$1.65 trillion in shadow debt: Big Tech plays a dangerous game with AI infrastructure
The largest technology corporations—Alphabet, Microsoft, Amazon, Meta, and Oracle—have accumulated hidden debt obligations totaling an astronomical $1.65 trillion. And this figure is not reflected in their official balance sheets. This refers to contracts for leasing GPUs, data centers, and joint ventures, which under current accounting rules are not considered debt until the facilities are operational.
The situation is particularly alarming at Meta, whose "hidden" liabilities have reached $420 billion—three times its official debt. At Oracle, off-balance-sheet risks have grown 30-fold in just four years. The Bank for International Settlements (BIS) has already characterized this practice as "shadow borrowing," and this definition sounds like a warning for the entire market.
Cash flow under threat
Alongside hidden debt, there is a dramatic compression of free cash flow (FCF) in the sector. According to my calculations, based on data from Bank of America, the aggregate FCF of the largest data center operators will fall from $191 billion in 2025 to $19 billion in 2026, and by 2027 it will turn negative, at approximately -$26 billion.
The reason is record capital expenditures on AI infrastructure. The Big Five have already spent over $725 billion on AI, and corporate debt issuance in the sector will reach about $175 billion—six times the average over the past five years. Companies are betting on a future monopoly: they are building not just infrastructure, but an essential link for everyone who will use AI, and they plan to profit from every request.
Alphabet appears to be the favorite here. Google owns the entire AI chain instead of leasing from others: Google Cloud revenue grew 63% year-over-year and exceeded $20 billion per quarter, its order backlog nearly doubled to $462 billion, and the company has started selling its proprietary TPU chips to external clients.
My analyst's perspective
The situation resembles a classic "all or nothing" scenario. If AI demand meets expectations, the winner will be the one who owns their own infrastructure. But if demand turns out weaker, the hidden trillions will hit investors who were unaware of the full scale of the bet. Facilities will have to be written down, and losses will fall on those who financed the construction through private credit and project bonds. The market is essentially playing Russian roulette with the balance sheets of the world's largest corporations.