US tech giants hide $1.65tn in AI debt off balance sheets
Alphabet, Microsoft, Amazon, Meta and Oracle are keeping $1.65 trillion in AI infrastructure debt off their official balance sheets, leaving global investors exposed to a sudden reckoning if artificial intelligence demand falters.
Five of the world’s largest technology companies have accumulated $1.65 trillion in hidden debt to fund their artificial intelligence infrastructure. A new analysis shows Alphabet, Microsoft, Amazon, Meta and Oracle are keeping this liability off their official balance sheets through separate legal entities, a figure that has surged roughly eightfold in four years.
The off-balance-sheet pile now exceeds the $1.35 trillion the five companies officially report. Meta alone holds roughly $420 billion in unrecorded debt, nearly triple its stated obligations, while Oracle’s hidden borrowings have grown thirtyfold since 2020.
The accounting mechanics involve routing the cost of chips, servers and power into joint ventures. For instance, Meta partnered with Blue Owl Capital to borrow $27 billion for its Hyperion data centre in Louisiana, but Meta avoids recording the liability by arguing it is not responsible for finding replacement tenants.
Similar structures exist across the sector. Oracle has $260 billion in future lease commitments that will eventually hit its books, and Nvidia carries $119 billion in purchase obligations. The industry is expected to spend more than $3 trillion on AI data centres through 2028, much of it financed against the hardware inside.
The timing is critical for global markets as four of the five firms report earnings in the next two weeks. Investors poring over the headline numbers will see manageable leverage, missing the buried footnote figures that represent more than half the true financial burden.
While legal and disclosed in regulatory filings, the use of off-balance-sheet vehicles mirrors the tactics that destroyed Enron 25 years ago. “Enron’s crime wasn’t having special purpose vehicles,” analyst Gil Luria noted. “Enron’s crime was hiding them.”
The central danger emerges when a facility goes live and its lease transfers onto the balance sheet. If AI revenue fails to match the massive capital outlays, data centres will be written down, shifting the losses directly onto the lenders and insurers that financed them.
Credit agencies are already sounding alarms. S&P downgraded Oracle over stretched leverage, while Morgan Stanley and Moody’s have flagged the systemic risk across the sector. “What if one of these companies was a house of cards,” asked accounting consultant Tom Selling, “and was propping itself up with this accounting treatment?”
The technology firms maintain that future earnings will comfortably service these obligations. Yet for any European pension fund, bank or institutional investor holding these assets, the hidden $1.65 trillion represents a massive blind spot in the financial architecture of the AI boom.
