The modern arms race in artificial intelligence has led to unexpected financial consequences. As noted by Nikkei Asian Review, the industry's main problem is not circular financing deals, but massive hidden borrowing. Tech giants, striving to ensure the development of computing infrastructure, are accumulating debts that do not appear in standard financial statements.
The top five leading American companies in the industry — Alphabet (Google), Microsoft, Amazon, Meta, and Oracle — have accumulated debt obligations totaling $1.65 trillion. However, only $1.35 trillion is openly reflected in their financial reports. The difference of $300 billion is not the result of fraud, but a consequence of established practice: borrowing costs for projects that have not yet been implemented do not appear on the balance sheet. However, once the corresponding data centers (DCs) are launched, these liabilities must be officially recorded, which will significantly increase the debt burden on the business.
The Scale of Hidden Liabilities
The situation with "off-balance sheet" debts varies from company to company, but in some cases, the figures are impressive. For Meta, the magnitude of such obligations is estimated at $420 billion. This is approximately three times higher than the official amount of debt obligations reflected on the company's balance sheet.
Even more rapid growth in hidden debt is observed at Oracle. Over four years, the company increased its "shadow" obligations by more than 30 times — to $273 billion. This surge is largely due to Oracle's desire to participate in the Stargate initiative, which implies colossal investments in building AI data centers in the United States.
Financial Schemes and Payback Risks
Market participants are actively issuing bonds and conducting additional stock issuances to raise capital. However, representatives of Nikkei emphasize that the share of "shadow borrowing" — obtaining funds from institutional investors on opaque terms — is also growing.
Tech giants are counting on reasonable payback periods. Microsoft, Amazon (AWS), and Alphabet, as of the end of March, forecasted potential revenue of $1.45 trillion. Nevertheless, if based on the amounts of hidden debts, they already exceed the potential financial return from these projects.
Furthermore, infrastructure construction costs are difficult to predict. A striking example is Meta's project in Louisiana. In 2025, the company announced plans to spend about $27 billion on the facility, but this month increased the budget to $50 billion. Meta's partner in this project is Blue Owl Capital. It sold 20% of the joint venture's shares, and Meta committed to leasing the data center under construction. Formally, this allows Meta to use computing resources without reflecting direct costs for purchasing the DC in the documents.
The Shadow of Enron and the Threat of Overcapacity
Financial experts warn of risks, comparing the current situation to the bankruptcy of Enron in 2001. The energy giant, working in a different sector, also attempted to engage in "shadow borrowing" through multiple shell companies, which ultimately led to its collapse.
There is also a risk that an oversupply of computing power will emerge in the market. In this case, the multi-billion dollar investments in building data centers simply will not be justified within a reasonable timeframe. Financial schemes in the AI sector are becoming less transparent, causing serious concern among analysts.
