$1.65 trillion in "shadow" debt: Big Tech builds an AI empire on hidden liabilities
The five largest technology corporations — Alphabet, Microsoft, Amazon, Meta, and Oracle — have accumulated $1.65 trillion in hidden debt obligations that are not reflected on their official balance sheets. This is happening amid a rapid contraction of free cash flow in the sector.
This refers to contracts for leasing GPUs, data centers, and stakes in joint ventures. Current accounting standards do not classify these obligations as debt until the facilities are put into operation. As a result, investors see only the tip of the iceberg.
Who is "hiding" how much?
The scale of hidden debt is extremely unevenly distributed. Meta (recognized as an extremist organization in Russia) has "hidden" $420 billion — three times its official debt. At Oracle, liabilities have grown 30-fold over four years. The Bank for International Settlements (BIS) has already dubbed this practice "shadow borrowing."
The entire scheme remains legal, but it fundamentally distorts the real picture for the market. Four out of five companies are reporting financial results within the next two weeks. Their official debt will look modest, while the $1.65 trillion from the notes to the financial statements will not make headlines.
The problem will fully manifest itself when data centers are launched. As soon as the facilities begin operations, the leases will hit the balance sheets all at once, and the financial reports will change dramatically. The worst-case scenario is weak demand for AI services. In this case, the facilities will have to be written down, and losses will fall on investors and insurers who financed the construction through private credit and project bonds, unaware of the full extent of the risks.
Cash flow is melting before our eyes
Bank of America has already warned of a liquidity crisis. The free cash flow of the largest data center operators will fall from $191 billion in 2025 to $19 billion in 2026, and will turn negative by approximately $26 billion in 2027.
The reason is record capital expenditures. The five corporations' spending on AI has already exceeded $725 billion in 2026, and the sector's debt issuance will reach about $175 billion — six times the average over the previous five years.
At the same time, the long-term forecast paints an opposite picture. Meta, Google, Microsoft, and Amazon could increase their combined free cash flow nearly tenfold — to about $650 billion by 2030. The bet is on a future monopoly: the companies aim not just to recoup infrastructure costs, but to become an essential link for anyone using AI and to profit from every query.
My analysis: This is a classic "do or die" dilemma. Big Tech is building AI infrastructure on debt, hiding half of its obligations in footnotes, and sacrificing cash flow for future control over the industry. If AI demand meets expectations, the winner will be the one that owns its infrastructure. If not, the "hidden" trillions will hit investors who didn't even know the full size of the bet. This makes the entire AI arms race one of the riskiest financial wagers of the decade.