Big Tech hides $1.65T in AI debt using Enron-style structures

The five technology giants—Alphabet, Microsoft, Amazon, Meta, and Oracle—have accumulated $1.65 trillion（約260兆円） in debt tied to AI infrastructure, according to a Nikkei study, yet this figure remains almost entirely hidden from their official balance sheets. The amount has surged roughly eightfold in four years and now exceeds the $1.35 trillion（約220兆円） these companies report as their total debt.

The mechanism mirrors structures Enron used before its 2001 collapse, though now legal under tightened regulations. Companies establish separate legal entities, often joint ventures, and place chip, server, and power infrastructure debt into those vehicles. Because the company is not technically the borrower, the obligations need not appear on its official accounts. Meta's Hyperion data center in Louisiana illustrates the pattern: Meta and Blue Owl Capital contributed equity to a separate structure that took on $27 billion（約4.3兆円） in debt. Meta serves as the sole tenant yet argues it bears no recording obligation because it is not responsible for finding replacement tenants. Similarly, Oracle carries $260 billion（約42兆円） in future lease commitments that will eventually reach its books, and Nvidia reports $119 billion（約19兆円） in purchase obligations.

The scale underscores the industry's AI buildout race. The sector is expected to spend more than $3 trillion（約480兆円） through 2028 on data center construction and equipment, much of it financed using this off-balance-sheet approach. Meta's off-balance-sheet debt alone totals about $420 billion（約67兆円）, nearly triple its reported debt. Oracle's hidden obligations have grown roughly thirtyfold in four years. Alphabet and Microsoft maintain their own off-book vehicles as well.

The risk emerges when reality deviates from expectations. When a data center goes live, its lease obligations move onto the balance sheet immediately. If AI demand falls short, facility values are marked down and losses crystallize for lenders and insurers. Credit rating agencies have begun flagging the risk: S&P has cut Oracle's credit rating citing stretched leverage, and both Morgan Stanley and Moody's have highlighted the broader exposure. Accounting consultant Tom Selling posed the stakes: "What if one of these companies was a house of cards and was propping itself up with this accounting treatment?" The timing amplifies the concern—four of the five companies report earnings within two weeks, and their official debt figures will appear tidy, while the $1.65 trillion（約260兆円） in obligations sits in footnotes where most investors will not read it.

The $1.65 trillion（約260兆円） in off-balance-sheet AI debt reflects an industry-wide financing model built around the expected $3 trillion（約480兆円） in data center spending through 2028. By packaging infrastructure obligations into separate legal entities—typically joint ventures where a tech giant is the sole tenant—companies can argue they bear no direct liability for the debt, keeping it off their official financial statements. Meta's Hyperion data center in Louisiana exemplifies the structure: Meta and Blue Owl Capital each contributed equity to a separate entity that took on $27 billion（約4.3兆円） in debt, yet Meta claims it need not record that liability because it is not obligated to find replacement tenants. This legal gray area has allowed the debt to balloon: Meta's off-balance-sheet obligations alone now total about $420 billion（約67兆円）, nearly triple its reported debt.

The timing of these disclosures matters acutely. With four of the five companies reporting earnings within two weeks, their official debt figures will appear manageable on press releases and investor calls, while the vastly larger obligations remain buried in footnotes. Credit agencies including S&P have already responded by cutting ratings—notably Oracle's—citing stretched leverage. The risk is asymmetric: if AI demand meets expectations, the leases will eventually flow through as planned and disappear into operating expenses. But if demand falters, the marked-down facility values and write-offs will crystallize losses simultaneously across lenders and insurers, potentially triggering contagion. The practice is legal precisely because disclosure exists; investors who read the footnotes can see the full picture. The concern is that the vast majority will not.