Five US Tech Giants' Hidden AI Debt Hits $1.65 Trillion, Nikkei Finds

Here's a question worth asking about any company's AI infrastructure spending: how much of it is really debt, even when it doesn't show up as debt on the balance sheet? For five of the biggest US tech companies, a new Nikkei analysis puts a number on the answer, and it's larger than their official debt combined.

Alphabet, Microsoft, Amazon, Meta, and Oracle's combined on-the-books debt comes to roughly $1.35 trillion. Their combined "hidden" AI-related debt, according to Nikkei's estimate published July 21, 2026, is $1.65 trillion, an eightfold increase in just four years.

What counts as "hidden"

Nikkei's analysis isn't describing accounting fraud. It's describing off-balance-sheet obligations, mainly long-term data center leases and multi-year GPU supply contracts, that behave financially like debt (the company has to keep paying regardless of whether the capacity gets used) without being recorded as debt under standard accounting rules. Because these commitments sit outside the debt line, they're harder for investors and analysts to see when assessing how leveraged a company really is.

The exposure isn't spread evenly. Nikkei estimates Meta's off-balance-sheet liabilities at roughly $420 billion, nearly three times its transparent, on-the-books debt. Oracle's hidden debt has grown faster in relative terms: an estimated $273 billion as of the end of May, more than thirty times what it was four years earlier, tracking the company's aggressive push into AI infrastructure leasing.

Why analysts are paying attention

This isn't just a Nikkei observation. Morgan Stanley has flagged the growth in data center lease contracts as a risk factor in investor research, and Moody's has separately raised concerns that pre-operational lease commitments, agreements signed for capacity that isn't yet generating revenue, could increase financial strain if AI demand growth slows before the leases come due.

Why this matters if you use AI at work

None of this changes what Claude, Gemini, or Copilot can do for your business tomorrow morning. But it's a useful reality check on how the infrastructure underneath these tools is actually being financed. A meaningful share of the current AI buildout is running on leases and supply contracts that don't show up as debt but function like it, which means the underlying economics are less transparent, and arguably less tested, than the balance sheets suggest. If you're building long-term dependencies on a specific AI vendor's infrastructure, or negotiating multi-year contracts of your own, it's worth remembering that a vendor's headline financial health may not tell the whole story.

Sources: Nikkei Asia