.

      This situation mirrors the Enron scandal, where money is obscured in off-balance-sheet structures, similar to the tactics Enron employed to conceal debt before its downfall 25 years ago. Although Enron's actions were fraudulent at that time, the introduction of stricter regulations and more comprehensive disclosures has made these practices legal today. The mechanisms remain unchanged; as analyst Gil Luria stated to Bloomberg Law, “Enron’s crime wasn’t having special purpose vehicles; it was hiding them.”

      The process is straightforward. Companies bundle debts related to components such as chips, servers, and energy into separate legal entities, often formed as joint ventures, so that the expenses do not reflect in their own financial statements. For instance, Meta’s Hyperion data center in Louisiana involved both Meta and Blue Owl Capital contributing equity to a distinct structure that assumed $27 billion in debt. Even though Meta is the only occupant, it claims that there is no obligation to record this debt, as it does not have to seek replacement tenants.

      Oracle has future lease commitments amounting to $260 billion that will eventually appear on its balance sheet, while Nvidia holds $119 billion in purchase obligations. Alphabet and Microsoft also maintain their financial vehicles off the books.

      The sheer magnitude of these figures is noteworthy. Meta’s hidden debt alone is approximately $420 billion, almost three times its reported liabilities, whereas Oracle's off-balance-sheet obligations have surged around thirtyfold in four years. This trend contributes to a larger goal, as the industry is projected to spend over $3 trillion on building and equipping AI data centers by 2028, much of which is financed against the chips within them.

      The timing of this issue is delicate, as four out of the five companies will announce earnings in the next two weeks, presenting a seemingly neat picture of their reported debt. However, the $1.65 trillion lurking in the footnotes is likely to be overlooked in headlines.

      The concern arises when a data center becomes operational, at which point its lease is incorporated onto the balance sheet all at once. Should the demand for AI fall short, the facility's value may decrease, resulting in losses for the lenders and insurers who financed it.

      Some market analysts are already expressing concerns. S&P has downgraded Oracle’s credit rating due to excessive leverage, and both Morgan Stanley and Moody’s have highlighted the broader implications. Accounting consultant Tom Selling posed a critical question: “What if one of these companies was a house of cards, propped up by this accounting treatment?”

      While all of this is not illegal, the companies assert that future profits will adequately cover their obligations. The details are available within the footnotes for those who take the time to look. However, this means that investors reviewing this week's earnings may only be aware of less than half of the actual leverage. In a year featuring significant discussions about market bubbles, the understated figures lurking off the page deserve attention.